prodigy 0.4.4

Turn ad-hoc Claude sessions into reproducible development pipelines with parallel AI agents
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
## Troubleshooting

Common retry configuration issues and their solutions.

### Issue: Retries Not Triggering

**Symptoms**:
- Command fails immediately without retry
- Only see one attempt in logs
- Expected retries don't happen

**Possible Causes and Solutions**:

#### 1. retry_on Matchers Too Specific

```yaml
# Problem: Error doesn't match any configured matcher
retry_config:
  attempts: 5
  retry_on:
    - network
    # Error is "connection timeout" but matcher only catches "network"
```

**Solution**: Use broader matchers or add custom patterns

```yaml
retry_config:
  attempts: 5
  retry_on:
    - network
    - timeout      # Add timeout matcher
    - pattern: "connection.*timeout"  # Or custom pattern
```

**Debug**: Check actual error message in logs and verify it matches one of your configured matchers (remember: matching is case-insensitive).

**Source**: ErrorMatcher matching logic in src/cook/retry_v2.rs:128-149

#### 2. Empty retry_on with Unexpected Error Filtering

```yaml
# Misconception: Empty retry_on means "don't retry"
# Reality: Empty retry_on means "retry ALL errors"
retry_config:
  attempts: 3
  # retry_on is empty - retries everything!
```

**Solution**: If you see retries happening when you don't expect them, check if `retry_on` is empty.

**Source**: Default behavior in src/cook/retry_v2.rs:42-43

#### 3. retry_config Missing Entirely

```yaml
commands:
  - shell: "curl https://api.example.com"
    # No retry_config - no retry happens
```

**Solution**: Add `retry_config` block to enable retry.

---

### Issue: Retries Happening But Taking Too Long

**Symptoms**:
- Workflow hangs during retries
- Total retry time exceeds expectations
- Delays grow too large

**Possible Causes and Solutions**:

#### 1. Exponential Backoff Without max_delay

```yaml
# Problem: Exponential growth uncapped
retry_config:
  attempts: 10
  backoff: exponential
  initial_delay: "1s"
  # No max_delay - delay can grow to minutes!
```

**Solution**: Always set `max_delay`

```yaml
retry_config:
  attempts: 10
  backoff: exponential
  initial_delay: "1s"
  max_delay: "30s"  # Cap delays at 30 seconds
```

**Default**: max_delay defaults to 30 seconds if not specified (src/cook/retry_v2.rs:64)

#### 2. Too Many Attempts Without retry_budget

```yaml
# Problem: High attempts can retry for hours
retry_config:
  attempts: 100
  backoff: exponential
```

**Solution**: Use `retry_budget` to cap total time

```yaml
retry_config:
  attempts: 100
  backoff: exponential
  retry_budget: "10m"  # Never exceed 10 minutes total
```

**Source**: retry_budget enforcement in tests at src/cook/retry_v2.rs:675-708

#### 3. Initial Delay Too High

```yaml
# Problem: First retry waits too long
retry_config:
  attempts: 3
  initial_delay: "60s"  # 1-minute delay before first retry!
```

**Solution**: Use shorter initial delay, let backoff grow

```yaml
retry_config:
  attempts: 3
  initial_delay: "1s"   # Start small
  backoff: exponential  # Let it grow
  max_delay: "30s"
```

---

### Issue: Circuit Breaker Always Open

**Symptoms**:
- Circuit breaker trips and stays open
- Requests fail immediately after threshold
- Recovery doesn't happen

**Possible Causes and Solutions**:

#### 1. Failure Threshold Too Low

```rust
// Problem: Circuit opens too easily
let executor = RetryExecutor::new(config)
    .with_circuit_breaker(
        1,                          // Opens after single failure!
        Duration::from_secs(30)
    );
```

**Solution**: Use appropriate threshold for failure rate

```rust
let executor = RetryExecutor::new(config)
    .with_circuit_breaker(
        5,                          // Open after 5 consecutive failures
        Duration::from_secs(30)
    );
```

**Source**: CircuitBreaker implementation in src/cook/retry_v2.rs:325-397

#### 2. Recovery Timeout Too Long

```rust
// Problem: Circuit stays open too long
let executor = RetryExecutor::new(config)
    .with_circuit_breaker(
        5,
        Duration::from_secs(600)    // 10-minute recovery time!
    );
```

**Solution**: Use shorter recovery timeout for faster testing

```rust
let executor = RetryExecutor::new(config)
    .with_circuit_breaker(
        5,
        Duration::from_secs(30)     // 30-second recovery attempts
    );
```

#### 3. No Success to Close Circuit

**Problem**: Circuit opens but downstream never recovers, so circuit never closes

**Solution**:
- Check if downstream service is actually recovering
- Verify circuit enters HalfOpen state and test requests succeed
- Monitor circuit state transitions with logging

---

### Issue: Thundering Herd (Multiple Parallel Agents Retrying Simultaneously)

**Symptoms**:
- Service overwhelmed during recovery
- All parallel agents retry at same time
- Cascading failures

**Problem**:

```yaml
# MapReduce with 10 parallel agents, no jitter
map:
  max_parallel: 10
  agent_template:
    - shell: "api-call.sh"
      retry_config:
        attempts: 5
        backoff: exponential
        jitter: false  # All agents retry at same time!
```

**Solution**: Enable jitter

```yaml
map:
  max_parallel: 10
  agent_template:
    - shell: "api-call.sh"
      retry_config:
        attempts: 5
        backoff: exponential
        jitter: true          # Randomize retry timing
        jitter_factor: 0.3    # 30% randomization
```

**Why**: Without jitter, all 10 agents calculate the same exponential delay and retry simultaneously.

**Source**: Jitter application in src/cook/retry_v2.rs:308-317

---

### Issue: Retrying Non-Transient Errors

**Symptoms**:
- Retrying 404 Not Found errors
- Retrying authentication failures (401)
- Wasting time on permanent failures

**Problem**:

```yaml
# Empty retry_on retries everything
retry_config:
  attempts: 5
  # Retries 404, 401, 400, etc. - permanent errors!
```

**Solution**: Use selective retry with error matchers

```yaml
retry_config:
  attempts: 5
  retry_on:
    - network        # Transient
    - timeout        # Transient
    - server_error   # Transient (5xx)
    # Don't retry 404, 401, 400, etc.
```

**Best Practice**: Only retry errors that might succeed on next attempt.

---

### Issue: Retry Budget Not Working as Expected

**Symptoms**:
- Retries exceed configured budget
- Budget seems ignored

**Possible Cause**: Misunderstanding retry_budget behavior

**How retry_budget Works**:
- Budget is checked **before each retry**
- If budget would be exceeded, retry stops
- Budget includes backoff delay time
- Budget does NOT include time for command execution itself

**Example**:

```yaml
retry_config:
  attempts: 10
  backoff: exponential
  initial_delay: "1s"
  max_delay: "60s"
  retry_budget: "2m"  # 2-minute budget
```

**Behavior**:
- If accumulated delays + next delay > 2 minutes → Stop
- Command execution time is NOT counted in budget
- If command takes 1 minute to execute each time, total time could be: 2m (budget) + (attempts * 1m execution) = ~12 minutes

**Source**: retry_budget field in src/cook/retry_v2.rs:46-47, tests at lines 675-708

---

### Issue: Fallback Command Also Failing

**Symptoms**:
- Primary command fails after retries
- Fallback command executes but also fails
- Workflow stops

**Problem**: Fallback command isn't reliable

```yaml
retry_config:
  attempts: 3
  on_failure:
    fallback:
      command: "curl https://backup-api.com/data"  # This can also fail!
```

**Solution**: Make fallback truly reliable

```yaml
retry_config:
  attempts: 3
  on_failure:
    fallback:
      command: "cat /cache/data.json"  # Local cache, very reliable
```

**Best Practice**: Fallback commands should be:
- Local operations (file reads, not network calls)
- Idempotent
- Very unlikely to fail
- Fast

---

### Issue: Retry Metrics Not Matching Expectations

**Symptoms**:
- Metrics show different attempt count than configured
- Unexpected success/failure counts

**Debugging**:

```rust
// Access retry metrics for debugging
let metrics = executor.metrics();
println!("Total attempts: {}", metrics.total_attempts);
println!("Successful: {}", metrics.successful_attempts);
println!("Failed: {}", metrics.failed_attempts);
println!("Retries: {:?}", metrics.retries);  // Vec<(attempt, delay)>
```

**Source**: RetryMetrics struct in src/cook/retry_v2.rs:399-422

**Check**:
- `total_attempts` = successful + failed
- `retries` vector shows actual delays used
- Compare with configured backoff strategy

---

### Debugging Workflow

When retry behavior is unexpected:

1. **Check retry_on matchers**:
   - Verify error message matches configured matchers
   - Remember matching is case-insensitive
   - Empty `retry_on` = retry all errors

2. **Check backoff configuration**:
   - Verify `max_delay` is set
   - Check `initial_delay` isn't too high
   - Ensure `backoff` strategy matches intent

3. **Check retry_budget**:
   - Remember budget is delay time, not total time
   - Budget checked before each retry
   - Command execution time NOT included

4. **Enable jitter for parallel workflows**:
   - Always use jitter in MapReduce
   - Set `jitter_factor` between 0.1 and 0.5

5. **Use selective retry**:
   - Don't retry permanent errors (404, 401, 400)
   - Use `retry_on` to specify transient errors only

6. **Monitor metrics**:
   - Access `RetryMetrics` for actual attempt counts
   - Verify delays match expectations
   - Check circuit breaker state transitions

---

### Getting Help

If retry behavior is still unclear:

1. **Check retry_v2.rs implementation**: src/cook/retry_v2.rs
2. **Review tests**: src/cook/retry_v2.rs:463-748 (comprehensive test coverage)
3. **Enable verbose logging**: Add logging around retry logic to see what's happening
4. **Test with simple cases**: Start with fixed backoff and 2-3 attempts to isolate issue

### See Also

- [Basic Retry Configuration]./basic-retry-configuration.md - Configuration fundamentals
- [Backoff Strategies]./backoff-strategies.md - Understanding backoff behavior
- [Conditional Retry with Error Matchers]./conditional-retry-with-error-matchers.md - Error matching details
- [Best Practices]./best-practices.md - Avoiding common pitfalls
- [Complete Examples]./complete-examples.md - Working configurations