aws-s3-transfer 1.0.0

Transfer Manager for Amazon S3 - a high-performance data plane client for uploading and downloading objects and directories.
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
/*
 * Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
 * SPDX-License-Identifier: Apache-2.0
 */

//! Upload transfer implementation for scheduler integration.
//!
//! # Failure handling
//!
//! Upload requests carry no adaptive latency deadline (timing the whole
//! part-send is size-blind and false-cancels large parts on slow links). The
//! bounds on a failing or stuck upload are:
//!
//! - **Transient transport** (connection IO error, client-side timeout,
//!   ENOBUFS-style dispatch failure): the SDK retries, and this module's outer
//!   [`retry`](crate::retry::retry) loop re-issues past a drained shared retry
//!   token bucket with a fast backoff.
//! - **Throttling** (503 `SlowDown`): the SDK's retry token bucket handles it
//!   first; when that bucket drains under a high fan-out the outer
//!   [`retry`](crate::retry::retry) loop re-issues with a hard throttle backoff,
//!   bounded by the loop's attempt budget. The per-bucket token bucket
//!   ([`bucket_retry_partition`](crate::retry::bucket_retry_partition)) is given
//!   a time-based refill so a drained budget recovers.
//! - **Mid-upload-body stall** (the peer stops making progress on the request
//!   body): stalled-stream protection ([`Handle::upload_override`](crate::client::Handle::upload_override)).
//! - **Response never arrives after the body is fully sent**: NOT bounded. This
//!   is the response-first-byte gap — stalled-stream protection watches
//!   request-body throughput, which has already completed, and the SDK sets no
//!   response/operation timeout. Bounding it needs a response-first-byte timeout
//!   measured from send completion (a signal the SDK does not currently expose).

use std::cmp;
use std::future::Future;
use std::pin::Pin;
use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
use std::sync::{Arc, Mutex, MutexGuard};
use std::task::{Context, Poll};

use aws_sdk_s3::primitives::ByteStream;
use aws_sdk_s3::types::{CompletedMultipartUpload, CompletedPart};
use bytes::Bytes;
use tokio_util::sync::CancellationToken;
use tracing::Instrument;

use crate::error::{Error, ErrorKind};
use crate::io::part_reader::{Builder as PartReaderBuilder, PartReadStart};
use crate::io::{InputStream, PartData, SizeHint};
use crate::operation::upload::context::{
    apply_content_length, validate_composite_expected_value_for_multipart,
    validate_content_md5_for_multipart, validate_distinct_part_numbers,
    validate_part_checksum_has_strategy, validate_part_number, validate_size_hint,
    AbortMpuProgress, MultipartCompletion, PartPlan, PartReadWake, PartTransferState,
    PendingPartRead, UploadFailure, UploadPartWork, UploadState,
};
#[cfg(test)]
use crate::operation::upload::observability::UploadTerminalOutcome;
use crate::operation::upload::observability::{
    UploadDiagnosticTimer, UploadEvent, UploadExecutionState, UploadMode, UploadObservability,
    UploadPartTiming, UploadRequestKind, UploadStateSnapshot, UploadTerminalReport,
};
use crate::operation::upload::part_body;
use crate::operation::upload::request::{
    copy_fields_to_mpu_request, copy_fields_to_upload_part_request,
};
use crate::operation::upload::{UploadInput, UploadOutput, UploadOutputBuilder};
use crate::telemetry::TARGET_TRANSFER_UPLOAD;
use crate::transfer::{IoRequest, PendingCause, PollWork, Transfer, TransferContext, WorkOutcome};
use crate::types::{AbortOutcome, AbortedUpload, BucketType, FailedMultipartUploadPolicy};

/// Upload-specific work data.
///
/// Every variant carries an `InFlight` guard, which counts the item in
/// `UploadTransferInner::in_flight` until it is dropped.
#[derive(Debug)]
pub(crate) enum UploadWork {
    CreateMPU(InFlight),
    UploadPart(UploadPartWork, InFlight),
    CompleteMPU(Box<CompleteMpuWork>, InFlight),
    PutObject { stream: InputStream, live: InFlight },
    AbortMPU { upload_id: String, live: InFlight },
}

impl UploadWork {
    /// Short name of the work kind for span fields.
    fn kind(&self) -> &'static str {
        match self {
            Self::CreateMPU(_) => "create_multipart_upload",
            Self::UploadPart(..) => "upload_part",
            Self::CompleteMPU(..) => "complete_multipart_upload",
            Self::PutObject { .. } => "put_object",
            Self::AbortMPU { .. } => "abort_multipart_upload",
        }
    }
}

/// Payload for CompleteMultipartUpload, moved out of `UploadState::Completing` when the request is
/// dispatched.
#[derive(Debug)]
pub(crate) struct CompleteMpuWork {
    parts: PartTransferState,
    response_builder: UploadOutputBuilder,
}

/// Counts one emitted work item in `UploadTransferInner::in_flight` until the
/// item is dropped: after `execute` finishes it, when a cancellation drops it
/// mid-request, or when the scheduler discards it unexecuted.
///
/// The transfer does not signal terminal while the count is above zero (see
/// `UploadTransfer::signal_terminal_when_drained`), so `join()`/`abort()` never
/// return while one of our requests is still on the wire or still holds its
/// source read or buffer.
pub(crate) struct InFlight {
    inner: Arc<UploadTransferInner>,
}

impl InFlight {
    /// Counts one work item. Must be called under the state lock, so a
    /// `poll_work` that reads the count under the lock never misses the item.
    fn new(inner: &Arc<UploadTransferInner>) -> Self {
        inner.in_flight.fetch_add(1, Ordering::SeqCst);
        Self {
            inner: Arc::clone(inner),
        }
    }
}

impl Drop for InFlight {
    fn drop(&mut self) {
        let prev = self.inner.in_flight.fetch_sub(1, Ordering::SeqCst);
        if prev != 1 {
            return;
        }
        // The last item to go signals terminal if a request handler has
        // already ended the transfer.
        if !self.inner.ctx.is_active() {
            self.inner.ctx.signal_terminal();
        } else if self.inner.abort_requested.load(Ordering::SeqCst) {
            // Aborting state waits for the count to reach zero before it
            // ends the transfer. Singal the context to continue as the
            // count dropped to zero
            self.inner.ctx.wake();
        }
    }
}

impl std::fmt::Debug for InFlight {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.write_str("InFlight")
    }
}

/// Maximum number of parts that a single S3 multipart upload supports
const MAX_PARTS: u64 = 10_000;

/// Initial capacity for the completed-part list when the content length is unknown.
///
/// There is no part count to size it by, so this only avoids a few early reallocations; the list
/// grows as parts complete. This matches the CRT reference's default.
const UNKNOWN_LENGTH_DEFAULT_NUM_PARTS: usize = 32;

/// Upload transfer that generates and executes upload work.
///
/// Cheap to clone - all state is behind `Arc`.
#[derive(Debug, Clone)]
pub(crate) struct UploadTransfer {
    inner: Arc<UploadTransferInner>,
}

/// Internal state for upload transfer.
#[derive(Debug)]
struct UploadTransferInner {
    /// Common transfer lifecycle management
    ctx: TransferContext,
    /// State machine for work progression
    state: Mutex<UploadState>,
    /// Optional upload-wide event and summary observation.
    observability: UploadObservability,
    /// The original request (body taken for processing)
    request: Arc<UploadInput>,
    /// Type of S3 bucket targeted by this operation.
    // TODO(vnext): unify bucket representation (name + kind) across operations.
    #[allow(dead_code)]
    bucket_type: BucketType,
    /// What S3 reported, for the handle to read once the transfer has ended.
    /// Taken alone, never while holding the state lock.
    result: Mutex<Option<UploadResult>>,
    /// Whether an abort has been requested, by the caller or by a failure.
    /// Set once and never cleared.
    abort_requested: AtomicBool,
    /// Interrupts in-flight UploadPart (with its part reads) and PutObject.
    ///
    /// TODO: this should be moved away and let the scheduler to know which
    /// work can be interrupt or not.
    abort_parts_token: CancellationToken,
    /// Work items emitted and not yet dropped (see `InFlight`).
    in_flight: AtomicUsize,
}

/// What S3 reported for an upload.
///
/// When Complete commits while AbortMultipartUpload is on the wire, both can
/// succeed. `Completed` then wins, in either order: the object exists.
#[derive(Debug)]
pub(crate) enum UploadResult {
    /// CompleteMultipartUpload or PutObject committed the object.
    Completed(UploadOutput),
    /// AbortMultipartUpload succeeded (upload ID and request charged).
    Aborted(AbortedUpload),
}

/// Whether an upload is sent as a multipart upload rather than one PutObject:
/// `size_hint` has no upper bound, the source is a `PartStream` (`mpu_only`), or
/// the upper bound is at least `mpu_threshold`.
fn uses_multipart(size_hint: SizeHint, mpu_only: bool, mpu_threshold: u64) -> bool {
    match size_hint.upper() {
        None => true,
        Some(upper) => mpu_only || upper >= mpu_threshold,
    }
}

impl UploadTransfer {
    pub(crate) fn try_new(
        ctx: TransferContext,
        bucket_type: BucketType,
        request: UploadInput,
        stream: InputStream,
    ) -> Result<Self, Error> {
        let size_hint = stream.size_hint();
        validate_size_hint(size_hint).map_err(crate::error::invalid_input)?;
        // A declared content length replaces the body's bounds, so the size checks while reading
        // and at completion hold the body to exactly that many bytes.
        let size_hint = apply_content_length(size_hint, request.content_length())
            .map_err(crate::error::invalid_input)?;
        if uses_multipart(
            size_hint,
            stream.is_mpu_only(),
            ctx.handle.mpu_threshold_bytes(),
        ) {
            validate_content_md5_for_multipart(&request).map_err(crate::error::invalid_input)?;
            validate_composite_expected_value_for_multipart(&request)
                .map_err(crate::error::invalid_input)?;
        }
        let observability =
            UploadObservability::new(ctx.handle.config.diagnostics().transfer(), size_hint);

        // Only equal bounds are a known total. A bounded stream reports no total while in progress;
        // its actual size is published after EOF.
        if size_hint.upper() == Some(size_hint.lower()) {
            ctx.set_total_bytes(size_hint.lower());
        }

        let inner = Arc::new(UploadTransferInner {
            ctx,
            state: Mutex::new(UploadState::PendingInit {
                stream: Some(stream),
                size_hint,
                init_in_flight: false,
            }),
            observability,
            request: Arc::new(request),
            bucket_type,
            result: Mutex::new(None),
            abort_requested: AtomicBool::new(false),
            abort_parts_token: CancellationToken::new(),
            in_flight: AtomicUsize::new(0),
        });

        Ok(Self { inner })
    }

    /// Access the transfer context.
    pub(crate) fn ctx(&self) -> &TransferContext {
        &self.inner.ctx
    }

    fn observe_event(&self, event: UploadEvent, state: &UploadState) {
        let snapshot = snapshot_state(state);
        self.inner
            .observability
            .observe_event(self.inner.ctx.id, event, snapshot);
    }

    /// Emits the upload terminal summary before notifying the owning handle.
    ///
    /// `state_snapshot` is the state the transfer ended in. Must be called
    /// without the state lock held: the report takes the observability lock
    /// and emits events.
    fn report_terminal(&self, state_snapshot: UploadStateSnapshot) {
        self.inner.ctx.finalize_terminal_metrics();
        if !self.inner.ctx.claim_terminal_report() {
            return;
        }
        let pending = self.inner.ctx.pending_stats().unwrap_or_default();
        let _ = self
            .inner
            .observability
            .report_terminal(UploadTerminalReport {
                transfer_id: self.inner.ctx.id,
                status: self.inner.ctx.transfer_status(),
                error_kind: self.inner.ctx.error_kind(),
                metrics: self.inner.ctx.metrics(),
                request_total: self.inner.ctx.metrics.request_metrics(),
                pending,
                state_snapshot,
            });
    }

    /// Take what S3 reported (used by the handle once the transfer has ended).
    /// `None` if nothing committed and no AbortMultipartUpload succeeded.
    pub(crate) fn take_result(&self) -> Option<UploadResult> {
        self.inner.result.lock().expect("lock poisoned").take()
    }

    /// Starts a soft abort for the caller's `abort()`. Idempotent.
    pub(crate) fn request_abort(&self) {
        let state = self
            .inner
            .state
            .lock()
            .unwrap_or_else(std::sync::PoisonError::into_inner);
        self.enter_aborting(state, None);
        self.signal_abort();
    }

    /// Interrupts parts and PutObject and wakes the transfer so `poll_work`
    /// drives `Aborting` to an end. Called after `enter_aborting`.
    fn signal_abort(&self) {
        self.inner.abort_requested.store(true, Ordering::SeqCst);
        self.inner.abort_parts_token.cancel();
        self.inner.ctx.wake();
    }

    /// Logs a multipart upload that the transfer ended without completing or
    /// aborting, so the caller can abort it.
    fn warn_upload_left_open(&self, upload_id: &str) {
        tracing::warn!(
            target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
            tid = %self.inner.ctx.id,
            upload_id = %upload_id,
            "upload ended before completing; the multipart upload may still be open"
        );
    }

    /// Whether `Aborting` should send AbortMultipartUpload once it has an ID.
    fn abort_mpu_policy(&self) -> AbortMpuProgress {
        match self
            .inner
            .request
            .failed_multipart_upload_policy
            .clone()
            .unwrap_or_default()
        {
            FailedMultipartUploadPolicy::Retain => AbortMpuProgress::NotNeeded,
            // `_` also covers variants added later (the enum is non_exhaustive).
            _ => AbortMpuProgress::ToSend,
        }
    }

    /// Signals terminal if no work item is left; otherwise the last
    /// `InFlight` guard to drop signals it.
    fn signal_terminal_when_drained(&self) {
        if self.inner.in_flight.load(Ordering::SeqCst) == 0 {
            self.inner.ctx.signal_terminal();
        }
    }

    /// Sets the state to `Done` and returns its snapshot, in one lock.
    fn set_done(&self) -> UploadStateSnapshot {
        let mut state = self.inner.state.lock().expect("lock poisoned");
        *state = UploadState::Done;
        snapshot_state(&state)
    }

    /// Ends a transfer in `Aborting`: as failed with the recorded failure if
    /// there is one, otherwise as cancelled.
    ///
    /// `state` is the held state lock; it is set to `Done` and released
    /// before the terminal report.
    fn end_aborting(&self, mut state: MutexGuard<'_, UploadState>) -> PollWork {
        let UploadState::Aborting {
            upload_id,
            abort_mpu,
            failure,
            ..
        } = std::mem::replace(&mut *state, UploadState::Done)
        else {
            unreachable!("end_aborting called outside Aborting");
        };
        let snapshot = match &failure {
            Some(failure) => failure.snapshot,
            None => snapshot_state(&state),
        };
        drop(state);
        match failure {
            Some(failure) => {
                // The upload is still open in S3 unless AbortMultipartUpload
                // removed it: the policy is `Retain`, or the abort failed.
                // `end_aborting` runs only once no abort is pending, so these
                // are the remaining outcomes.
                let error = match upload_id {
                    Some(upload_id) if abort_mpu != AbortMpuProgress::Done => {
                        failure.error.with_upload_id(upload_id)
                    }
                    _ => failure.error,
                };
                self.inner.ctx.set_failed(error);
            }
            None => {
                // An upload ID with no AbortMultipartUpload to send means the
                // policy is `Retain`: report the open upload to the caller.
                // Stored before the terminal signal that `abort()` waits on.
                if let (Some(upload_id), AbortMpuProgress::NotNeeded) = (upload_id, abort_mpu) {
                    self.inner
                        .result
                        .lock()
                        .expect("lock poisoned")
                        .get_or_insert(UploadResult::Aborted(AbortedUpload {
                            upload_id: Some(upload_id),
                            outcome: AbortOutcome::Retained,
                            ..Default::default()
                        }));
                }
                self.inner.ctx.set_cancelled();
            }
        }
        self.report_terminal(snapshot);
        self.inner.ctx.signal_terminal();
        PollWork::Done
    }

    /// Moves the upload to `Aborting`, keeping the upload ID and which of
    /// CreateMultipartUpload and CompleteMultipartUpload are on the wire. The
    /// body and part state are dropped.
    ///
    /// `failure` becomes the transfer's outcome unless an earlier failure was
    /// recorded, and its request is no longer waited for. An upload already
    /// `Aborting` only records the failure. `Done`, or a transfer that has
    /// ended, is left as it is and the failure is dropped, logging an upload
    /// ID the failed request carries.
    ///
    /// The previous state is dropped after `state` is unlocked. It can own the
    /// caller's `PartStream`, whose `Drop` must not run under the lock.
    fn enter_aborting(
        &self,
        mut state: MutexGuard<'_, UploadState>,
        failure: Option<(UploadFailure, FailedRequest)>,
    ) {
        if !self.inner.ctx.is_active() {
            drop(state);
            // `on_terminal` drops the state. A CreateMultipartUpload that
            // failed after the transfer ended can carry the only copy of the
            // upload ID.
            if let Some((
                _,
                FailedRequest::CreateMpu {
                    upload_id: Some(upload_id),
                },
            )) = &failure
            {
                self.warn_upload_left_open(upload_id);
            }
            return;
        }
        let aborting = |upload_id, create_in_flight, complete_mpu_in_flight, abort_mpu| {
            UploadState::Aborting {
                upload_id,
                create_in_flight,
                complete_mpu_in_flight,
                abort_mpu,
                failure: None,
            }
        };
        let next = match &*state {
            // Create is on the wire and is not interrupted: if S3 accepted it,
            // there is an upload to abort, and only its response carries the
            // ID. The Create handler fills `upload_id` in when it returns.
            UploadState::PendingInit {
                init_in_flight: true,
                ..
            } => Some(aborting(None, true, false, self.abort_mpu_policy())),
            // Nothing was sent, so nothing exists on S3.
            UploadState::PendingInit {
                init_in_flight: false,
                ..
            } => Some(aborting(None, false, false, AbortMpuProgress::NotNeeded)),
            // Parts were interrupted by `abort_parts_token`. Their results no
            // longer matter, so the part state (reader, parked reads,
            // completed list) is dropped. From `Completing`, Complete has not
            // been sent and now never will be: the object cannot exist.
            UploadState::Transferring { upload_id, .. }
            | UploadState::Completing { upload_id, .. } => Some(aborting(
                Some(upload_id.clone()),
                false,
                false,
                self.abort_mpu_policy(),
            )),
            // Complete is (or is about to be) on the wire and is not
            // interrupted. Abort the upload right away, so that if S3 has not
            // committed yet the abort wins, and let Complete's response
            // decide the outcome. S3 accepts the abort whether or not Complete
            // committed.
            UploadState::CompleteInFlight { upload_id } => Some(aborting(
                Some(upload_id.clone()),
                false,
                true,
                self.abort_mpu_policy(),
            )),
            // PutObject is interrupted like a part (see `execute`), not waited
            // for: there is no upload to abort, and the caller gets `Cancelled`
            // even though S3 may already have stored the object. The
            // `Aborting` arm still waits for the dropped request to be freed
            // (its `InFlight` count).
            UploadState::PutObjectInFlight => {
                Some(aborting(None, false, false, AbortMpuProgress::NotNeeded))
            }
            UploadState::Aborting { .. } | UploadState::Done => None,
        };
        // Dropped once `state` is unlocked.
        let previous = next.map(|next| std::mem::replace(&mut *state, next));
        if let (
            Some((failure, failed)),
            UploadState::Aborting {
                upload_id,
                create_in_flight,
                complete_mpu_in_flight,
                abort_mpu,
                failure: recorded,
            },
        ) = (failure, &mut *state)
        {
            // The first failure is the one reported.
            if recorded.is_none() {
                *recorded = Some(failure);
            }
            // The failed request is still marked as on the wire, whether
            // `Aborting` was entered above or earlier.
            match failed {
                FailedRequest::Other => {}
                FailedRequest::CreateMpu { upload_id: id } => {
                    *create_in_flight = false;
                    if id.is_some() {
                        *upload_id = id;
                    }
                }
                FailedRequest::CompleteMpu => *complete_mpu_in_flight = false,
                FailedRequest::AbortMpu => *abort_mpu = AbortMpuProgress::Failed,
            }
        }
        drop(state);
        drop(previous);
    }

    /// Poll for the next work item.
    ///
    /// Returns:
    /// - `PollWork::Ready { .. }` - work available to execute
    /// - `PollWork::Pending` - waiting for in-flight work to complete
    /// - `PollWork::Done` - transfer complete
    #[tracing::instrument(
        level = "trace",
        target = TARGET_TRANSFER_UPLOAD,
        skip_all,
        fields(tid = %self.inner.ctx.id),
    )]
    pub(crate) fn poll_work(&self) -> PollWork {
        if !self.inner.ctx.is_active() {
            return PollWork::Done;
        }

        let mut state = self.inner.state.lock().expect("lock poisoned");

        loop {
            match &mut *state {
                UploadState::PendingInit {
                    init_in_flight,
                    size_hint,
                    stream,
                } => {
                    if *init_in_flight {
                        self.inner
                            .ctx
                            .set_pending(PendingCause::in_flight_work("create_multipart_upload"));
                        return PollWork::Pending;
                    }

                    let use_mpu = uses_multipart(
                        *size_hint,
                        stream.as_ref().is_some_and(|stream| stream.is_mpu_only()),
                        self.inner.ctx.handle.mpu_threshold_bytes(),
                    );
                    return if use_mpu {
                        *init_in_flight = true;
                        self.inner.observability.select_mode(UploadMode::Multipart);
                        self.inner.observability.observe_event(
                            self.inner.ctx.id,
                            UploadEvent::MultipartUploadScheduled,
                            UploadStateSnapshot::inactive(
                                UploadExecutionState::CreateMultipartUploadInFlight,
                            ),
                        );
                        PollWork::ready(IoRequest {
                            data: Some(Box::new(UploadWork::CreateMPU(InFlight::new(&self.inner)))),
                        })
                    } else {
                        let stream = stream.take().expect("stream already taken");
                        *state = UploadState::PutObjectInFlight;
                        self.inner.observability.select_mode(UploadMode::PutObject);
                        self.observe_event(UploadEvent::PutObjectScheduled, &state);
                        PollWork::ready(IoRequest {
                            data: Some(Box::new(UploadWork::PutObject {
                                stream,
                                live: InFlight::new(&self.inner),
                            })),
                        })
                    };
                }
                UploadState::Transferring { parts, .. } => {
                    if let Some(work) = parts.schedule_part() {
                        let event = if work.is_resumed() {
                            UploadEvent::SourceWakeScheduled
                        } else {
                            UploadEvent::NewPartScheduled
                        };
                        self.inner.observability.observe_event(
                            self.inner.ctx.id,
                            event,
                            parts.snapshot(),
                        );
                        return PollWork::ready(IoRequest {
                            data: Some(Box::new(UploadWork::UploadPart(
                                work,
                                InFlight::new(&self.inner),
                            ))),
                        });
                    }
                    if let Some(work) = parts.maybe_start_empty_object_part() {
                        self.inner.observability.observe_event(
                            self.inner.ctx.id,
                            UploadEvent::EmptyObjectPartScheduled,
                            parts.snapshot(),
                        );
                        return PollWork::ready(IoRequest {
                            data: Some(Box::new(UploadWork::UploadPart(
                                work,
                                InFlight::new(&self.inner),
                            ))),
                        });
                    }
                    let snapshot = parts.snapshot();
                    let pending_reason = parts.pending_reason();
                    if parts.is_complete() {
                        self.inner.observability.observe_event(
                            self.inner.ctx.id,
                            UploadEvent::CompletionReady,
                            snapshot,
                        );
                    }
                    if try_begin_completing(&mut state) {
                        continue;
                    }
                    self.inner.observability.observe_state(snapshot);
                    self.inner.ctx.set_pending(pending_reason);
                    return PollWork::Pending;
                }
                UploadState::Completing { .. } => {
                    // Move the fields out by value; the placeholder is overwritten before the
                    // state lock is released, so no reader observes it.
                    let UploadState::Completing {
                        upload_id,
                        parts,
                        response_builder,
                    } = std::mem::replace(&mut *state, UploadState::Done)
                    else {
                        unreachable!("state changed under lock");
                    };
                    *state = UploadState::CompleteInFlight { upload_id };
                    return PollWork::ready(IoRequest {
                        data: Some(Box::new(UploadWork::CompleteMPU(
                            Box::new(CompleteMpuWork {
                                parts,
                                response_builder,
                            }),
                            InFlight::new(&self.inner),
                        ))),
                    });
                }
                UploadState::CompleteInFlight { .. } => {
                    self.inner
                        .ctx
                        .set_pending(PendingCause::in_flight_work("complete_multipart_upload"));
                    return PollWork::Pending;
                }
                UploadState::PutObjectInFlight => {
                    self.inner
                        .ctx
                        .set_pending(PendingCause::in_flight_work("put_object"));
                    return PollWork::Pending;
                }
                UploadState::Aborting {
                    upload_id,
                    create_in_flight,
                    complete_mpu_in_flight,
                    abort_mpu,
                    ..
                } => {
                    // An upload exists and AbortMultipartUpload is still to be sent.
                    if let (Some(id), AbortMpuProgress::ToSend) = (upload_id.as_ref(), *abort_mpu) {
                        let upload_id = id.clone();
                        *abort_mpu = AbortMpuProgress::InFlight;
                        return PollWork::ready(IoRequest {
                            data: Some(Box::new(UploadWork::AbortMPU {
                                upload_id,
                                live: InFlight::new(&self.inner),
                            })),
                        });
                    }
                    // Wait while a request whose response decides the outcome,
                    // or any other emitted item, is outstanding.
                    let pending_inflight = match (
                        *create_in_flight,
                        *complete_mpu_in_flight,
                        *abort_mpu == AbortMpuProgress::InFlight,
                    ) {
                        (true, _, _) => Some("create_multipart_upload"),
                        (_, true, _) => Some("complete_multipart_upload_during_abort"),
                        (_, _, true) => Some("abort_multipart_upload"),
                        _ => {
                            if self.inner.in_flight.load(Ordering::SeqCst) > 0 {
                                Some("upload_parts")
                            } else {
                                None
                            }
                        }
                    };
                    if let Some(pending_inflight) = pending_inflight {
                        self.inner
                            .ctx
                            .set_pending(PendingCause::in_flight_work(pending_inflight));
                        return PollWork::Pending;
                    }
                    // Nothing is outstanding, and no commit succeeded (a commit
                    // moves the state to `Done` itself): the upload ends as
                    // failed if a failure was recorded, otherwise as cancelled.
                    return self.end_aborting(state);
                }
                UploadState::Done => return PollWork::Done,
            }
        }
    }

    #[tracing::instrument(
        level = "trace",
        target = TARGET_TRANSFER_UPLOAD,
        skip_all,
        fields(tid = %self.inner.ctx.id, work = tracing::field::Empty),
    )]
    pub(crate) async fn execute(&self, work: &mut IoRequest) -> WorkOutcome {
        // A work item executes once, so execute takes its payload by value.
        let data = work.take_data::<UploadWork>();
        tracing::Span::current().record("work", data.kind());
        match data {
            // Each arm keeps its `InFlight` guard until the request future is
            // gone, so the count is released only after that.
            UploadWork::CreateMPU(_live) => self.execute_create_mpu().await,
            UploadWork::UploadPart(part, live) => {
                // Race the part against the parts-only abort token; the loser
                // is dropped at its current `.await`. `biased` checks the
                // token first on every wake-up, so once the abort fires a part
                // never acts on a response it got later.
                let token = self.inner.abort_parts_token.clone();
                let outcome = tokio::select! {
                    biased;
                    _ = token.cancelled() => WorkOutcome::Cancelled,
                    outcome = self.execute_upload_part(part) => outcome,
                };
                // The part future is gone by the end of the `let` above, so
                // release the count only now.
                drop(live);
                outcome
            }
            UploadWork::CompleteMPU(completion, _live) => {
                self.execute_complete_mpu(*completion).await
            }
            UploadWork::PutObject { stream, live } => {
                // Same race as UploadPart: an abort drops PutObject wherever
                // it is (body upload, retry backoff or response wait) and the
                // upload ends as `Cancelled`. The object may still have been
                // stored if S3 had the whole body.
                let token = self.inner.abort_parts_token.clone();
                let outcome = tokio::select! {
                    biased;
                    _ = token.cancelled() => WorkOutcome::Cancelled,
                    outcome = self.execute_put_object(stream) => outcome,
                };
                drop(live);
                outcome
            }
            UploadWork::AbortMPU {
                upload_id,
                live: _live,
            } => self.execute_abort_mpu(upload_id).await,
        }
    }

    async fn execute_create_mpu(&self) -> WorkOutcome {
        let outcome = self.do_execute_create_mpu().await;
        // state changed - try to wake if we were pending
        self.inner.ctx.try_wake();
        outcome
    }

    /// Checks, without opening it, that a file-backed source names a regular
    /// file before CreateMultipartUpload is sent, so a source that is already
    /// non-regular fails without creating a multipart upload. The part reader
    /// checks the opened file again after Create returns.
    ///
    /// The stream leaves `PendingInit` while the metadata is read, so the read
    /// runs without the state lock, and returns afterwards. If the upload left
    /// `PendingInit` in the meantime, the stream is dropped and Create proceeds
    /// as it would for an abort that lands while Create is on the wire.
    ///
    /// Returns the failed outcome when the check fails, and `None` otherwise.
    fn check_file_source(&self) -> Option<WorkOutcome> {
        let stream = {
            let mut state = self.inner.state.lock().expect("lock poisoned");
            match &mut *state {
                UploadState::PendingInit { stream, .. }
                    if stream.as_ref().is_some_and(InputStream::is_file_backed) =>
                {
                    stream.take().expect("stream checked above")
                }
                _ => return None,
            }
        };
        let checked = stream.check_file_type();
        let mut state = self.inner.state.lock().expect("lock poisoned");
        match checked {
            // Create has not been sent, so no upload exists to abort.
            Err(error) => {
                Some(self.fail(state, error, FailedRequest::CreateMpu { upload_id: None }))
            }
            Ok(()) => {
                if let UploadState::PendingInit { stream: slot, .. } = &mut *state {
                    *slot = Some(stream);
                }
                None
            }
        }
    }

    async fn do_execute_create_mpu(&self) -> WorkOutcome {
        if let Some(outcome) = self.check_file_source() {
            return outcome;
        }

        let client = self.inner.ctx.s3_client();

        let mpu_req =
            copy_fields_to_mpu_request(&self.inner.request, client.create_multipart_upload());

        let req_metrics = self
            .inner
            .observability
            .start_request(&self.inner.ctx, UploadRequestKind::CreateMultipartUpload);
        let response = mpu_req
            .customize()
            .config_override(
                self.inner
                    .ctx
                    .bucket_partition_override(self.inner.request.bucket()),
            )
            .send()
            .instrument(tracing::trace_span!(
                target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
                "send-create-multipart-upload",
            ))
            .await;
        let _ = req_metrics.finish();
        let resp = match response {
            Ok(resp) => resp,
            // Create's response is in, so it is no longer on the wire. No
            // upload was created, so there is nothing to abort.
            Err(e) => {
                let state = self.inner.state.lock().expect("lock poisoned");
                return self.fail(
                    state,
                    crate::error::service_error(e),
                    FailedRequest::CreateMpu { upload_id: None },
                );
            }
        };

        let upload_id = resp.upload_id().expect("upload_id present").to_string();
        let response_builder = crate::sdk_v1::upload_output_from_create_multipart_upload(&resp);

        let (stream, size_hint) = {
            let mut state = self.inner.state.lock().expect("lock poisoned");
            match &mut *state {
                UploadState::PendingInit {
                    stream, size_hint, ..
                } => (stream.take().expect("stream already taken"), *size_hint),
                // An abort landed while Create was on the wire. Hand the ID to
                // the abort instead of starting the parts; `execute_create_mpu`
                // wakes the transfer after we return.
                UploadState::Aborting {
                    upload_id: slot,
                    create_in_flight,
                    ..
                } => {
                    *slot = Some(upload_id);
                    *create_in_flight = false;
                    return WorkOutcome::Success { data: None };
                }
                // The transfer ended while Create was on the wire. Nothing
                // aborts the multipart upload it opened.
                UploadState::Done => {
                    drop(state);
                    self.warn_upload_left_open(&upload_id);
                    return WorkOutcome::Cancelled;
                }
                _ => panic!("unexpected state for create_mpu"),
            }
        };
        let reads_until_eof = stream.is_mpu_only();

        // An upper bound can size parts so a manager-produced body fits S3's part limit. A stream
        // without one uses the configured size and enforces the limit as parts are produced.
        let part_size = match size_hint.upper() {
            Some(upper) => cmp::max(
                self.inner.ctx.handle.part_size_bytes(),
                upper.div_ceil(MAX_PARTS),
            ),
            None => self.inner.ctx.handle.part_size_bytes(),
        };
        let plan = if reads_until_eof {
            PartPlan::UntilEof { size_hint }
        } else {
            let declared_object_size = size_hint
                .upper()
                .expect("byte and file streams have an exact size");
            debug_assert_eq!(size_hint.lower(), declared_object_size);
            PartPlan::Known {
                total_parts: declared_object_size.div_ceil(part_size),
                declared_object_size,
            }
        };

        tracing::trace!(
            target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
            "upload request using multipart upload with part size: {part_size} bytes",
        );

        let part_reader = Arc::new(
            match PartReaderBuilder::new()
                .stream(stream)
                .part_size(part_size.try_into().expect("valid part size"))
                .direct_io(self.inner.ctx.handle.runtime.components().direct_io())
                .buffer_pool(self.inner.ctx.handle.buffer_pool.clone())
                .metrics(std::sync::Arc::clone(&self.inner.ctx.metrics))
                .telemetry(std::sync::Arc::clone(&self.inner.ctx.handle.telemetry))
                .build()
            {
                Ok(reader) => reader,
                // The upload exists but the state does not have its ID yet.
                // Hand it to the abort so AbortMultipartUpload is sent.
                Err(e) => {
                    let state = self.inner.state.lock().expect("lock poisoned");
                    return self.fail(
                        state,
                        e,
                        FailedRequest::CreateMpu {
                            upload_id: Some(upload_id),
                        },
                    );
                }
            },
        );

        let completed_parts_capacity = match &plan {
            PartPlan::Known { total_parts, .. } => *total_parts as usize,
            PartPlan::UntilEof { size_hint } => size_hint
                .upper()
                .map_or(UNKNOWN_LENGTH_DEFAULT_NUM_PARTS, |upper| {
                    upper.div_ceil(part_size) as usize
                }),
        };
        let snapshot = {
            let mut state = self.inner.state.lock().expect("lock poisoned");
            // The abort, or the end of the transfer, can also land after the
            // first lock above, while the part reader was being built with the
            // lock released. Same handling; the reader is dropped (closing the
            // source) after the lock is released when we return.
            match &mut *state {
                UploadState::Aborting {
                    upload_id: slot,
                    create_in_flight,
                    ..
                } => {
                    *slot = Some(upload_id);
                    *create_in_flight = false;
                    return WorkOutcome::Success { data: None };
                }
                UploadState::Done => {
                    drop(state);
                    self.warn_upload_left_open(&upload_id);
                    return WorkOutcome::Cancelled;
                }
                _ => {}
            }
            *state = UploadState::Transferring {
                upload_id,
                parts: PartTransferState::new(
                    part_reader,
                    plan,
                    completed_parts_capacity,
                    self.inner.observability.clone(),
                ),
                response_builder,
            };
            snapshot_state(&state)
        };
        self.inner.observability.multipart_created();
        self.inner.observability.observe_event(
            self.inner.ctx.id,
            UploadEvent::MultipartUploadCreated,
            snapshot,
        );

        tracing::debug!(
            target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
            total_parts = size_hint.upper().map(|upper| upper.div_ceil(part_size)),
            part_size,
            "MPU created, transferring",
        );

        WorkOutcome::Success { data: None }
    }

    async fn execute_upload_part(&self, mut work: UploadPartWork) -> WorkOutcome {
        if work.is_empty_object() {
            tracing::debug!(
                target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
                tid = %self.inner.ctx.id,
                "empty source satisfied its size hint; uploading required zero-byte part",
            );
            return self
                .send_part(PartData::new(1, Bytes::new()), work.take_timing())
                .await;
        }

        let (part_reader, reads_until_eof) = {
            let state = self.inner.state.lock().expect("lock poisoned");
            match &*state {
                UploadState::Transferring { parts, .. } => {
                    (Arc::clone(&parts.part_reader), parts.reads_until_eof())
                }
                other => return abandoned_part(other, "for read_part"),
            }
        };

        let (mut future, wake, mut timing) = match work.take_pending_read() {
            Some(read) => (read.future, read.wake, read.timing),
            None => {
                let future = match part_reader
                    .start_part_read()
                    .instrument(tracing::trace_span!(
                        target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
                        "read-upload-body",
                    ))
                    .await
                {
                    PartReadStart::Ready(future) => future,
                    PartReadStart::Blocked => {
                        let mut state = self.inner.state.lock().expect("lock poisoned");
                        let UploadState::Transferring { parts, .. } = &mut *state else {
                            return abandoned_part(&state, "while retracting upload part");
                        };
                        parts.retract_scheduled_part();
                        let snapshot = parts.snapshot();
                        drop(state);
                        self.inner.observability.observe_event(
                            self.inner.ctx.id,
                            UploadEvent::CustomSourceBlocked,
                            snapshot,
                        );
                        return WorkOutcome::Yielded;
                    }
                    PartReadStart::Finished => {
                        self.finish_read_without_part();
                        return WorkOutcome::Yielded;
                    }
                };
                (
                    future,
                    PartReadWake::new(self.inner.ctx.waker()),
                    work.take_timing(),
                )
            }
        };

        let result = {
            let task_waker = std::task::Waker::from(Arc::clone(&wake));
            let mut task_context = Context::from_waker(&task_waker);
            Pin::new(&mut future).poll(&mut task_context)
        };
        let data = match result {
            Poll::Pending => {
                let mut state = self.inner.state.lock().expect("lock poisoned");
                let UploadState::Transferring { parts, .. } = &mut *state else {
                    return abandoned_part(&state, "while parking upload part read");
                };
                parts.record_read_pending(&mut timing);
                let parked_wake = Arc::clone(&wake);
                let observation = timing.observation();
                parts.park_read(PendingPartRead {
                    future,
                    wake,
                    timing,
                });
                let snapshot = parts.snapshot();
                drop(state);
                self.inner.observability.observe_source_pending(
                    self.inner.ctx.id,
                    observation,
                    snapshot,
                );
                parked_wake.requeue_if_notified();
                return WorkOutcome::Yielded;
            }
            Poll::Ready(Ok(Some(data))) => data,
            Poll::Ready(Ok(None)) => {
                return self.on_end_of_stream();
            }
            Poll::Ready(Err(error)) => {
                let state = self.inner.state.lock().expect("lock poisoned");
                return self.fail(state, error.into(), FailedRequest::Other);
            }
        };

        {
            let mut state = self.inner.state.lock().expect("lock poisoned");
            let UploadState::Transferring { parts, .. } = &mut *state else {
                return abandoned_part(&state, "while recording upload part");
            };
            let parts_read = match parts.observe_part(data.data.len() as u64) {
                Ok(parts_read) => parts_read,
                Err(error) => {
                    return self.fail(
                        state,
                        crate::error::invalid_input(error),
                        FailedRequest::Other,
                    );
                }
            };
            if reads_until_eof && parts_read > MAX_PARTS {
                return self.fail(
                    state,
                    Error::new(
                        ErrorKind::InputInvalid,
                        format!(
                            "stream produced more than the maximum of {MAX_PARTS} parts; combine \
                             source data into fewer, larger parts"
                        ),
                    ),
                    FailedRequest::Other,
                );
            }
            if let Err(error) = validate_part_number(data.part_number) {
                return self.fail(
                    state,
                    crate::error::invalid_input(error),
                    FailedRequest::Other,
                );
            }
            if let Err(error) = validate_part_checksum_has_strategy(&self.inner.request, &data) {
                return self.fail(
                    state,
                    crate::error::invalid_input(error),
                    FailedRequest::Other,
                );
            }
        }

        // A custom source that was blocked is available again. Refill the work withheld while the
        // retained source operation was pending so reading the next part can overlap this upload.
        self.inner.ctx.try_wake();

        self.send_part(data, timing).await
    }

    /// Records source end-of-stream.
    ///
    /// Empty-part synthesis is deferred until every source read and UploadPart operation drains.
    /// At that point transfer state can distinguish no source parts from an explicit zero-byte part.
    fn on_end_of_stream(&self) -> WorkOutcome {
        let (transitioned, snapshot) = {
            let mut state = self.inner.state.lock().expect("lock poisoned");
            let UploadState::Transferring { parts, .. } = &mut *state else {
                return abandoned_part(&state, "at upload end-of-stream");
            };
            let transitioned = parts.record_end_of_stream();
            (transitioned, parts.snapshot())
        };
        if transitioned {
            self.inner.observability.observe_event(
                self.inner.ctx.id,
                UploadEvent::SourceExhausted,
                snapshot,
            );
        }

        // End-of-stream may close dispatch while `poll_work` is parked behind the source gate.
        self.inner.ctx.try_wake();

        self.finish_read_without_part();
        WorkOutcome::Success { data: None }
    }

    /// Uploads one part and records it for CompleteMultipartUpload.
    ///
    /// This is shared by ordinary source output and the empty-object part, keeping request
    /// retries, accounting, and completion behavior identical.
    async fn send_part(&self, data: PartData, timing: UploadPartTiming) -> WorkOutcome {
        let part_number = data.part_number;
        let presentation_segments = data.data.segment_count();
        let content_length = data.data.len() as i64;
        let bytes_sent = content_length as u64;

        let (upload_id, source_observation, request_start_snapshot) = {
            let mut state = self.inner.state.lock().expect("lock poisoned");
            match &mut *state {
                UploadState::Transferring {
                    upload_id, parts, ..
                } => {
                    let source_observation = parts.begin_upload(presentation_segments, timing);
                    (upload_id.clone(), source_observation, parts.snapshot())
                }
                other => return abandoned_part(other, "for send_part"),
            }
        };

        let part_num_i32 = i32::try_from(part_number)
            .expect("part numbers are checked against 1..=10,000 before UploadPart");
        self.inner.observability.observe_part_started(
            self.inner.ctx.id,
            part_number,
            bytes_sent,
            presentation_segments,
            source_observation,
            request_start_snapshot,
        );

        let sdk_body = part_body::sdk_body(data.data);
        let checksum = data.checksum;

        // Retry transient transport errors and throttles (the classifier picks
        // the backoff: fast for transient, hard for a throttle). The SDK normally
        // retries an UploadPart dispatch over the rewindable in-memory body, but a
        // concurrent ENOBUFS-style burst or a throttle storm can exhaust its
        // shared retry token bucket and surface parts un-recovered; this outer
        // loop re-issues those, landing after in-flight parts refill the quota.
        //
        // No adaptive latency deadline on the upload path: timing the whole
        // part-send (body-push + response) is size-blind and false-cancels large
        // parts on slow links. Stalled-stream protection (`upload_override`)
        // bounds a mid-upload-body stall; a response that never arrives after the
        // body is fully sent is not bounded here (see the module docs on the
        // response-first-byte gap).
        let mut req_metrics = self
            .inner
            .observability
            .start_request(&self.inner.ctx, UploadRequestKind::UploadPart);
        let retry_classify = crate::retry::classify_upload_part_retry;
        let result = crate::retry::retry(req_metrics.metrics_mut(), retry_classify, |_hedge| {
            let body = sdk_body
                .try_clone()
                .expect("UploadPart SdkBody must be retryable");
            let req = copy_fields_to_upload_part_request(
                &self.inner.request,
                self.inner
                    .ctx
                    .s3_client()
                    .upload_part()
                    .upload_id(&upload_id)
                    .part_number(part_num_i32)
                    .content_length(content_length)
                    .body(ByteStream::new(body)),
                checksum.as_ref(),
            );
            async move {
                req.customize()
                    .config_override(self.inner.ctx.upload_override(self.inner.request.bucket()))
                    .disable_payload_signing()
                    .send()
                    .instrument(tracing::trace_span!(
                        target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
                        "send-upload-part",
                        part_number,
                    ))
                    .await
                    .map_err(|e| crate::retry::GuardError::Inner(crate::error::service_error(e)))
            }
        })
        .instrument(tracing::trace_span!(
            target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
            "upload-part",
            tid = %self.inner.ctx.id,
            part_number
        ))
        .await;
        let request_elapsed = req_metrics.finish().elapsed;
        // The original retry body retains every immutable payload owner even
        // after the successful request clone has been consumed. Release it
        // before publishing part completion: the final part can wake
        // CompleteMultipartUpload on another worker, and that operation may
        // signal `join()` before this work item otherwise unwinds.
        drop(sdk_body);
        // Validate the response after the retry loop, so a rejected response is not re-sent.
        let result = result.and_then(|resp| validate_upload_part_response(resp, part_number));
        let resp = match result {
            Ok(resp) => resp,
            Err(e) => {
                self.inner.observability.complete_upload();
                // Outside `Transferring` (an abort started) there is no part
                // snapshot to report, but the error is still recorded. The
                // part event is emitted after the lock is released; the
                // terminal report cannot come first, because it waits for this
                // work item to drop.
                let state = self.inner.state.lock().expect("lock poisoned");
                let snapshot = match &*state {
                    UploadState::Transferring { parts, .. } => Some(parts.snapshot()),
                    _ => None,
                };
                let outcome = self.fail(state, e, FailedRequest::Other);
                if let Some(snapshot) = snapshot {
                    self.inner.observability.observe_part_failed(
                        self.inner.ctx.id,
                        part_number,
                        bytes_sent,
                        presentation_segments,
                        request_elapsed,
                        snapshot,
                    );
                }
                return outcome;
            }
        };

        let completed = CompletedPart::builder()
            .part_number(part_num_i32)
            .set_e_tag(resp.e_tag.clone())
            .set_checksum_crc32(resp.checksum_crc32.clone())
            .set_checksum_crc32_c(resp.checksum_crc32_c.clone())
            .set_checksum_crc64_nvme(resp.checksum_crc64_nvme.clone())
            .set_checksum_sha1(resp.checksum_sha1.clone())
            .set_checksum_sha256(resp.checksum_sha256.clone())
            .build();

        let (should_wake, completion_snapshot) = {
            let mut state = self.inner.state.lock().expect("lock poisoned");
            let UploadState::Transferring { parts, .. } = &mut *state else {
                return abandoned_part(&state, "while completing upload part");
            };
            parts.complete_part(completed, bytes_sent);
            let should_wake = try_begin_completing(&mut state);
            let snapshot = snapshot_state(&state);
            (should_wake, snapshot)
        };

        self.inner.observability.observe_part_completed(
            self.inner.ctx.id,
            part_number,
            bytes_sent,
            presentation_segments,
            request_elapsed,
            completion_snapshot,
        );

        self.inner.ctx.record_io(&crate::metrics::IoSample {
            network_tx: bytes_sent,
            ..Default::default()
        });

        if should_wake {
            self.inner.ctx.try_wake();
        }

        WorkOutcome::Success { data: None }
    }

    /// Retires scheduled source work that produced no UploadPart request.
    fn finish_read_without_part(&self) {
        let mut state = self.inner.state.lock().expect("lock poisoned");
        let UploadState::Transferring { parts, .. } = &mut *state else {
            abandoned_part(&state, "while completing empty upload read");
            return;
        };
        let empty_object_ready = parts.finish_read_without_part();
        if empty_object_ready || try_begin_completing(&mut state) {
            drop(state);
            self.inner.ctx.try_wake();
        }
    }

    async fn execute_put_object(&self, stream: InputStream) -> WorkOutcome {
        use crate::operation::upload::request::copy_fields_to_put_object_request;

        {
            let state = self.inner.state.lock().expect("lock poisoned");
            // The upload was aborted or ended after dispatch: never send
            // PutObject. `execute` normally catches this first (the token is
            // checked before this future is polled); this is the backstop.
            if matches!(&*state, UploadState::Aborting { .. } | UploadState::Done) {
                return WorkOutcome::Cancelled;
            }
        }
        let content_length = stream
            .size_hint()
            .upper()
            .expect("content length must be known for PutObject");

        let is_file_backed = stream.is_file_backed();
        let direct_io = self.inner.ctx.handle.runtime.components().direct_body_io();

        // Hand the request body off to the SDK as a retryable `SdkBody`:
        // in-memory sources ride `SdkBody::from(Bytes)`'s built-in rebuild path;
        // file-backed sources go through `DirectFileBody` / `OffloadedFileBody`
        // (one stable file identity, a fresh cursor per retry, and pooled
        // bounded chunks).
        // The body must stay a native `SdkBody` (not a custom wrapper) so the SDK
        // keeps its in-memory checksum path — wrapping would force aws-chunked
        // trailer encoding and change the checksum framing.
        let sdk_body =
            match stream.into_sdk_body(direct_io, self.inner.ctx.handle.buffer_pool.clone()) {
                Ok(body) => body,
                Err(error) => {
                    let state = self.inner.state.lock().expect("lock poisoned");
                    return self.fail(state, error, FailedRequest::Other);
                }
            };

        let transfer_id = self.inner.ctx.id;
        tracing::trace!(
            target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
            tid = %transfer_id,
            content_length,
            is_file_backed,
            "put_object.send_enter",
        );

        // Retry transient transport and throttles (same rationale as UploadPart).
        // No adaptive latency deadline; a mid-upload-body stall is bounded by
        // stalled-stream protection (`upload_override`), a post-send response-wait
        // is not (see the module docs on the response-first-byte gap).
        let mut req_metrics = self
            .inner
            .observability
            .start_request(&self.inner.ctx, UploadRequestKind::PutObject);
        let retry_classify = crate::retry::classify_upload_part_retry;
        let result = crate::retry::retry(req_metrics.metrics_mut(), retry_classify, |_hedge| {
            let body = sdk_body
                .try_clone()
                .expect("PutObject SdkBody must be retryable");
            let put_req = copy_fields_to_put_object_request(
                &self.inner.request,
                self.inner
                    .ctx
                    .s3_client()
                    .put_object()
                    .body(ByteStream::new(body)),
            );
            async move {
                put_req
                    .customize()
                    .config_override(self.inner.ctx.upload_override(self.inner.request.bucket()))
                    .disable_payload_signing()
                    .send()
                    .instrument(tracing::trace_span!(
                        target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
                        "send-put-object",
                    ))
                    .await
                    .map_err(|e| crate::retry::GuardError::Inner(crate::error::service_error(e)))
            }
        })
        .instrument(tracing::trace_span!(
            target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
            "put-object",
            tid = %transfer_id
        ))
        .await;
        let _ = req_metrics.finish();
        let resp = match result {
            Ok(resp) => {
                tracing::trace!(
                    target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
                    tid = %transfer_id,
                    "put_object.send_exit_ok",
                );
                resp
            }
            Err(e) => {
                tracing::debug!(
                    target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
                    tid = %transfer_id,
                    error = %e,
                    "put_object.send_exit_err",
                );
                let state = self.inner.state.lock().expect("lock poisoned");
                return self.fail(state, e, FailedRequest::Other);
            }
        };

        // A successful response means the SDK fully read the body and
        // the bytes made it to the wire. For file-backed sources, the
        // body implementations above have therefore read `content_length`
        // bytes from disk. We attribute the metric here (rather than
        // per-chunk inside the body) so that there is a single semantic
        // anchor — "recorded when the SDK confirms success" — consistent
        // with how `network_tx` is attributed.
        let disk_read = if is_file_backed { content_length } else { 0 };
        self.inner.ctx.record_io(&crate::metrics::IoSample {
            network_tx: content_length,
            disk_read,
            ..Default::default()
        });

        // The output carries the transfer's final metrics, so its I/O and
        // finish are recorded before the snapshot. The result is stored
        // before `set_completed`, because the last `InFlight` guard signals
        // terminal once the transfer is inactive.
        self.inner.ctx.finalize_terminal_metrics();
        let result = crate::sdk_v1::upload_output_from_put_object(&resp)
            .metrics(self.inner.ctx.metrics())
            .build()
            .expect("valid response");

        *self.inner.result.lock().expect("lock poisoned") = Some(UploadResult::Completed(result));

        let snapshot = self.set_done();
        self.inner.ctx.set_completed();
        self.report_terminal(snapshot);
        self.signal_terminal_when_drained();

        WorkOutcome::Success { data: None }
    }

    async fn execute_complete_mpu(&self, completion: CompleteMpuWork) -> WorkOutcome {
        let CompleteMpuWork {
            parts,
            response_builder,
        } = completion;
        let transfer_diagnostics = self.inner.ctx.handle.config.diagnostics().transfer();
        let completion_timer = UploadDiagnosticTimer::start(transfer_diagnostics);
        let upload_id = {
            let mut state = self.inner.state.lock().expect("lock poisoned");
            match &mut *state {
                UploadState::CompleteInFlight { upload_id } => upload_id.clone(),
                // The abort landed after `poll_work` dispatched us but before we
                // started. Never send Complete, so the object cannot exist, and
                // tell the abort it no longer has to wait for us.
                UploadState::Aborting {
                    complete_mpu_in_flight,
                    ..
                } => {
                    *complete_mpu_in_flight = false;
                    drop(state); // unlock before waking
                    self.inner.ctx.try_wake();
                    return WorkOutcome::Cancelled;
                }
                // The transfer has ended: never send Complete.
                UploadState::Done => return WorkOutcome::Cancelled,
                _ => panic!("unexpected state for complete_mpu"),
            }
        };
        // The part state is held by this work item rather than by `CompleteInFlight`,
        // so its measurements come from here.
        self.inner.observability.observe_event(
            self.inner.ctx.id,
            UploadEvent::MultipartCompletionStarted,
            parts
                .snapshot()
                .with_state(UploadExecutionState::CompleteMultipartUploadInFlight),
        );

        let MultipartCompletion {
            part_reader,
            plan,
            mut completed_parts,
            bytes_read,
            final_snapshot,
        } = parts.into_completion();
        completed_parts.sort_by_key(|p| p.part_number);
        // Sorted, so a number used twice is adjacent. Checked here rather than per part: the
        // assembled list is the first place a repeat is visible without tracking every number.
        // Complete is not sent on these errors, but the upload is still open.
        if let Err(error) = validate_distinct_part_numbers(&completed_parts) {
            return self.complete_mpu_failed(crate::error::invalid_input(error));
        }

        let object_size = match plan.mpu_object_size(bytes_read) {
            Ok(object_size) => object_size,
            Err(error) => return self.complete_mpu_failed(crate::error::invalid_input(error)),
        };
        self.inner.ctx.set_total_bytes(object_size);

        let base_req = self
            .inner
            .ctx
            .s3_client()
            .complete_multipart_upload()
            .upload_id(&upload_id)
            .mpu_object_size(object_size as i64)
            .multipart_upload(
                CompletedMultipartUpload::builder()
                    .set_parts(Some(completed_parts))
                    .build(),
            );

        let complete_req = super::request::copy_fields_to_complete_mpu_request(
            &self.inner.request,
            base_req,
            || async { part_reader.full_object_checksum().await },
        )
        .await;
        // The transfer may have ended while the request was prepared.
        if !self.inner.ctx.is_active() {
            return WorkOutcome::Cancelled;
        }

        let req_metrics = self
            .inner
            .observability
            .start_request(&self.inner.ctx, UploadRequestKind::CompleteMultipartUpload);
        let response = complete_req
            .customize()
            .config_override(
                self.inner
                    .ctx
                    .bucket_partition_override(self.inner.request.bucket()),
            )
            .send()
            .instrument(tracing::trace_span!(
                target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
                "send-complete-multipart-upload",
            ))
            .await;
        let _ = req_metrics.finish();
        let resp = match response {
            Ok(resp) => resp,
            Err(e) => {
                let error = crate::error::service_error(e);
                // The TM SEP requires a warning with the upload ID and cause.
                tracing::warn!(
                    target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
                    upload_id = %upload_id,
                    error = %error,
                    "CompleteMultipartUpload failed"
                );
                return self.complete_mpu_failed(error);
            }
        };
        self.inner.observability.finish_multipart(
            final_snapshot.with_state(UploadExecutionState::CompleteMultipartUploadInFlight),
            completion_timer.elapsed(),
        );

        // The output carries the transfer's final metrics; see the PutObject
        // completion for the ordering.
        self.inner.ctx.finalize_terminal_metrics();
        let result = crate::sdk_v1::update_upload_output_from_complete_multipart_upload(
            response_builder,
            &resp,
        )
        .metrics(self.inner.ctx.metrics())
        .build()
        .expect("valid response");

        *self.inner.result.lock().expect("lock poisoned") = Some(UploadResult::Completed(result));
        let snapshot = self.set_done();
        self.inner.ctx.set_completed();
        self.report_terminal(snapshot);
        self.signal_terminal_when_drained();

        WorkOutcome::Success { data: None }
    }

    /// Handles a CompleteMultipartUpload error, or an error that stopped
    /// Complete from being sent.
    ///
    /// During a caller's abort, `NoSuchUpload` is taken to mean the abort won,
    /// and the upload ends `Cancelled`: S3 returns `NoSuchUpload` to a Complete
    /// sent after AbortMultipartUpload. It does not prove that no object
    /// exists: a Complete retried after an earlier attempt committed can also
    /// be answered with `NoSuchUpload`. Every other error, and any error
    /// outside an abort, fails the transfer, which aborts the upload.
    fn complete_mpu_failed(&self, error: Error) -> WorkOutcome {
        let no_such_upload = error.code() == Some("NoSuchUpload");
        let mut state = self.inner.state.lock().expect("lock poisoned");
        if no_such_upload {
            if let UploadState::Aborting {
                complete_mpu_in_flight,
                ..
            } = &mut *state
            {
                *complete_mpu_in_flight = false;
                drop(state); // unlock before waking
                self.inner.ctx.try_wake(); // the `Aborting` arm ends the transfer
                return WorkOutcome::Success { data: None };
            }
        }
        // Complete's response is in (or it was never sent), so the abort does
        // not wait for it.
        self.fail(state, error, FailedRequest::CompleteMpu)
    }

    /// Sends AbortMultipartUpload for an upload in `UploadState::Aborting`.
    async fn execute_abort_mpu(&self, upload_id: String) -> WorkOutcome {
        let ctx = &self.inner.ctx;
        // Complete committed while this request waited for a slot (always the
        // case at concurrency 1, where Complete holds the only slot). The
        // object exists and `abort()` reports `AlreadyCompleted`, so there is
        // no upload left to abort.
        if ctx.is_completed() {
            return WorkOutcome::Success { data: None };
        }
        let response = crate::sdk_v1::copy_upload_input_fields_to_abort_multipart_upload(
            &self.inner.request,
            ctx.s3_client()
                .abort_multipart_upload()
                .upload_id(&upload_id),
        )
        .customize()
        .config_override(
            ctx.handle
                .bucket_partition_override(self.inner.request.bucket()),
        )
        .send()
        .instrument(tracing::trace_span!(
            target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
            "send-abort-multipart-upload",
        ))
        .await;

        match response {
            Ok(resp) => {
                // Taken alone, never while holding the state lock, so the two
                // mutexes have no ordering to get wrong. A `Completed` result
                // already stored is kept: Complete committed first.
                self.inner
                    .result
                    .lock()
                    .expect("lock poisoned")
                    .get_or_insert(UploadResult::Aborted(AbortedUpload {
                        upload_id: Some(upload_id),
                        request_charged: resp
                            .request_charged
                            .as_ref()
                            .map(crate::sdk_v1::request_charged_from_sdk),
                        ..Default::default()
                    }));
                let mut state = self.inner.state.lock().expect("lock poisoned");
                // `Done` here means the upload ended first; nothing to record.
                if let UploadState::Aborting { abort_mpu, .. } = &mut *state {
                    *abort_mpu = AbortMpuProgress::Done;
                }
                drop(state);
                ctx.try_wake();
                WorkOutcome::Success { data: None }
            }
            Err(e) => {
                if ctx.is_completed() {
                    // Complete committed first, so the abort had nothing to do
                    // and its error says nothing about the object.
                    return WorkOutcome::Success { data: None };
                }
                let error = crate::error::service_error(e);
                tracing::warn!(
                    target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
                    upload_id = %upload_id,
                    error = %error,
                    "AbortMultipartUpload failed; the multipart upload may still be open"
                );
                let state = self.inner.state.lock().expect("lock poisoned");
                // Not sent again. The transfer ends as failed, with this
                // error unless an earlier failure started the abort.
                self.fail(state, error, FailedRequest::AbortMpu)
            }
        }
    }

    /// Handles an error from any request or source read: records `error` and
    /// starts the abort.
    ///
    /// The transfer moves to `Aborting`, which interrupts parts and PutObject
    /// still running, sends AbortMultipartUpload for an open upload unless
    /// the policy is to retain it, and waits for every request still on the
    /// wire. Then it ends as failed with the first recorded failure. An error
    /// that arrives after the upload ended (`Done`) is dropped.
    ///
    /// Returns `WorkOutcome::Failed` while the transfer is still active, so
    /// the scheduler sees the error classification.
    ///
    /// `state` is the held state lock; it is released before the abort is
    /// signalled. `failed` names the request that failed, so `Aborting` stops
    /// waiting for it.
    fn fail(
        &self,
        state: MutexGuard<'_, UploadState>,
        error: Error,
        failed: FailedRequest,
    ) -> WorkOutcome {
        tracing::debug!(
            target: crate::telemetry::TARGET_TRANSFER_UPLOAD,
            %error,
            "upload failed",
        );
        let classification = crate::scheduler::classify_error(&error);
        let snapshot = snapshot_state(&state);
        self.enter_aborting(state, Some((UploadFailure { error, snapshot }, failed)));
        self.signal_abort();
        WorkOutcome::Failed { classification }
    }
}

/// The request that failed, so that `Aborting` stops waiting for it.
enum FailedRequest {
    /// A part, a source read or PutObject. Each is counted by its own
    /// `InFlight`, so `Aborting` tracks nothing else for it.
    Other,
    /// CreateMultipartUpload returned. `upload_id` is set when S3 created the
    /// upload before the failure, so that upload is aborted.
    CreateMpu { upload_id: Option<String> },
    /// CompleteMultipartUpload returned, or was never sent.
    CompleteMpu,
    /// AbortMultipartUpload returned. It is not sent again.
    AbortMpu,
}

/// Returns a copyable view of upload execution while the state lock is held.
fn snapshot_state(state: &UploadState) -> UploadStateSnapshot {
    match state {
        UploadState::PendingInit {
            init_in_flight: false,
            ..
        } => UploadStateSnapshot::inactive(UploadExecutionState::PendingInitialization),
        UploadState::PendingInit {
            init_in_flight: true,
            ..
        } => UploadStateSnapshot::inactive(UploadExecutionState::CreateMultipartUploadInFlight),
        UploadState::Transferring { parts, .. } => parts.snapshot(),
        UploadState::Completing { parts, .. } => parts
            .snapshot()
            .with_state(UploadExecutionState::MultipartCompletionPending),
        // The part state moved into the dispatched work item, so no part
        // measurements are available here.
        UploadState::CompleteInFlight { .. } => {
            UploadStateSnapshot::inactive(UploadExecutionState::CompleteMultipartUploadInFlight)
        }
        UploadState::PutObjectInFlight => {
            UploadStateSnapshot::inactive(UploadExecutionState::PutObjectInFlight)
        }
        UploadState::Aborting { .. } => {
            UploadStateSnapshot::inactive(UploadExecutionState::Aborting)
        }
        UploadState::Done => UploadStateSnapshot::inactive(UploadExecutionState::Done),
    }
}

/// Moves a fully drained multipart transfer into its completion phase.
///
/// The state replacement elects exactly one caller even when source EOF and a network completion
/// race to retire the final work item.
/// Checks that a successful `UploadPart` response carries what completing the upload needs.
///
/// `CompleteMultipartUpload` identifies each part by its ETag, so a response without one, or with
/// an empty one, is a failed part: the completed list could not identify it, and the object would
/// be stored without that evidence. aws-c-s3 treats it the same way
/// (`AWS_ERROR_S3_MISSING_ETAG`). The error carries the operation name and the response's request
/// ids, as an error from a failed request does.
fn validate_upload_part_response(
    resp: aws_sdk_s3::operation::upload_part::UploadPartOutput,
    part_number: u64,
) -> Result<aws_sdk_s3::operation::upload_part::UploadPartOutput, Error> {
    if resp.e_tag().is_some_and(|e_tag| !e_tag.is_empty()) {
        return Ok(resp);
    }
    Err(crate::error::invalid_service_response(
        "UploadPart",
        &resp,
        format!("UploadPart returned no ETag for part {part_number}"),
    ))
}

/// Returns `WorkOutcome::Cancelled` for part work that succeeded, or produced
/// nothing, after the state left `Transferring`.
///
/// After an abort (by the caller or by a failure), a part already past its
/// last `.await` can still reach one of its state locks. There is no part
/// state left to record its result in, so the result is dropped. Part errors
/// never reach here; they go to `fail()`.
///
/// # Panics
///
/// Panics unless `state` is `Aborting` or `Done`; any other state is a bug.
fn abandoned_part(state: &UploadState, step: &str) -> WorkOutcome {
    assert!(
        matches!(state, UploadState::Aborting { .. } | UploadState::Done),
        "unexpected state {step}"
    );
    WorkOutcome::Cancelled
}

/// Returns the ID of the multipart upload `state` leaves open if the upload
/// ends now: one created, not committed, and not closed by
/// AbortMultipartUpload.
fn open_upload_id(state: &UploadState) -> Option<String> {
    match state {
        UploadState::Transferring { upload_id, .. }
        | UploadState::Completing { upload_id, .. }
        | UploadState::CompleteInFlight { upload_id } => Some(upload_id.clone()),
        UploadState::Aborting {
            upload_id,
            abort_mpu,
            ..
        } if *abort_mpu != AbortMpuProgress::Done => upload_id.clone(),
        _ => None,
    }
}

fn try_begin_completing(state: &mut UploadState) -> bool {
    let ready = matches!(
        state,
        UploadState::Transferring { parts, .. } if parts.is_complete()
    );
    if !ready {
        return false;
    }

    // Move the fields out by value; the placeholder is overwritten before the state lock is
    // released, so no reader observes it.
    let UploadState::Transferring {
        upload_id,
        parts,
        response_builder,
    } = std::mem::replace(state, UploadState::Done)
    else {
        unreachable!("completion readiness changed under lock");
    };
    *state = UploadState::Completing {
        upload_id,
        parts,
        response_builder,
    };
    true
}

impl Transfer for UploadTransfer {
    fn ctx(&self) -> &TransferContext {
        UploadTransfer::ctx(self)
    }

    fn poll_work(&self) -> PollWork {
        UploadTransfer::poll_work(self)
    }

    fn execute<'a>(
        &'a self,
        work: &'a mut IoRequest,
    ) -> Pin<Box<dyn Future<Output = WorkOutcome> + Send + 'a>> {
        Box::pin(UploadTransfer::execute(self, work))
    }

    fn on_terminal(&self) {
        // Terminal cleanup runs even if a panic poisoned the state lock.
        let mut state = self
            .inner
            .state
            .lock()
            .unwrap_or_else(std::sync::PoisonError::into_inner);
        // The summary reports the state the transfer ended in.
        let snapshot = snapshot_state(&state);
        // An ended transfer is never polled again, so nothing sends
        // AbortMultipartUpload for it.
        let left_open = open_upload_id(&state);
        // Nothing is left to drive. Work still executing finds `Done` and
        // stands down.
        let previous = std::mem::replace(&mut *state, UploadState::Done);
        drop(state);
        // Interrupted parts drop the part readers they hold.
        self.inner.abort_parts_token.cancel();
        if let Some(upload_id) = left_open {
            self.warn_upload_left_open(&upload_id);
            // A failure forced from outside, such as a worker panic, is stored
            // before this hook runs and before the terminal signal `join()`
            // waits on, so the ID reaches the error `join()` returns.
            if self.inner.ctx.is_failed() {
                self.inner.ctx.amend_error(|err| match err.upload_id() {
                    Some(_) => err,
                    None => err.with_upload_id(upload_id),
                });
            }
        }
        self.report_terminal(snapshot);
        // Last, after the lock and the transfer's own cleanup: it can own the
        // caller's `PartStream`, whose `Drop` is caller code.
        drop(previous);
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::io::InputStream;
    use crate::scheduler::test_util::{assert_done, assert_pending, assert_ready};
    use aws_sdk_s3::operation::abort_multipart_upload::AbortMultipartUploadOutput;
    use aws_sdk_s3::operation::complete_multipart_upload::CompleteMultipartUploadOutput;
    use aws_sdk_s3::operation::create_multipart_upload::CreateMultipartUploadOutput;
    use aws_sdk_s3::operation::upload_part::UploadPartOutput;
    use aws_smithy_mocks::{mock, mock_client, RuleMode};
    use aws_smithy_runtime_api::client::orchestrator::HttpResponse;
    use aws_smithy_runtime_api::http::StatusCode;
    use aws_smithy_types::body::SdkBody;

    fn create_test_transfer(s3_client: aws_sdk_s3::Client, content: Vec<u8>) -> UploadTransfer {
        create_test_transfer_with_stream(s3_client, InputStream::from(content))
    }

    fn create_test_transfer_with_stream(
        s3_client: aws_sdk_s3::Client,
        stream: InputStream,
    ) -> UploadTransfer {
        create_test_transfer_with_detail(s3_client, stream, 2)
    }

    fn create_test_transfer_with_detail(
        s3_client: aws_sdk_s3::Client,
        stream: InputStream,
        detail: u64,
    ) -> UploadTransfer {
        create_test_transfer_with_policy(s3_client, stream, detail, None)
    }

    fn create_test_transfer_with_policy(
        s3_client: aws_sdk_s3::Client,
        stream: InputStream,
        detail: u64,
        policy: Option<FailedMultipartUploadPolicy>,
    ) -> UploadTransfer {
        let handle = crate::client::Handle::test_handle_tokio(
            crate::Config::builder()
                .sdk_client(s3_client)
                .diagnostics_for_test(crate::config::MemoryDiagnosticsConfig::default(), detail)
                .build()
                .expect("valid config"),
        );

        let mut input = UploadInput::builder()
            .bucket("test-bucket")
            .key("test-key")
            .body(stream)
            .set_failed_multipart_upload_policy(policy)
            .build()
            .unwrap();
        let stream = input.take_body();

        let (ctx, _completion_rx) = TransferContext::new(handle);
        UploadTransfer::try_new(ctx, BucketType::Standard, input, stream).unwrap()
    }

    fn mock_s3_client_for_mpu() -> aws_sdk_s3::Client {
        let create_mpu = mock!(aws_sdk_s3::Client::create_multipart_upload).then_output(|| {
            CreateMultipartUploadOutput::builder()
                .upload_id("test-upload-id")
                .build()
        });

        let upload_part = mock!(aws_sdk_s3::Client::upload_part)
            .then_output(|| UploadPartOutput::builder().e_tag("test-etag").build());

        let complete_mpu = mock!(aws_sdk_s3::Client::complete_multipart_upload).then_output(|| {
            CompleteMultipartUploadOutput::builder()
                .e_tag("final-etag")
                .build()
        });

        mock_client!(
            aws_sdk_s3,
            RuleMode::MatchAny,
            &[create_mpu, upload_part, complete_mpu]
        )
    }

    fn abort_mpu_ok() -> aws_smithy_mocks::Rule {
        mock!(aws_sdk_s3::Client::abort_multipart_upload)
            .then_output(|| AbortMultipartUploadOutput::builder().build())
    }

    /// Runs the AbortMultipartUpload a failure emits, then checks the
    /// transfer has ended.
    async fn finish_failure_abort(transfer: &UploadTransfer) {
        let mut abort = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut abort).await,
            WorkOutcome::Success { .. }
        ));
        assert_done(transfer.poll_work());
    }

    fn service_error_response() -> HttpResponse {
        HttpResponse::new(
            StatusCode::try_from(412).unwrap(),
            SdkBody::from("<Error><Code>PreconditionFailed</Code></Error>"),
        )
    }

    /// A 403 `AccessDenied`, so a test can tell it apart from the 412 in
    /// `service_error_response`.
    fn access_denied_response() -> HttpResponse {
        HttpResponse::new(
            StatusCode::try_from(403).unwrap(),
            SdkBody::from("<Error><Code>AccessDenied</Code></Error>"),
        )
    }

    fn create_mpu_ok() -> aws_smithy_mocks::Rule {
        mock!(aws_sdk_s3::Client::create_multipart_upload).then_output(|| {
            CreateMultipartUploadOutput::builder()
                .upload_id("test-upload-id")
                .build()
        })
    }

    /// AbortMultipartUpload that succeeds only for `test-upload-id`.
    fn abort_mpu_for_test_upload() -> aws_smithy_mocks::Rule {
        mock!(aws_sdk_s3::Client::abort_multipart_upload)
            .match_requests(|req| req.upload_id() == Some("test-upload-id"))
            .then_output(|| AbortMultipartUploadOutput::builder().build())
    }

    /// The `Aborting` fields that decide when and how the abort ends.
    #[derive(Debug, PartialEq)]
    struct AbortingView {
        upload_id: Option<String>,
        create_in_flight: bool,
        complete_mpu_in_flight: bool,
        abort_mpu: AbortMpuProgress,
        failure_recorded: bool,
    }

    /// Panics if the transfer is not in `Aborting`.
    fn aborting_view(transfer: &UploadTransfer) -> AbortingView {
        let state = transfer.inner.state.lock().expect("lock poisoned");
        let UploadState::Aborting {
            upload_id,
            create_in_flight,
            complete_mpu_in_flight,
            abort_mpu,
            failure,
        } = &*state
        else {
            panic!("expected Aborting, got {:?}", snapshot_state(&state).state);
        };
        AbortingView {
            upload_id: upload_id.clone(),
            create_in_flight: *create_in_flight,
            complete_mpu_in_flight: *complete_mpu_in_flight,
            abort_mpu: *abort_mpu,
            failure_recorded: failure.is_some(),
        }
    }

    /// The error the transfer ended with.
    fn final_error(transfer: &UploadTransfer) -> crate::error::Error {
        transfer
            .ctx()
            .take_error()
            .expect("transfer should end as failed")
    }

    #[test]
    fn upload_state_snapshot_classifies_non_multipart_execution() {
        let pending = UploadState::PendingInit {
            stream: None,
            size_hint: crate::io::SizeHint::exact(0),
            init_in_flight: false,
        };
        assert_eq!(
            snapshot_state(&pending),
            UploadStateSnapshot::inactive(UploadExecutionState::PendingInitialization)
        );

        let creating = UploadState::PendingInit {
            stream: None,
            size_hint: crate::io::SizeHint::exact(0),
            init_in_flight: true,
        };
        assert_eq!(
            snapshot_state(&creating),
            UploadStateSnapshot::inactive(UploadExecutionState::CreateMultipartUploadInFlight)
        );

        let completion_in_flight = UploadState::CompleteInFlight {
            upload_id: "test-upload-id".to_string(),
        };
        assert_eq!(
            snapshot_state(&completion_in_flight),
            UploadStateSnapshot::inactive(UploadExecutionState::CompleteMultipartUploadInFlight)
        );
        assert_eq!(
            snapshot_state(&UploadState::PutObjectInFlight),
            UploadStateSnapshot::inactive(UploadExecutionState::PutObjectInFlight)
        );
        assert_eq!(
            snapshot_state(&UploadState::Done),
            UploadStateSnapshot::inactive(UploadExecutionState::Done)
        );
    }

    #[derive(Debug)]
    struct BlockingPartStream {
        ready: Arc<std::sync::atomic::AtomicBool>,
        source_waker: Arc<Mutex<Option<std::task::Waker>>>,
        yielded: bool,
        /// Dropped with the stream, so a `Weak` to it reports whether the stream is still owned.
        _probe: Option<Arc<()>>,
    }

    impl crate::io::PartStream for BlockingPartStream {
        fn poll_part(
            mut self: Pin<&mut Self>,
            cx: &mut std::task::Context<'_>,
            _stream_cx: &crate::io::StreamContext,
        ) -> std::task::Poll<Option<std::io::Result<PartData>>> {
            if self.yielded {
                return std::task::Poll::Ready(None);
            }
            if self.ready.load(std::sync::atomic::Ordering::Acquire) {
                self.yielded = true;
                return std::task::Poll::Ready(Some(Ok(PartData::new(
                    1,
                    Bytes::from_static(b"ready"),
                ))));
            }
            *self.source_waker.lock().expect("source waker poisoned") = Some(cx.waker().clone());
            std::task::Poll::Pending
        }

        fn size_hint(&self) -> crate::io::SizeHint {
            crate::io::SizeHint::exact(32 * 1024 * 1024)
        }
    }

    #[derive(Debug)]
    struct FailingPartStream;

    impl crate::io::PartStream for FailingPartStream {
        fn poll_part(
            self: Pin<&mut Self>,
            _cx: &mut std::task::Context<'_>,
            _stream_cx: &crate::io::StreamContext,
        ) -> std::task::Poll<Option<std::io::Result<PartData>>> {
            std::task::Poll::Ready(Some(Err(std::io::Error::other("injected source failure"))))
        }

        fn size_hint(&self) -> crate::io::SizeHint {
            crate::io::SizeHint::exact(5)
        }
    }

    #[cfg_attr(miri, ignore)]
    #[test]
    fn test_poll_work_unknown_length_routes_to_multipart() {
        #[derive(Debug)]
        struct EmptyUnknown;

        impl crate::io::PartStream for EmptyUnknown {
            fn poll_part(
                self: Pin<&mut Self>,
                _cx: &mut std::task::Context<'_>,
                _stream_cx: &crate::io::StreamContext,
            ) -> std::task::Poll<Option<std::io::Result<PartData>>> {
                std::task::Poll::Ready(None)
            }

            fn size_hint(&self) -> crate::io::SizeHint {
                crate::io::SizeHint::default()
            }
        }

        let transfer = create_test_transfer_with_stream(
            mock_client!(aws_sdk_s3, []),
            InputStream::from_part_stream(EmptyUnknown),
        );
        let mut work = assert_ready(transfer.poll_work());
        assert!(matches!(
            work.data_mut::<UploadWork>(),
            UploadWork::CreateMPU(_)
        ));
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn empty_unknown_length_stream_uploads_one_empty_part() {
        #[derive(Debug)]
        struct EmptyUnknown;

        impl crate::io::PartStream for EmptyUnknown {
            fn poll_part(
                self: Pin<&mut Self>,
                _cx: &mut std::task::Context<'_>,
                _stream_cx: &crate::io::StreamContext,
            ) -> std::task::Poll<Option<std::io::Result<PartData>>> {
                std::task::Poll::Ready(None)
            }

            fn size_hint(&self) -> crate::io::SizeHint {
                crate::io::SizeHint::default()
            }
        }

        let transfer = create_test_transfer_with_stream(
            mock_s3_client_for_mpu(),
            InputStream::from_part_stream(EmptyUnknown),
        );

        let mut create = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut create).await,
            WorkOutcome::Success { .. }
        ));

        let mut upload = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut upload).await,
            WorkOutcome::Success { .. }
        ));

        let mut empty_object = assert_ready(transfer.poll_work());
        let UploadWork::UploadPart(work, _) = empty_object.data_mut::<UploadWork>() else {
            panic!("drained empty source did not schedule its required zero-byte part");
        };
        assert!(work.is_empty_object());
        assert!(matches!(
            transfer.execute(&mut empty_object).await,
            WorkOutcome::Success { .. }
        ));

        let mut complete = assert_ready(transfer.poll_work());
        assert!(matches!(
            complete.data_mut::<UploadWork>(),
            UploadWork::CompleteMPU(..)
        ));
        assert!(matches!(
            transfer.execute(&mut complete).await,
            WorkOutcome::Success { .. }
        ));

        crate::scheduler::test_util::assert_done(transfer.poll_work());
        assert!(transfer.take_result().is_some());
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn end_of_stream_waits_for_an_earlier_source_part_to_retire() {
        #[derive(Debug)]
        struct UnknownSource;

        impl crate::io::PartStream for UnknownSource {
            fn poll_part(
                self: Pin<&mut Self>,
                _cx: &mut std::task::Context<'_>,
                _stream_cx: &crate::io::StreamContext,
            ) -> std::task::Poll<Option<std::io::Result<PartData>>> {
                unreachable!("this test drives transfer accounting directly")
            }

            fn size_hint(&self) -> crate::io::SizeHint {
                crate::io::SizeHint::default()
            }
        }

        let transfer = create_test_transfer_with_stream(
            mock_s3_client_for_mpu(),
            InputStream::from_part_stream(UnknownSource),
        );
        let mut create = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut create).await,
            WorkOutcome::Success { .. }
        ));

        let mut state = transfer.inner.state.lock().expect("lock poisoned");
        let UploadState::Transferring { parts, .. } = &mut *state else {
            panic!("upload did not enter transferring state");
        };

        // Model the ordering that exposed the race:
        //
        //   part N returns data ── gate opens ── part N+1 observes EOF
        //          │                                  │
        //          └──── accounting is delayed ───────┘
        //
        // EOF retires its own work first. The earlier source part remains in flight, so the
        // transfer must not mistake the temporarily empty accounting state for an empty source.
        let mut earlier_part = parts.schedule_part().expect("earlier source part");
        let _end_of_stream = parts.schedule_part().expect("end-of-stream read");
        assert!(parts.record_end_of_stream());
        assert!(!parts.record_end_of_stream());
        let empty_object_ready = parts.finish_read_without_part();

        assert!(!empty_object_ready);
        assert!(!parts.is_complete());

        parts.observe_part(5).unwrap();
        parts.begin_upload(1, earlier_part.take_timing());
        parts.complete_part(CompletedPart::builder().part_number(1).build(), 5);

        assert!(parts.maybe_start_empty_object_part().is_none());
        assert!(parts.is_complete());
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn blocked_custom_source_retains_one_read_and_retracts_waiters() {
        let ready = Arc::new(std::sync::atomic::AtomicBool::new(false));
        let source_waker = Arc::new(Mutex::new(None));
        let stream = BlockingPartStream {
            ready: Arc::clone(&ready),
            source_waker: Arc::clone(&source_waker),
            yielded: false,
            _probe: None,
        };
        let transfer = create_test_transfer_with_stream(
            mock_s3_client_for_mpu(),
            InputStream::from_part_stream(stream),
        );

        let mut create = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut create).await,
            WorkOutcome::Success { .. }
        ));

        let mut first = assert_ready(transfer.poll_work());
        let mut second = assert_ready(transfer.poll_work());
        let mut third = assert_ready(transfer.poll_work());

        assert!(matches!(
            transfer.execute(&mut first).await,
            WorkOutcome::Yielded
        ));
        {
            let state = transfer.inner.state.lock().expect("lock poisoned");
            let UploadState::Transferring { parts, .. } = &*state else {
                panic!("upload left transferring state");
            };
            assert_eq!(parts.test_counts(), (3, 3, 1));
            assert_eq!(
                parts.snapshot(),
                UploadStateSnapshot {
                    state: UploadExecutionState::Transferring,
                    parts_dispatched: 3,
                    parts_in_flight: 3,
                    uploads_in_flight: 0,
                    pending_reads: 1,
                    completed_parts: 0,
                    bytes_read: 0,
                    bytes_uploaded: 0,
                    eof: false,
                    dispatch_closed: false,
                }
            );
        }

        assert!(matches!(
            transfer.execute(&mut second).await,
            WorkOutcome::Yielded
        ));
        assert!(matches!(
            transfer.execute(&mut third).await,
            WorkOutcome::Yielded
        ));
        {
            let state = transfer.inner.state.lock().expect("lock poisoned");
            let UploadState::Transferring { parts, .. } = &*state else {
                panic!("upload left transferring state");
            };
            assert_eq!(parts.test_counts(), (1, 1, 1));
            let summary = parts.test_summary();
            assert_eq!(summary.read_pending_polls, 1);
            assert_eq!(summary.read_pending_parts, 0);
        }
        assert_pending(transfer.poll_work());

        ready.store(true, std::sync::atomic::Ordering::Release);
        source_waker
            .lock()
            .expect("source waker poisoned")
            .take()
            .expect("source did not register a waker")
            .wake();

        let mut resumed = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut resumed).await,
            WorkOutcome::Success { .. }
        ));
        {
            let state = transfer.inner.state.lock().expect("lock poisoned");
            let UploadState::Transferring { parts, .. } = &*state else {
                panic!("upload left transferring state");
            };
            assert_eq!(parts.test_counts(), (1, 0, 0));
            let summary = parts.test_summary();
            assert_eq!(summary.parts_completed, 1);
            assert_eq!(summary.bytes_uploaded, 5);
            assert_eq!(summary.read_pending_polls, 1);
            assert_eq!(summary.read_pending_parts, 1);
            assert_eq!(summary.presentation_segments, 1);
            assert_eq!(summary.max_presentation_segments, 1);
        }

        let mut next = assert_ready(transfer.poll_work());
        assert!(matches!(
            next.data_mut::<UploadWork>(),
            UploadWork::UploadPart(..)
        ));
    }

    // FIXME: crossbeam-epoch is incompatible with miri (https://github.com/crossbeam-rs/crossbeam/issues/1181)
    #[cfg_attr(miri, ignore)]
    #[test]
    fn test_poll_work_initial_state_returns_create_mpu() {
        let s3_client = mock_client!(aws_sdk_s3, []);
        let content = vec![0u8; 16 * 1024 * 1024];
        let transfer = create_test_transfer(s3_client, content);

        let mut work = assert_ready(transfer.poll_work());
        let data = work.data_mut::<UploadWork>();
        assert!(matches!(data, UploadWork::CreateMPU(_)));
    }

    // FIXME: crossbeam-epoch is incompatible with miri (https://github.com/crossbeam-rs/crossbeam/issues/1181)
    #[cfg_attr(miri, ignore)]
    #[test]
    fn test_poll_work_pending_while_init_in_flight() {
        let s3_client = mock_client!(aws_sdk_s3, []);
        let content = vec![0u8; 16 * 1024 * 1024];
        let transfer = create_test_transfer(s3_client, content);

        let _work = assert_ready(transfer.poll_work());
        assert_pending(transfer.poll_work());
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_poll_work_generates_parts_after_create_mpu() {
        let s3_client = mock_s3_client_for_mpu();
        let content = vec![0u8; 16 * 1024 * 1024];
        let transfer = create_test_transfer(s3_client, content);

        let mut work = assert_ready(transfer.poll_work());
        transfer.execute(&mut work).await;

        let mut work1 = assert_ready(transfer.poll_work());
        let data1 = work1.data_mut::<UploadWork>();
        assert!(matches!(data1, UploadWork::UploadPart(..)));

        let mut work2 = assert_ready(transfer.poll_work());
        let data2 = work2.data_mut::<UploadWork>();
        assert!(matches!(data2, UploadWork::UploadPart(..)));

        assert_pending(transfer.poll_work());
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_execute_create_mpu_transitions_to_transferring() {
        let s3_client = mock_s3_client_for_mpu();
        let content = vec![0u8; 16 * 1024 * 1024];
        let transfer = create_test_transfer(s3_client, content);

        let mut work = assert_ready(transfer.poll_work());

        let outcome = transfer.execute(&mut work).await;
        assert!(matches!(outcome, WorkOutcome::Success { .. }));

        let mut next = assert_ready(transfer.poll_work());
        let data = next.data_mut::<UploadWork>();
        assert!(matches!(data, UploadWork::UploadPart(..)));
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_execute_full_mpu_flow() {
        let s3_client = mock_s3_client_for_mpu();
        let content = vec![0u8; 16 * 1024 * 1024];
        let transfer = create_test_transfer(s3_client, content);

        // 1. CreateMPU
        let mut work = assert_ready(transfer.poll_work());
        transfer.execute(&mut work).await;

        // 2. UploadPart 1 (read+send in one call)
        let mut work = assert_ready(transfer.poll_work());
        transfer.execute(&mut work).await;

        // 3. UploadPart 2 (read+send in one call)
        let mut work = assert_ready(transfer.poll_work());
        transfer.execute(&mut work).await;

        {
            let state = transfer.inner.state.lock().expect("lock poisoned");
            let UploadState::Completing { parts, .. } = &*state else {
                panic!("upload did not enter completing state");
            };
            assert_eq!(
                snapshot_state(&state).state,
                UploadExecutionState::MultipartCompletionPending
            );
            let summary = parts.test_summary();
            assert_eq!(summary.parts_completed, 2);
            assert_eq!(summary.last_active_snapshot.parts_in_flight, 0);
            assert_eq!(summary.last_active_snapshot.uploads_in_flight, 0);
            assert_eq!(summary.last_active_snapshot.pending_reads, 0);
            assert_eq!(summary.segmented_parts, 0);
            assert_eq!(summary.presentation_segments, 2);
            assert_eq!(summary.max_presentation_segments, 1);
            assert_eq!(summary.read_pending_polls, 0);
            assert_eq!(summary.read_pending_parts, 0);
        }

        // 4. CompleteMPU
        let mut work = assert_ready(transfer.poll_work());
        let data = work.data_mut::<UploadWork>();
        assert!(matches!(data, UploadWork::CompleteMPU(..)));
        transfer.execute(&mut work).await;

        // 5. Should be Done
        use crate::scheduler::test_util::assert_done;
        assert_done(transfer.poll_work());

        // 6. Result should be available
        let result = transfer.take_result();
        assert!(result.is_some());

        let request_metrics = transfer.ctx().metrics.request_metrics();
        assert_eq!(request_metrics.requests, 4);
        assert_eq!(request_metrics.retry_reissues, 0);
        assert_eq!(request_metrics.throttle_reissues, 0);
        assert_eq!(request_metrics.hedge_reissues, 0);

        let summary = transfer
            .inner
            .observability
            .test_terminal_summary()
            .expect("completed MPU should emit a terminal summary");
        assert_eq!(summary.outcome, UploadTerminalOutcome::Completed);
        assert_eq!(summary.mode, UploadMode::Multipart);
        assert_eq!(summary.request_total.requests, 4);
        assert_eq!(summary.requests.create_multipart_upload.requests, 1);
        assert_eq!(summary.requests.upload_part.requests, 2);
        assert_eq!(summary.requests.complete_multipart_upload.requests, 1);
        assert_eq!(
            summary
                .multipart
                .expect("multipart summary")
                .parts_completed,
            2
        );
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_basic_upload_object() {
        use aws_sdk_s3::operation::put_object::PutObjectOutput;
        let put_object = mock!(aws_sdk_s3::Client::put_object)
            .then_output(|| PutObjectOutput::builder().e_tag("test-etag").build());
        let s3_client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[put_object]);
        // Small content below MPU threshold (default 8MB)
        let content = vec![0u8; 1024];
        let transfer = create_test_transfer(s3_client, content);

        let mut work = assert_ready(transfer.poll_work());
        let data = work.data_mut::<UploadWork>();
        assert!(matches!(data, UploadWork::PutObject { .. }));

        let outcome = transfer.execute(&mut work).await;
        assert!(matches!(outcome, WorkOutcome::Success { .. }));

        assert!(transfer.take_result().is_some());

        // PutObject must record network_tx so the adaptive controller sees throughput
        assert!(
            !transfer.ctx().handle.telemetry.io_counters.is_idle(),
            "PutObject should record network_tx to IOCounters"
        );
        assert_eq!(transfer.ctx().metrics.request_metrics().requests, 1);

        let summary = transfer
            .inner
            .observability
            .test_terminal_summary()
            .expect("completed PutObject should emit a terminal summary");
        assert_eq!(summary.outcome, UploadTerminalOutcome::Completed);
        assert_eq!(summary.mode, UploadMode::PutObject);
        assert_eq!(summary.request_total.requests, 1);
        assert_eq!(summary.requests.put_object.requests, 1);
        assert_eq!(summary.multipart, None);
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn diagnostic_levels_preserve_put_object_outcome() {
        for detail in 0..=2 {
            let put_object = mock!(aws_sdk_s3::Client::put_object).then_output(|| {
                aws_sdk_s3::operation::put_object::PutObjectOutput::builder().build()
            });
            let transfer = create_test_transfer_with_detail(
                mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[put_object]),
                InputStream::from(vec![0u8; 1024]),
                detail,
            );

            let mut work = assert_ready(transfer.poll_work());
            assert!(matches!(
                transfer.execute(&mut work).await,
                WorkOutcome::Success { .. }
            ));
            assert_eq!(
                transfer.ctx().transfer_status(),
                crate::types::TransferStatus::Completed
            );
            assert_eq!(
                transfer
                    .inner
                    .observability
                    .test_terminal_summary()
                    .is_some(),
                detail != 0
            );
        }
    }

    // Builds a client handle: dropping the scheduler's ready set is not miri-clean.
    #[cfg_attr(miri, ignore)]
    #[test]
    fn external_terminal_paths_emit_one_upload_summary() {
        let transfer = create_test_transfer(
            mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[]),
            vec![0u8; 1024],
        );
        let _work = assert_ready(transfer.poll_work());
        transfer.ctx().set_cancelled();

        Transfer::on_terminal(&transfer);
        Transfer::on_terminal(&transfer);

        let summary = transfer
            .inner
            .observability
            .test_terminal_summary()
            .expect("external cancellation should emit a terminal summary");
        assert_eq!(summary.outcome, UploadTerminalOutcome::Cancelled);
        assert_eq!(summary.mode, UploadMode::PutObject);
        assert_eq!(summary.request_total.requests, 0);
    }

    // Builds a client handle: dropping the scheduler's ready set is not miri-clean.
    #[cfg_attr(miri, ignore)]
    #[test]
    fn on_terminal_tolerates_poisoned_state() {
        let transfer = create_test_transfer(
            mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[]),
            vec![0u8; 1024],
        );
        let _work = assert_ready(transfer.poll_work());
        std::thread::scope(|scope| {
            let poisoner = scope.spawn(|| {
                let _guard = transfer.inner.state.lock().expect("lock poisoned");
                panic!("unexpected state");
            });
            assert!(poisoner.join().is_err());
        });
        assert!(
            transfer.inner.state.lock().is_err(),
            "state should be poisoned"
        );
        transfer.ctx().set_failed(Error::new(
            ErrorKind::RuntimeError,
            "worker panic during execute",
        ));

        let cleanup = std::panic::catch_unwind(std::panic::AssertUnwindSafe(|| {
            Transfer::on_terminal(&transfer)
        }));
        assert!(
            cleanup.is_ok(),
            "on_terminal must tolerate a poisoned state"
        );

        let summary = transfer
            .inner
            .observability
            .test_terminal_summary()
            .expect("poisoned state should still emit a terminal summary");
        assert_eq!(summary.outcome, UploadTerminalOutcome::Failed);
        let state = transfer
            .inner
            .state
            .lock()
            .unwrap_or_else(std::sync::PoisonError::into_inner);
        assert!(
            matches!(*state, UploadState::Done),
            "on_terminal must replace a poisoned state"
        );
    }

    // Builds a client handle: dropping the scheduler's ready set is not miri-clean.
    #[cfg_attr(miri, ignore)]
    #[test]
    fn dropped_request_measurement_reaches_upload_request_kind_summary() {
        let transfer = create_test_transfer(
            mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[]),
            vec![0u8; 1024],
        );
        {
            let mut measurement = transfer
                .inner
                .observability
                .start_request(&transfer.inner.ctx, UploadRequestKind::PutObject);
            measurement
                .metrics_mut()
                .record_retry_reissue(std::time::Duration::from_micros(3));
        }
        transfer.ctx().set_failed(Error::new(
            ErrorKind::RuntimeError,
            "simulated worker failure",
        ));
        Transfer::on_terminal(&transfer);

        let summary = transfer
            .inner
            .observability
            .test_terminal_summary()
            .expect("worker failure should emit a terminal summary");
        assert_eq!(summary.outcome, UploadTerminalOutcome::Failed);
        assert_eq!(summary.request_total.requests, 1);
        assert_eq!(summary.requests.put_object.requests, 1);
        assert_eq!(summary.requests.put_object.retry_reissues, 1);
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn create_multipart_failure_emits_terminal_summary() {
        let create_mpu = mock!(aws_sdk_s3::Client::create_multipart_upload)
            .then_http_response(service_error_response);
        let transfer = create_test_transfer(
            mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[create_mpu]),
            vec![0u8; 16 * 1024 * 1024],
        );

        let mut work = assert_ready(transfer.poll_work());
        let outcome = transfer.execute(&mut work).await;
        assert!(matches!(outcome, WorkOutcome::Failed { .. }));
        // No upload was created, so there is nothing to abort.
        assert_done(transfer.poll_work());

        let summary = transfer
            .inner
            .observability
            .test_terminal_summary()
            .expect("CreateMultipartUpload failure should emit a terminal summary");
        assert_eq!(summary.outcome, UploadTerminalOutcome::Failed);
        assert_eq!(summary.mode, UploadMode::Multipart);
        assert_eq!(summary.requests.create_multipart_upload.requests, 1);
        assert_eq!(summary.requests.upload_part.requests, 0);
        assert_eq!(summary.multipart, None);
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn put_object_failure_emits_terminal_summary() {
        let put_object =
            mock!(aws_sdk_s3::Client::put_object).then_http_response(service_error_response);
        let transfer = create_test_transfer(
            mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[put_object]),
            vec![0u8; 1024],
        );

        let mut work = assert_ready(transfer.poll_work());
        let outcome = transfer.execute(&mut work).await;
        assert!(matches!(outcome, WorkOutcome::Failed { .. }));
        assert_done(transfer.poll_work());

        let summary = transfer
            .inner
            .observability
            .test_terminal_summary()
            .expect("PutObject failure should emit a terminal summary");
        assert_eq!(summary.outcome, UploadTerminalOutcome::Failed);
        assert_eq!(summary.mode, UploadMode::PutObject);
        assert_eq!(summary.requests.put_object.requests, 1);
        assert_eq!(summary.multipart, None);
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn source_failure_emits_terminal_summary() {
        let create_mpu = mock!(aws_sdk_s3::Client::create_multipart_upload).then_output(|| {
            CreateMultipartUploadOutput::builder()
                .upload_id("test-upload-id")
                .build()
        });
        let transfer = create_test_transfer_with_stream(
            mock_client!(
                aws_sdk_s3,
                RuleMode::MatchAny,
                &[create_mpu, abort_mpu_ok()]
            ),
            InputStream::from_part_stream(FailingPartStream),
        );

        let mut create = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut create).await,
            WorkOutcome::Success { .. }
        ));
        let mut part = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut part).await,
            WorkOutcome::Failed { .. }
        ));
        finish_failure_abort(&transfer).await;

        let summary = transfer
            .inner
            .observability
            .test_terminal_summary()
            .expect("source failure should emit a terminal summary");
        assert_eq!(summary.outcome, UploadTerminalOutcome::Failed);
        assert_eq!(summary.requests.create_multipart_upload.requests, 1);
        assert_eq!(summary.requests.upload_part.requests, 0);
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn upload_part_failure_emits_terminal_summary() {
        let create_mpu = mock!(aws_sdk_s3::Client::create_multipart_upload).then_output(|| {
            CreateMultipartUploadOutput::builder()
                .upload_id("test-upload-id")
                .build()
        });
        let upload_part =
            mock!(aws_sdk_s3::Client::upload_part).then_http_response(service_error_response);
        let transfer = create_test_transfer(
            mock_client!(
                aws_sdk_s3,
                RuleMode::MatchAny,
                &[create_mpu, upload_part, abort_mpu_ok()]
            ),
            vec![0u8; 16 * 1024 * 1024],
        );

        let mut create = assert_ready(transfer.poll_work());
        transfer.execute(&mut create).await;
        let mut part = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut part).await,
            WorkOutcome::Failed { .. }
        ));
        finish_failure_abort(&transfer).await;

        let summary = transfer
            .inner
            .observability
            .test_terminal_summary()
            .expect("UploadPart failure should emit a terminal summary");
        assert_eq!(summary.outcome, UploadTerminalOutcome::Failed);
        assert_eq!(summary.requests.create_multipart_upload.requests, 1);
        assert_eq!(summary.requests.upload_part.requests, 1);
        let multipart = summary.multipart.expect("multipart state");
        assert_eq!(multipart.parts_completed, 0);
        assert_eq!(multipart.last_active_snapshot.uploads_in_flight, 0);
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn complete_multipart_failure_emits_terminal_summary() {
        let create_mpu = mock!(aws_sdk_s3::Client::create_multipart_upload).then_output(|| {
            CreateMultipartUploadOutput::builder()
                .upload_id("test-upload-id")
                .build()
        });
        let upload_part = mock!(aws_sdk_s3::Client::upload_part)
            .then_output(|| UploadPartOutput::builder().e_tag("test-etag").build());
        let complete_mpu = mock!(aws_sdk_s3::Client::complete_multipart_upload)
            .then_http_response(service_error_response);
        let transfer = create_test_transfer(
            mock_client!(
                aws_sdk_s3,
                RuleMode::MatchAny,
                &[create_mpu, upload_part, complete_mpu, abort_mpu_ok()]
            ),
            vec![0u8; 16 * 1024 * 1024],
        );

        let mut create = assert_ready(transfer.poll_work());
        transfer.execute(&mut create).await;
        for _ in 0..2 {
            let mut part = assert_ready(transfer.poll_work());
            transfer.execute(&mut part).await;
        }
        let mut complete = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut complete).await,
            WorkOutcome::Failed { .. }
        ));
        finish_failure_abort(&transfer).await;

        let summary = transfer
            .inner
            .observability
            .test_terminal_summary()
            .expect("CompleteMultipartUpload failure should emit a terminal summary");
        assert_eq!(summary.outcome, UploadTerminalOutcome::Failed);
        assert_eq!(summary.requests.create_multipart_upload.requests, 1);
        assert_eq!(summary.requests.upload_part.requests, 2);
        assert_eq!(summary.requests.complete_multipart_upload.requests, 1);
        assert_eq!(
            summary.multipart.expect("multipart state").parts_completed,
            2
        );
    }

    /// A failed Create leaves no upload to abort: the transfer ends as failed
    /// without AbortMultipartUpload.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn create_failure_ends_without_abort_mpu() {
        let create_mpu = mock!(aws_sdk_s3::Client::create_multipart_upload)
            .then_http_response(service_error_response);
        let transfer = create_test_transfer(
            mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[create_mpu]),
            vec![0u8; 16 * 1024 * 1024],
        );

        let mut create = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut create).await,
            WorkOutcome::Failed { .. }
        ));
        assert_eq!(
            aborting_view(&transfer),
            AbortingView {
                upload_id: None,
                create_in_flight: false,
                complete_mpu_in_flight: false,
                abort_mpu: AbortMpuProgress::ToSend,
                failure_recorded: true,
            }
        );
        assert_done(transfer.poll_work());
        assert_eq!(final_error(&transfer).code(), Some("PreconditionFailed"));
    }

    /// A caller's abort while Create is on the wire waits for Create; when
    /// Create then fails, the transfer ends as failed, not cancelled.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn create_failure_during_caller_abort_ends_as_failed() {
        let create_mpu = mock!(aws_sdk_s3::Client::create_multipart_upload)
            .then_http_response(service_error_response);
        let transfer = create_test_transfer(
            mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[create_mpu]),
            vec![0u8; 16 * 1024 * 1024],
        );

        let mut create = assert_ready(transfer.poll_work());
        transfer.request_abort();
        assert_pending(transfer.poll_work());
        assert_eq!(
            aborting_view(&transfer),
            AbortingView {
                upload_id: None,
                create_in_flight: true,
                complete_mpu_in_flight: false,
                abort_mpu: AbortMpuProgress::ToSend,
                failure_recorded: false,
            }
        );

        assert!(matches!(
            transfer.execute(&mut create).await,
            WorkOutcome::Failed { .. }
        ));
        assert_eq!(
            aborting_view(&transfer),
            AbortingView {
                upload_id: None,
                create_in_flight: false,
                complete_mpu_in_flight: false,
                abort_mpu: AbortMpuProgress::ToSend,
                failure_recorded: true,
            }
        );
        assert_done(transfer.poll_work());
        assert_eq!(final_error(&transfer).code(), Some("PreconditionFailed"));
    }

    /// A multipart file input whose metadata cannot be read fails before
    /// CreateMultipartUpload is sent, so there is no upload to abort.
    /// ref: RUST-1674
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn file_open_failure_fails_before_create() {
        use std::io::Write;
        let mut tmp = tempfile::NamedTempFile::new().unwrap();
        tmp.write_all(&vec![0u8; 16 * 1024 * 1024]).unwrap();
        let create_mpu = create_mpu_ok();
        let abort_mpu = abort_mpu_for_test_upload();
        let transfer = create_test_transfer_with_stream(
            mock_client!(
                aws_sdk_s3,
                RuleMode::MatchAny,
                &[create_mpu.clone(), abort_mpu.clone()]
            ),
            InputStream::from_path(tmp.path()).unwrap(),
        );
        // The file's metadata is read before Create is sent; removing the file
        // makes that read fail.
        tmp.close().unwrap();

        let mut create = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut create).await,
            WorkOutcome::Failed { .. }
        ));
        assert_done(transfer.poll_work());
        assert_eq!(create_mpu.num_calls(), 0);
        assert_eq!(abort_mpu.num_calls(), 0);
        assert_eq!(
            final_error(&transfer).kind(),
            &crate::error::ErrorKind::IOError
        );
    }

    /// Helper function to replace the regular file at `path` with a FIFO, so the path
    /// names a non-regular file only after the input stream read its metadata.
    /// Android's app sandbox denies creating the FIFO.
    #[cfg(all(unix, not(target_os = "android")))]
    fn replace_with_fifo(path: &std::path::Path) {
        std::fs::remove_file(path).unwrap();
        nix::unistd::mkfifo(path, nix::sys::stat::Mode::S_IRWXU).unwrap();
    }

    /// Helper function to assert on not_regular_file_error
    #[cfg(all(unix, not(target_os = "android")))]
    fn assert_not_regular_file_error(error: &crate::error::Error) {
        use aws_smithy_types::error::display::DisplayErrorContext;

        assert_eq!(error.kind(), &crate::error::ErrorKind::InputInvalid);
        let message = DisplayErrorContext(error).to_string();
        assert!(message.contains("is not a regular file"), "{message}");
    }

    /// A file input that no longer opens as a regular file fails PutObject
    /// before the request is sent.
    /// ref: RUST-1674
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    #[cfg(all(unix, not(target_os = "android")))]
    async fn put_object_rejects_a_source_replaced_by_a_fifo() {
        // Create a regular file, that it would pass InputStream check
        let directory = tempfile::tempdir().unwrap();
        let path = directory.path().join("body");
        std::fs::write(&path, vec![0u8; 1024]).unwrap();
        let stream = InputStream::from_path(&path).unwrap();

        // Replace it with FIFO
        replace_with_fifo(&path);

        // Then make sure we validate the file before sending the request.
        let put_object = mock!(aws_sdk_s3::Client::put_object)
            .then_output(|| aws_sdk_s3::operation::put_object::PutObjectOutput::builder().build());
        let transfer = create_test_transfer_with_stream(
            mock_client!(
                aws_sdk_s3,
                RuleMode::MatchAny,
                std::slice::from_ref(&put_object)
            ),
            stream,
        );

        let mut work = assert_ready(transfer.poll_work());
        assert!(matches!(
            work.data_mut::<UploadWork>(),
            UploadWork::PutObject { .. }
        ));
        assert!(matches!(
            transfer.execute(&mut work).await,
            WorkOutcome::Failed { .. }
        ));
        assert_done(transfer.poll_work());
        assert_eq!(put_object.num_calls(), 0);
        assert_not_regular_file_error(&final_error(&transfer));
    }

    /// A multipart file input that no longer opens as a regular file fails
    /// before CreateMultipartUpload is sent, so no multipart upload exists to
    /// abort or to be left open, whatever the failed-upload policy.
    /// ref: RUST-1674
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    #[cfg(all(unix, not(target_os = "android")))]
    async fn multipart_rejects_a_source_replaced_by_a_fifo() {
        for policy in [None, Some(FailedMultipartUploadPolicy::Retain)] {
            // Create a regular file, that it would pass InputStream check
            let directory = tempfile::tempdir().unwrap();
            let path = directory.path().join("body");
            std::fs::write(&path, vec![0u8; 16 * 1024 * 1024]).unwrap();
            let stream = InputStream::from_path(&path).unwrap();

            // Replace the regular file with fifo
            replace_with_fifo(&path);

            let create_mpu = create_mpu_ok();
            let upload_part = mock!(aws_sdk_s3::Client::upload_part)
                .then_output(|| UploadPartOutput::builder().e_tag("test-etag").build());
            let abort_mpu = abort_mpu_for_test_upload();
            let transfer = create_test_transfer_with_policy(
                mock_client!(
                    aws_sdk_s3,
                    RuleMode::MatchAny,
                    &[create_mpu.clone(), upload_part.clone(), abort_mpu.clone()]
                ),
                stream,
                2,
                policy.clone(),
            );

            let mut create = assert_ready(transfer.poll_work());
            assert!(matches!(
                transfer.execute(&mut create).await,
                WorkOutcome::Failed { .. }
            ));
            assert_done(transfer.poll_work());
            assert_eq!(create_mpu.num_calls(), 0, "{policy:?}");
            assert_eq!(upload_part.num_calls(), 0, "{policy:?}");
            assert_eq!(abort_mpu.num_calls(), 0, "{policy:?}");
            let error = final_error(&transfer);
            assert_eq!(error.upload_id(), None, "{policy:?}");
            assert_not_regular_file_error(&error);
        }
    }

    /// A failed UploadPart aborts the upload and reports the part's error.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn upload_part_failure_aborts_the_upload() {
        let upload_part =
            mock!(aws_sdk_s3::Client::upload_part).then_http_response(service_error_response);
        let abort_mpu = abort_mpu_for_test_upload();
        let transfer = create_test_transfer(
            mock_client!(
                aws_sdk_s3,
                RuleMode::MatchAny,
                &[create_mpu_ok(), upload_part, abort_mpu.clone()]
            ),
            vec![0u8; 16 * 1024 * 1024],
        );

        let mut create = assert_ready(transfer.poll_work());
        transfer.execute(&mut create).await;
        let mut part = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut part).await,
            WorkOutcome::Failed { .. }
        ));
        assert_eq!(
            aborting_view(&transfer),
            AbortingView {
                upload_id: Some("test-upload-id".to_owned()),
                create_in_flight: false,
                complete_mpu_in_flight: false,
                abort_mpu: AbortMpuProgress::ToSend,
                failure_recorded: true,
            }
        );
        finish_failure_abort(&transfer).await;
        assert_eq!(abort_mpu.num_calls(), 1);
        assert_eq!(final_error(&transfer).code(), Some("PreconditionFailed"));
    }

    /// Under `Retain`, a failed UploadPart leaves the upload open.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn upload_part_failure_with_retain_sends_no_abort_mpu() {
        let upload_part =
            mock!(aws_sdk_s3::Client::upload_part).then_http_response(service_error_response);
        let abort_mpu = abort_mpu_for_test_upload();
        let transfer = create_test_transfer_with_policy(
            mock_client!(
                aws_sdk_s3,
                RuleMode::MatchAny,
                &[create_mpu_ok(), upload_part, abort_mpu.clone()]
            ),
            InputStream::from(vec![0u8; 16 * 1024 * 1024]),
            2,
            Some(FailedMultipartUploadPolicy::Retain),
        );

        let mut create = assert_ready(transfer.poll_work());
        transfer.execute(&mut create).await;
        let mut part = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut part).await,
            WorkOutcome::Failed { .. }
        ));
        assert_eq!(
            aborting_view(&transfer),
            AbortingView {
                upload_id: Some("test-upload-id".to_owned()),
                create_in_flight: false,
                complete_mpu_in_flight: false,
                abort_mpu: AbortMpuProgress::NotNeeded,
                failure_recorded: true,
            }
        );
        assert_done(transfer.poll_work());
        assert_eq!(abort_mpu.num_calls(), 0);
        assert_eq!(final_error(&transfer).code(), Some("PreconditionFailed"));
    }

    /// Under `Retain`, an abort before CreateMultipartUpload is sent leaves no
    /// upload to retain, so it stores no result: `abort()` reports
    /// `Cancelled` without an upload ID.
    #[cfg_attr(miri, ignore)]
    #[test]
    fn abort_before_create_with_retain_stores_no_result() {
        let transfer = create_test_transfer_with_policy(
            mock_client!(aws_sdk_s3, []),
            InputStream::from(vec![0u8; 16 * 1024 * 1024]),
            2,
            Some(FailedMultipartUploadPolicy::Retain),
        );

        transfer.request_abort();
        assert_done(transfer.poll_work());
        assert!(transfer.take_result().is_none());
    }

    /// A failed Complete aborts the upload; the abort does not wait for
    /// Complete, whose response is already in.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn complete_failure_aborts_the_upload() {
        let upload_part = mock!(aws_sdk_s3::Client::upload_part)
            .then_output(|| UploadPartOutput::builder().e_tag("test-etag").build());
        let complete_mpu = mock!(aws_sdk_s3::Client::complete_multipart_upload)
            .then_http_response(service_error_response);
        let abort_mpu = abort_mpu_for_test_upload();
        let transfer = create_test_transfer(
            mock_client!(
                aws_sdk_s3,
                RuleMode::MatchAny,
                &[
                    create_mpu_ok(),
                    upload_part,
                    complete_mpu,
                    abort_mpu.clone()
                ]
            ),
            vec![0u8; 16 * 1024 * 1024],
        );

        let mut create = assert_ready(transfer.poll_work());
        transfer.execute(&mut create).await;
        for _ in 0..2 {
            let mut part = assert_ready(transfer.poll_work());
            transfer.execute(&mut part).await;
        }
        let mut complete = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut complete).await,
            WorkOutcome::Failed { .. }
        ));
        assert_eq!(
            aborting_view(&transfer),
            AbortingView {
                upload_id: Some("test-upload-id".to_owned()),
                create_in_flight: false,
                complete_mpu_in_flight: false,
                abort_mpu: AbortMpuProgress::ToSend,
                failure_recorded: true,
            }
        );
        finish_failure_abort(&transfer).await;
        assert_eq!(abort_mpu.num_calls(), 1);
        assert_eq!(final_error(&transfer).code(), Some("PreconditionFailed"));
    }

    /// When the AbortMPU a failure started also fails, it is not sent again
    /// and the transfer reports the first failure.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn abort_mpu_failure_after_part_failure_keeps_part_error() {
        let upload_part =
            mock!(aws_sdk_s3::Client::upload_part).then_http_response(service_error_response);
        let abort_mpu = mock!(aws_sdk_s3::Client::abort_multipart_upload)
            .then_http_response(access_denied_response);
        let transfer = create_test_transfer(
            mock_client!(
                aws_sdk_s3,
                RuleMode::MatchAny,
                &[create_mpu_ok(), upload_part, abort_mpu.clone()]
            ),
            vec![0u8; 16 * 1024 * 1024],
        );

        let mut create = assert_ready(transfer.poll_work());
        transfer.execute(&mut create).await;
        let mut part = assert_ready(transfer.poll_work());
        transfer.execute(&mut part).await;
        let mut abort = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut abort).await,
            WorkOutcome::Failed { .. }
        ));
        assert_eq!(
            aborting_view(&transfer),
            AbortingView {
                upload_id: Some("test-upload-id".to_owned()),
                create_in_flight: false,
                complete_mpu_in_flight: false,
                abort_mpu: AbortMpuProgress::Failed,
                failure_recorded: true,
            }
        );
        assert_done(transfer.poll_work());
        assert_eq!(abort_mpu.num_calls(), 1);
        assert_eq!(final_error(&transfer).code(), Some("PreconditionFailed"));
    }

    /// When the AbortMPU of a caller's abort fails, the transfer ends as
    /// failed with the AbortMPU error instead of cancelled.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn abort_mpu_failure_during_caller_abort_ends_as_failed() {
        let abort_mpu = mock!(aws_sdk_s3::Client::abort_multipart_upload)
            .then_http_response(access_denied_response);
        let transfer = create_test_transfer(
            mock_client!(
                aws_sdk_s3,
                RuleMode::MatchAny,
                &[create_mpu_ok(), abort_mpu.clone()]
            ),
            vec![0u8; 16 * 1024 * 1024],
        );

        let mut create = assert_ready(transfer.poll_work());
        transfer.execute(&mut create).await;
        transfer.request_abort();
        let mut abort = assert_ready(transfer.poll_work());
        assert_eq!(
            aborting_view(&transfer),
            AbortingView {
                upload_id: Some("test-upload-id".to_owned()),
                create_in_flight: false,
                complete_mpu_in_flight: false,
                abort_mpu: AbortMpuProgress::InFlight,
                failure_recorded: false,
            }
        );
        assert!(matches!(
            transfer.execute(&mut abort).await,
            WorkOutcome::Failed { .. }
        ));
        assert_eq!(aborting_view(&transfer).abort_mpu, AbortMpuProgress::Failed);
        assert_done(transfer.poll_work());
        assert_eq!(abort_mpu.num_calls(), 1);
        assert_eq!(final_error(&transfer).code(), Some("AccessDenied"));
    }

    /// Regression: when the upload source is a file (`InputStream::from_path`),
    /// the PutObject code path must record `disk_read` equal to the payload
    /// size. Previously only `network_tx` was recorded, leaving the plural
    /// `upload_objects` metrics' `disk_read` at zero for all small-file
    /// children.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_put_object_records_disk_read_for_file_source() {
        use aws_sdk_s3::operation::put_object::PutObjectOutput;
        use std::io::Write;

        let put_object = mock!(aws_sdk_s3::Client::put_object)
            .then_output(|| PutObjectOutput::builder().e_tag("test-etag").build());
        let s3_client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[put_object]);

        let mut tmp = tempfile::NamedTempFile::new().unwrap();
        let payload = vec![0u8; 1024];
        tmp.write_all(&payload).unwrap();
        tmp.flush().unwrap();

        let handle = crate::client::Handle::test_handle_tokio(
            crate::Config::builder()
                .sdk_client(s3_client)
                .build()
                .expect("valid config"),
        );
        let mut input = UploadInput::builder()
            .bucket("test-bucket")
            .key("test-key")
            .body(InputStream::from_path(tmp.path()).unwrap())
            .build()
            .unwrap();
        let stream = input.take_body();
        let (ctx, _completion_rx) = TransferContext::new(handle);
        let transfer = UploadTransfer::try_new(ctx, BucketType::Standard, input, stream).unwrap();

        let mut work = assert_ready(transfer.poll_work());
        assert!(matches!(
            work.data_mut::<UploadWork>(),
            UploadWork::PutObject { .. }
        ));
        let outcome = transfer.execute(&mut work).await;
        assert!(matches!(outcome, WorkOutcome::Success { .. }));

        let metrics = transfer.ctx().metrics();
        assert_eq!(
            payload.len() as u64,
            metrics.network_tx,
            "network_tx should equal the file size"
        );
        assert_eq!(
            payload.len() as u64,
            metrics.disk_read,
            "disk_read should equal the file size for a file-backed PutObject"
        );
    }

    /// Complement to the file-backed test: in-memory (`RawInputStream::Buf`)
    /// uploads must NOT inflate `disk_read`, since no bytes were read from
    /// disk.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_put_object_does_not_record_disk_read_for_memory_source() {
        use aws_sdk_s3::operation::put_object::PutObjectOutput;

        let put_object = mock!(aws_sdk_s3::Client::put_object)
            .then_output(|| PutObjectOutput::builder().e_tag("test-etag").build());
        let s3_client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[put_object]);
        let content = vec![0u8; 1024];
        let transfer = create_test_transfer(s3_client, content.clone());

        let mut work = assert_ready(transfer.poll_work());
        let outcome = transfer.execute(&mut work).await;
        assert!(matches!(outcome, WorkOutcome::Success { .. }));

        let metrics = transfer.ctx().metrics();
        assert_eq!(content.len() as u64, metrics.network_tx);
        assert_eq!(
            0, metrics.disk_read,
            "in-memory source must not report disk_read"
        );
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_basic_mpu() {
        let s3_client = mock_s3_client_for_mpu();
        let content = vec![0u8; 16 * 1024 * 1024];
        let transfer = create_test_transfer(s3_client, content);

        // CreateMPU
        let mut work = assert_ready(transfer.poll_work());
        assert!(matches!(
            work.data_mut::<UploadWork>(),
            UploadWork::CreateMPU(_)
        ));
        transfer.execute(&mut work).await;

        // Upload parts
        let mut work1 = assert_ready(transfer.poll_work());
        transfer.execute(&mut work1).await;
        let mut work2 = assert_ready(transfer.poll_work());
        transfer.execute(&mut work2).await;

        // CompleteMPU
        let mut work = assert_ready(transfer.poll_work());
        assert!(matches!(
            work.data_mut::<UploadWork>(),
            UploadWork::CompleteMPU(..)
        ));
        transfer.execute(&mut work).await;

        assert!(transfer.take_result().is_some());
    }

    /// Returns a body over a [`BlockingPartStream`], and a `Weak` that upgrades only while the
    /// stream has not been dropped.
    ///
    /// With `ready`, the stream yields one five-byte part and then ends. Otherwise it stays
    /// pending, as a producer waiting for the upload to take its next part does.
    fn probed_stream(ready: bool) -> (InputStream, std::sync::Weak<()>) {
        let probe = Arc::new(());
        let weak = Arc::downgrade(&probe);
        let stream = BlockingPartStream {
            ready: Arc::new(std::sync::atomic::AtomicBool::new(ready)),
            source_waker: Arc::new(Mutex::new(None)),
            yielded: false,
            _probe: Some(probe),
        };
        (InputStream::from_part_stream(stream), weak)
    }

    /// Creates the multipart upload, then executes one source read that parks on the pending
    /// source.
    async fn park_one_source_read(transfer: &UploadTransfer) {
        let mut create = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut create).await,
            WorkOutcome::Success { .. }
        ));
        let mut read = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut read).await,
            WorkOutcome::Yielded
        ));
        let state = transfer.inner.state.lock().expect("lock poisoned");
        assert_eq!(snapshot_state(&state).pending_reads, 1);
    }

    // ref: awslabs/aws-s3-transfer-manager-rs#226
    /// A source read parked on a pending source is released when another work item fails the
    /// upload, rather than retained for as long as the transfer lives.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn failure_releases_parked_source_read() {
        let (stream, probe) = probed_stream(false);
        let transfer = create_test_transfer_with_stream(mock_s3_client_for_mpu(), stream);
        park_one_source_read(&transfer).await;

        assert!(matches!(
            transfer.fail(
                transfer.inner.state.lock().expect("lock poisoned"),
                Error::new(ErrorKind::RuntimeError, "sibling failure"),
                FailedRequest::Other,
            ),
            WorkOutcome::Failed { .. }
        ));

        assert!(
            probe.upgrade().is_none(),
            "failed upload retained its parked source read"
        );
    }

    // ref: awslabs/aws-s3-transfer-manager-rs#226
    /// A failed UploadPart releases a source that no read holds. Between reads the stream waits
    /// in the part reader, so the failure must release the reader, not only the parked reads.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn upload_part_failure_releases_idle_source() {
        let create_mpu = mock!(aws_sdk_s3::Client::create_multipart_upload).then_output(|| {
            CreateMultipartUploadOutput::builder()
                .upload_id("test-upload-id")
                .build()
        });
        let upload_part =
            mock!(aws_sdk_s3::Client::upload_part).then_http_response(service_error_response);
        let (stream, probe) = probed_stream(true);
        let transfer = create_test_transfer_with_stream(
            mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[create_mpu, upload_part]),
            stream,
        );

        let mut create = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut create).await,
            WorkOutcome::Success { .. }
        ));
        let mut part = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut part).await,
            WorkOutcome::Failed { .. }
        ));

        assert!(
            probe.upgrade().is_none(),
            "failed upload retained its idle source"
        );
    }

    /// A source whose first poll fails the upload and then returns `Pending`: another work item
    /// failing while a read is in progress.
    #[derive(Debug)]
    struct FailsUploadWhilePolledStream {
        /// Taken on the first poll, which also breaks the transfer-to-stream cycle.
        transfer: Arc<Mutex<Option<UploadTransfer>>>,
        _probe: Arc<()>,
    }

    impl crate::io::PartStream for FailsUploadWhilePolledStream {
        fn poll_part(
            self: Pin<&mut Self>,
            _cx: &mut std::task::Context<'_>,
            _stream_cx: &crate::io::StreamContext,
        ) -> std::task::Poll<Option<std::io::Result<PartData>>> {
            let transfer = self.transfer.lock().expect("transfer slot poisoned").take();
            if let Some(transfer) = transfer {
                assert!(matches!(
                    transfer.fail(
                        transfer.inner.state.lock().expect("lock poisoned"),
                        Error::new(ErrorKind::RuntimeError, "sibling failure"),
                        FailedRequest::Other,
                    ),
                    WorkOutcome::Failed { .. }
                ));
            }
            std::task::Poll::Pending
        }

        fn size_hint(&self) -> crate::io::SizeHint {
            crate::io::SizeHint::default()
        }
    }

    // ref: awslabs/aws-s3-transfer-manager-rs#226
    /// A read whose source returns `Pending` after another work item failed the upload is not
    /// parked: the read reports `Cancelled`, and its stream is dropped with the part reader rather
    /// than retained.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn read_pending_after_failure_is_not_parked() {
        let probe = Arc::new(());
        let weak = Arc::downgrade(&probe);
        let slot = Arc::new(Mutex::new(None));
        let stream = FailsUploadWhilePolledStream {
            transfer: Arc::clone(&slot),
            _probe: probe,
        };
        let transfer = create_test_transfer_with_stream(
            mock_s3_client_for_mpu(),
            InputStream::from_part_stream(stream),
        );
        *slot.lock().expect("transfer slot poisoned") = Some(transfer.clone());

        let mut create = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut create).await,
            WorkOutcome::Success { .. }
        ));
        let mut read = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut read).await,
            WorkOutcome::Cancelled
        ));

        let state = transfer.inner.state.lock().expect("lock poisoned");
        assert_eq!(snapshot_state(&state).pending_reads, 0);
        drop(state);
        assert!(
            weak.upgrade().is_none(),
            "a read pending after the failure retained its source"
        );
    }

    // ref: awslabs/aws-s3-transfer-manager-rs#226
    /// Ending the transfer from outside (cancellation or a worker panic) releases a source read
    /// parked on a pending source, rather than retaining it for as long as the handle is held.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn external_termination_releases_parked_source_read() {
        let (stream, probe) = probed_stream(false);
        let transfer = create_test_transfer_with_stream(mock_s3_client_for_mpu(), stream);
        park_one_source_read(&transfer).await;

        transfer.ctx().set_cancelled();
        Transfer::on_terminal(&transfer);

        assert!(
            probe.upgrade().is_none(),
            "cancelled upload retained its parked source read"
        );
    }

    // ref: awslabs/aws-s3-transfer-manager-rs#226
    /// A read dispatched before the transfer ended from outside, and executed after, starts no
    /// source operation.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn read_started_after_termination_is_cancelled() {
        let (stream, probe) = probed_stream(false);
        let transfer = create_test_transfer_with_stream(mock_s3_client_for_mpu(), stream);
        let mut create = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut create).await,
            WorkOutcome::Success { .. }
        ));
        let mut read = assert_ready(transfer.poll_work());

        transfer.ctx().set_cancelled();
        Transfer::on_terminal(&transfer);
        assert!(probe.upgrade().is_none());

        assert!(matches!(
            transfer.execute(&mut read).await,
            WorkOutcome::Cancelled
        ));
    }

    // ref: awslabs/aws-s3-transfer-manager-rs#226
    /// Ending the transfer from outside while CreateMultipartUpload is in flight drops the body.
    /// When Create then succeeds, it stands down without installing a part reader.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn create_multipart_after_termination_is_cancelled() {
        let (stream, probe) = probed_stream(false);
        let transfer = create_test_transfer_with_stream(mock_s3_client_for_mpu(), stream);
        let mut create = assert_ready(transfer.poll_work());

        transfer.ctx().set_cancelled();
        Transfer::on_terminal(&transfer);
        assert!(probe.upgrade().is_none());

        assert!(matches!(
            transfer.execute(&mut create).await,
            WorkOutcome::Cancelled
        ));
        let state = transfer.inner.state.lock().expect("lock poisoned");
        assert!(matches!(&*state, UploadState::Done));
    }

    /// `open_upload_id` reports a multipart upload that was created, has not committed, and has
    /// not been closed by AbortMultipartUpload.
    #[test]
    fn open_upload_id_reports_only_uploads_left_open() {
        let id = || Some("test-upload-id".to_owned());
        let aborting = |upload_id, abort_mpu| UploadState::Aborting {
            upload_id,
            create_in_flight: false,
            complete_mpu_in_flight: false,
            abort_mpu,
            failure: None,
        };
        let complete_in_flight = UploadState::CompleteInFlight {
            upload_id: "test-upload-id".to_owned(),
        };
        assert_eq!(open_upload_id(&complete_in_flight), id());
        for abort_mpu in [
            AbortMpuProgress::NotNeeded,
            AbortMpuProgress::ToSend,
            AbortMpuProgress::InFlight,
            AbortMpuProgress::Failed,
        ] {
            assert_eq!(
                open_upload_id(&aborting(id(), abort_mpu)),
                id(),
                "{abort_mpu:?}"
            );
        }
        assert_eq!(
            open_upload_id(&aborting(id(), AbortMpuProgress::Done)),
            None
        );
        assert_eq!(
            open_upload_id(&aborting(None, AbortMpuProgress::ToSend)),
            None
        );
        let creating = UploadState::PendingInit {
            stream: None,
            size_hint: crate::io::SizeHint::default(),
            init_in_flight: true,
        };
        assert_eq!(open_upload_id(&creating), None);
        assert_eq!(open_upload_id(&UploadState::PutObjectInFlight), None);
        assert_eq!(open_upload_id(&UploadState::Done), None);
    }

    /// `open_upload_id` reports the upload while parts are transferring, and once they have all
    /// completed and CompleteMultipartUpload has not been dispatched.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn open_upload_id_reports_an_upload_with_parts() {
        let (stream, _probe) = probed_stream(false);
        let transferring = create_test_transfer_with_stream(mock_s3_client_for_mpu(), stream);
        park_one_source_read(&transferring).await;
        {
            let state = transferring.inner.state.lock().expect("lock poisoned");
            assert!(matches!(&*state, UploadState::Transferring { .. }));
            assert_eq!(open_upload_id(&state).as_deref(), Some("test-upload-id"));
        }

        let upload_part = mock!(aws_sdk_s3::Client::upload_part)
            .then_output(|| UploadPartOutput::builder().e_tag("test-etag").build());
        let completing = create_test_transfer(
            mock_client!(
                aws_sdk_s3,
                RuleMode::MatchAny,
                &[create_mpu_ok(), upload_part]
            ),
            vec![0u8; 16 * 1024 * 1024],
        );
        let mut create = assert_ready(completing.poll_work());
        assert!(matches!(
            completing.execute(&mut create).await,
            WorkOutcome::Success { .. }
        ));
        for _ in 0..2 {
            let mut part = assert_ready(completing.poll_work());
            assert!(matches!(
                completing.execute(&mut part).await,
                WorkOutcome::Success { .. }
            ));
        }
        let state = completing.inner.state.lock().expect("lock poisoned");
        assert!(matches!(&*state, UploadState::Completing { .. }));
        assert_eq!(open_upload_id(&state).as_deref(), Some("test-upload-id"));
    }

    // ref: awslabs/aws-s3-transfer-manager-rs#226
    /// A CompleteMultipartUpload dispatched before the transfer ended from outside, and executed
    /// after, is never sent.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn complete_multipart_after_termination_is_not_sent() {
        let upload_part = mock!(aws_sdk_s3::Client::upload_part)
            .then_output(|| UploadPartOutput::builder().e_tag("test-etag").build());
        let complete_mpu = mock!(aws_sdk_s3::Client::complete_multipart_upload)
            .then_output(|| CompleteMultipartUploadOutput::builder().build());
        let transfer = create_test_transfer(
            mock_client!(
                aws_sdk_s3,
                RuleMode::MatchAny,
                &[create_mpu_ok(), upload_part, complete_mpu.clone()]
            ),
            vec![0u8; 16 * 1024 * 1024],
        );
        let mut create = assert_ready(transfer.poll_work());
        assert!(matches!(
            transfer.execute(&mut create).await,
            WorkOutcome::Success { .. }
        ));
        for _ in 0..2 {
            let mut part = assert_ready(transfer.poll_work());
            assert!(matches!(
                transfer.execute(&mut part).await,
                WorkOutcome::Success { .. }
            ));
        }
        let mut complete = assert_ready(transfer.poll_work());

        transfer.ctx().set_cancelled();
        Transfer::on_terminal(&transfer);

        assert!(matches!(
            transfer.execute(&mut complete).await,
            WorkOutcome::Cancelled
        ));
        assert_eq!(complete_mpu.num_calls(), 0);
        assert!(transfer.take_result().is_none());
    }

    // ref: awslabs/aws-s3-transfer-manager-rs#226
    /// A PutObject that starts after the transfer ended from outside, before `abort_parts_token`
    /// interrupts it, is never sent.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn put_object_after_termination_is_not_sent() {
        use aws_sdk_s3::operation::put_object::PutObjectOutput;
        let put_object = mock!(aws_sdk_s3::Client::put_object)
            .then_output(|| PutObjectOutput::builder().e_tag("test-etag").build());
        let transfer = create_test_transfer(
            mock_client!(
                aws_sdk_s3,
                RuleMode::MatchAny,
                std::slice::from_ref(&put_object)
            ),
            vec![0u8; 1024],
        );
        let mut work = assert_ready(transfer.poll_work());
        let UploadWork::PutObject {
            stream,
            live: _live,
        } = work.take_data::<UploadWork>()
        else {
            panic!("expected PutObject");
        };

        transfer.ctx().set_cancelled();
        Transfer::on_terminal(&transfer);

        // Called past the token race in `execute`, as work that starts before `on_terminal`
        // cancels the token is.
        assert!(matches!(
            transfer.execute_put_object(stream).await,
            WorkOutcome::Cancelled
        ));
        assert_eq!(put_object.num_calls(), 0);
    }

    /// A source that stays pending and, when dropped, records whether the upload-state lock was
    /// free.
    #[derive(Debug)]
    struct ChecksStateLockOnDrop {
        /// Taken when dropped, which also breaks the transfer-to-stream cycle.
        transfer: Arc<Mutex<Option<UploadTransfer>>>,
        lock_was_free: Arc<Mutex<Option<bool>>>,
    }

    impl crate::io::PartStream for ChecksStateLockOnDrop {
        fn poll_part(
            self: Pin<&mut Self>,
            _cx: &mut std::task::Context<'_>,
            _stream_cx: &crate::io::StreamContext,
        ) -> std::task::Poll<Option<std::io::Result<PartData>>> {
            std::task::Poll::Pending
        }

        fn size_hint(&self) -> crate::io::SizeHint {
            crate::io::SizeHint::default()
        }
    }

    impl Drop for ChecksStateLockOnDrop {
        fn drop(&mut self) {
            let transfer = self.transfer.lock().expect("transfer slot poisoned").take();
            if let Some(transfer) = transfer {
                let free = transfer.inner.state.try_lock().is_ok();
                *self.lock_was_free.lock().expect("result poisoned") = Some(free);
            }
        }
    }

    /// Parks the only read of an upload, ends the upload with `end`, and asserts that the
    /// caller's stream was dropped with the upload-state lock free.
    ///
    /// No other work holds the part reader, so the state that `end` replaces holds the last owner
    /// of the stream.
    async fn assert_stream_dropped_outside_state_lock(end: impl FnOnce(&UploadTransfer)) {
        let slot = Arc::new(Mutex::new(None));
        let lock_was_free = Arc::new(Mutex::new(None));
        let stream = ChecksStateLockOnDrop {
            transfer: Arc::clone(&slot),
            lock_was_free: Arc::clone(&lock_was_free),
        };
        let transfer = create_test_transfer_with_stream(
            mock_s3_client_for_mpu(),
            InputStream::from_part_stream(stream),
        );
        *slot.lock().expect("transfer slot poisoned") = Some(transfer.clone());
        park_one_source_read(&transfer).await;

        end(&transfer);

        assert_eq!(
            *lock_was_free.lock().expect("result poisoned"),
            Some(true),
            "the stream was not dropped, or was dropped under the upload-state lock"
        );
    }

    // ref: awslabs/aws-s3-transfer-manager-rs#226
    /// Aborting an upload drops the caller's stream after the upload-state lock is released. The
    /// stream's `Drop` is caller code, and must not run under the lock.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn abort_drops_the_stream_outside_the_state_lock() {
        assert_stream_dropped_outside_state_lock(UploadTransfer::request_abort).await;
    }

    // ref: awslabs/aws-s3-transfer-manager-rs#226
    /// Ending the transfer from outside drops the caller's stream after the upload-state lock is
    /// released.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn termination_drops_the_stream_outside_the_state_lock() {
        assert_stream_dropped_outside_state_lock(|transfer| {
            transfer.ctx().set_cancelled();
            Transfer::on_terminal(transfer);
        })
        .await;
    }
}