brokk-mj-controller 2.10.0

Daemon-side controller, session manager, and web server for Mjolnir
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
//! Session close, force-stop, and permanent-destruction transitions.

use std::time::Duration;

use anyhow::{Context, Result, bail, ensure};

use crate::session_manager::{SessionManagerControl, new_command_id};
use mj_core::state::{CheckpointMetadata, SessionRecord, SessionState};

use crate::targets::{self, CommandExecutor, ProcessExecutor, ProvisionStage, ProvisionStageGuard};
use mj_core::relay::{RelayCommand, RelayExecutionState};

use super::backend::backend_locator;
use super::checkpoint::{
    CheckpointExportPolicy, LatchExclusivity, prune_replaced_checkpoint,
    release_projection_behind_checkpoint, verify_installed_checkpoint_gate, wait_for_relay_closed,
};
use super::worker_restart::WorkerRestartLeftNoWorker;
use super::worktree::{cleanup_managed_worktree, retire_managed_worktree};
use super::{Controller, now, persist_session_record_transition_or_restore};

/// What destroying a session does with its managed worktree's git branch.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum BranchDisposition {
    /// Delete the branch with the rest of the session. What a destroy does.
    Delete,
    /// Leave the branch in the repository. What archiving does, because the
    /// branch may hold work the user still wants.
    Keep,
}

impl Controller {
    /// Checkpoint, ask the harness to close, and only then tear down the exact
    /// provisioned target. Checkpoint failure is deliberately non-destructive,
    /// except when the checkpoint's worker restart left no live worker: that
    /// records `Error` and keeps the target for a later resume or forced close.
    pub async fn close_session(&mut self, session_id: &str) -> Result<()> {
        self.close_session_controlled(session_id, &ProcessExecutor)
            .await
    }

    pub async fn close_session_controlled(
        &mut self,
        session_id: &str,
        executor: &(impl CommandExecutor + Sync),
    ) -> Result<()> {
        if self
            .close_session_controlled_with_manager(session_id, executor, None, None)
            .await?
        {
            self.cleanup_stopped_target(session_id, executor)?;
        }
        Ok(())
    }

    pub async fn close_session_managed_controlled(
        &mut self,
        session_id: &str,
        executor: &(impl CommandExecutor + Sync),
        manager: &SessionManagerControl,
    ) -> Result<bool> {
        self.close_session_controlled_with_manager(session_id, executor, Some(manager), None)
            .await
    }

    pub(super) async fn close_session_for_move(
        &mut self,
        session_id: &str,
        executor: &(impl CommandExecutor + Sync),
        manager: &SessionManagerControl,
        operation: &mut mj_core::state::MoveOperation,
        preparation: Option<&mj_core::state::MovePreparation>,
    ) -> Result<bool> {
        self.prepare_move_source_checkpoint(session_id, executor, manager, operation)
            .await?;
        self.close_session_controlled_with_manager(
            session_id,
            executor,
            Some(manager),
            Some((operation, preparation)),
        )
        .await
    }

    async fn close_session_controlled_with_manager(
        &mut self,
        session_id: &str,
        executor: &(impl CommandExecutor + Sync),
        manager: Option<&SessionManagerControl>,
        move_intent: Option<(
            &mut mj_core::state::MoveOperation,
            Option<&mj_core::state::MovePreparation>,
        )>,
    ) -> Result<bool> {
        let previous = self
            .state
            .sessions
            .get(session_id)
            .with_context(|| format!("unknown session {session_id}"))?
            .clone();
        let record = self.state.sessions.get_mut(session_id).unwrap();
        // Persist the close intent before beginning its checkpoint. A process
        // exit anywhere below must leave enough state for the next controller
        // to retry the close, even when no checkpoint has been installed yet.
        apply_close_checkpoint_started(record, now());
        self.persist_session_transition_or_restore(
            session_id,
            &previous,
            "persist closing state before checkpointing the session",
        )?;

        // Close seals the relay at the exact latched cursor, so this checkpoint
        // keeps its exclusive connection until the relay reports Closed.
        let mut latched = match self
            .checkpoint_session_latched(
                session_id,
                executor,
                manager,
                LatchExclusivity::HoldThroughClose,
                CheckpointExportPolicy::ReuseUnchangedArchive,
            )
            .await
        {
            Ok(latched) => latched,
            Err(error) => {
                let record = self.state.sessions.get_mut(session_id).unwrap();
                // The target is kept even when the restart left no worker: a
                // forced destroy and a resume's pre-clean both use it to tear
                // down the dead container.
                apply_close_checkpoint_failure(record, &previous, &error, now());
                return Err(
                    self.persist_failed_checkpoint_state_or_restore(session_id, &previous, error)
                );
            }
        };

        let artifact = latched.artifact.clone();
        let record = self.state.sessions.get_mut(session_id).unwrap();
        record.state = SessionState::Closing;
        record.native_session_id = Some(artifact.native_session_id.clone());
        record.checkpoint = Some(artifact.metadata.clone());
        record.updated_at = now();
        record.last_error = None;
        record.last_checkpoint_error = None;
        self.persist_checkpoint_transition_or_restore(
            session_id,
            &previous,
            "persist verified checkpoint and closing state before sealing the relay",
        )?;
        if let Some((operation, preparation)) = move_intent {
            // The source is still behind an unsealed barrier. A destination
            // preflight error must release it and leave its processes alive.
            if let Err(error) = self.validate_move_checkpoint(operation, preparation, executor) {
                let record = self.state.sessions.get_mut(session_id).unwrap();
                record.state = previous.state;
                record.last_error = Some(format!("{error:#}"));
                self.persist_session_transition_or_restore(
                    session_id,
                    &previous,
                    "restore source after move preflight failure",
                )?;
                return Err(error);
            }
            operation.checkpoint = Some(artifact.metadata.clone());
            operation.updated_at = now();
            crate::database::save_move_operation(operation)?;
        }
        prune_replaced_checkpoint(previous.checkpoint.as_ref(), &artifact.metadata);
        // A stopping session will not checkpoint again, so this is its last
        // chance to release what its checkpoint now covers.
        release_projection_behind_checkpoint(session_id, &artifact.metadata);

        let close_command_id = new_command_id("close")?;
        let barrier_command_id = latched.barrier_command_id.clone();
        let close_result = {
            let _closing = ProvisionStageGuard::new(executor, ProvisionStage::Closing);
            latched
                .relay
                .connection_mut()
                .submit(
                    close_command_id,
                    RelayCommand::Close {
                        barrier_command_id: barrier_command_id.clone(),
                        expected: latched.cursor.clone(),
                    },
                )
                .await
        };
        if let Err(error) = close_result {
            self.record_interrupted_close(session_id, &error)?;
            return Err(error.context("seal verified checkpoint for close"));
        }
        let close_result = {
            let _closing = ProvisionStageGuard::new(executor, ProvisionStage::Closing);
            latched
                .relay
                .connection_mut()
                .submit(
                    new_command_id("checkpoint-complete")?,
                    RelayCommand::CompleteCheckpoint { barrier_command_id },
                )
                .await
        };
        if let Err(error) = close_result {
            self.record_interrupted_close(session_id, &error)?;
            return Err(error.context("release verified close checkpoint"));
        }
        let close_result = {
            let _closing = ProvisionStageGuard::new(executor, ProvisionStage::Closing);
            wait_for_relay_closed(latched.relay.connection_mut()).await
        };
        if let Err(error) = close_result {
            self.record_interrupted_close(session_id, &error)?;
            return Err(error);
        }
        latched.relay.release();

        match self.destroy_after_verified_checkpoint(session_id, &artifact.metadata, executor) {
            Ok(deferred) => Ok(deferred),
            Err(error) => {
                self.record_interrupted_close(session_id, &error)?;
                Err(error)
            }
        }
    }

    /// Resume the durable closing state after a controller restart. If the
    /// relay had accepted Close, wait for it and destroy through the exact
    /// installed checkpoint gate. If it had not, take a fresh checkpoint;
    /// the previously installed archive may have become stale after EOF
    /// released its barrier.
    pub async fn recover_interrupted_close_managed(
        &mut self,
        session_id: &str,
        executor: &(impl CommandExecutor + Sync),
        manager: &SessionManagerControl,
    ) -> Result<bool> {
        let (state, verified) = {
            let session = self
                .state
                .sessions
                .get(session_id)
                .with_context(|| format!("unknown session {session_id}"))?;
            ensure!(
                matches!(
                    session.state,
                    SessionState::Closing | SessionState::Destroying
                ),
                "session {session_id} has no interrupted close to recover"
            );
            (session.state, session.checkpoint.clone())
        };
        if state == SessionState::Destroying {
            let verified = verified.context("destroying session has no verified checkpoint")?;
            return self.destroy_after_verified_checkpoint(session_id, &verified, executor);
        }
        ensure!(
            state == SessionState::Closing,
            "session {session_id} has no relay close to recover"
        );
        let handle = manager
            .wait_for_session(session_id, Duration::from_secs(5))
            .await?;
        let mut lease = handle.lease_connection().await?;
        let execution = lease.connection_mut().sync().await?.operational.execution;
        match execution {
            RelayExecutionState::Closed => {}
            RelayExecutionState::Closing => {
                let _closing = ProvisionStageGuard::new(executor, ProvisionStage::Closing);
                wait_for_relay_closed(lease.connection_mut()).await?;
            }
            RelayExecutionState::Idle | RelayExecutionState::Running => {
                lease.release();
                return self
                    .close_session_controlled_with_manager(
                        session_id,
                        executor,
                        Some(manager),
                        None,
                    )
                    .await;
            }
        }
        lease.release();
        let verified = verified.context("closed relay has no verified checkpoint")?;
        self.destroy_after_verified_checkpoint(session_id, &verified, executor)
    }

    fn record_interrupted_close(&mut self, session_id: &str, error: &anyhow::Error) -> Result<()> {
        let record = self.state.sessions.get_mut(session_id).unwrap();
        apply_interrupted_close_error(record, error, &now());
        self.persist_session_state(session_id)
    }

    /// Execute cleanup only after the close state machine has installed a
    /// verified checkpoint on the record.
    fn destroy_after_verified_checkpoint(
        &mut self,
        session_id: &str,
        verified: &CheckpointMetadata,
        executor: &impl CommandExecutor,
    ) -> Result<bool> {
        self.destroy_after_verified_checkpoint_with(
            session_id,
            verified,
            executor,
            crate::database::save_lifecycle_session,
        )
    }

    fn destroy_after_verified_checkpoint_with(
        &mut self,
        session_id: &str,
        verified: &CheckpointMetadata,
        executor: &impl CommandExecutor,
        persist: impl Fn(&SessionRecord) -> Result<()>,
    ) -> Result<bool> {
        let target_mutex = crate::recovery_gate::worker_target_mutex(session_id);
        let _target_guard = target_mutex.lock().map_err(|_| {
            anyhow::anyhow!("worker target ownership lock poisoned for {session_id}")
        })?;
        let session = self
            .state
            .sessions
            .get(session_id)
            .with_context(|| format!("unknown session {session_id}"))?
            .clone();
        ensure!(
            matches!(
                session.state,
                SessionState::Closing | SessionState::Destroying
            ),
            "refusing to destroy session {session_id}: it is not closing or destroying"
        );
        ensure!(
            session.checkpoint.as_ref() == Some(verified),
            "refusing to destroy session {session_id}: verified checkpoint gate is stale"
        );
        if session.state == SessionState::Closing {
            let record = self.state.sessions.get_mut(session_id).unwrap();
            record.state = SessionState::Destroying;
            record.updated_at = now();
            record.last_error = None;
            persist_session_record_transition_or_restore(
                &mut self.state,
                session_id,
                &session,
                "persist destroying state before target cleanup",
                &persist,
            )?;
        }

        let destroying = self
            .state
            .sessions
            .get(session_id)
            .expect("destroying session disappeared")
            .clone();
        {
            let _verifying = ProvisionStageGuard::new(executor, ProvisionStage::Verifying);
            verify_installed_checkpoint_gate(session_id, verified)?;
        }
        // The reviewer's native session lives on the target that is about to
        // go. Recording that now, before the target is torn down, is what
        // stops a resumed session from trying to reload a conversation that no
        // longer exists; its transcript is kept for reference either way.
        if let Err(error) = crate::database::lose_reviewer_continuity(session_id) {
            tracing::warn!(
                session_id,
                error = format!("{error:#}"),
                "could not record that the second-opinion conversation ends with this target"
            );
        }
        let locator = destroying
            .target
            .as_ref()
            .context("session has no target")?;
        let backend = backend_locator(locator, &destroying, &self.config)?;
        let deferred = if self.state.subagents.contains_key(session_id) {
            targets::borrowed_worker_cleanup_plan(&backend, session_id)?.execute(executor)?;
            false
        } else if let Some(plan) = targets::quiesce_plan(&backend, session_id)? {
            plan.execute(executor)?;
            true
        } else {
            execute_target_cleanup(&backend, session_id, executor)?;
            false
        };
        if let Some(worktree) = &destroying.managed_worktree {
            retire_managed_worktree(executor, worktree)
                .context("retire managed raw-session worktree after verified close")?;
        }
        let record = self.state.sessions.get_mut(session_id).unwrap();
        record.state = SessionState::Stopped;
        if !deferred {
            record.target = None;
        }
        record.updated_at = now();
        record.last_error = None;
        persist_session_record_transition_or_restore(
            &mut self.state,
            session_id,
            &destroying,
            "persist stopped state after target cleanup",
            &persist,
        )?;
        Ok(deferred)
    }

    /// Finish storage cleanup for a stopped Podman target retained by the
    /// quiescence transition. The locator stays durable until every command
    /// succeeds, making daemon restart and explicit retry idempotent.
    pub fn cleanup_stopped_target(
        &mut self,
        session_id: &str,
        executor: &impl CommandExecutor,
    ) -> Result<()> {
        self.cleanup_stopped_target_with(
            session_id,
            executor,
            crate::database::save_lifecycle_session,
        )
    }

    fn cleanup_stopped_target_with(
        &mut self,
        session_id: &str,
        executor: &impl CommandExecutor,
        persist: impl Fn(&SessionRecord) -> Result<()>,
    ) -> Result<()> {
        let target_mutex = crate::recovery_gate::worker_target_mutex(session_id);
        let _target_guard = target_mutex.lock().map_err(|_| {
            anyhow::anyhow!("worker target ownership lock poisoned for {session_id}")
        })?;
        let previous = self
            .state
            .sessions
            .get(session_id)
            .with_context(|| format!("unknown session {session_id}"))?
            .clone();
        ensure!(
            previous.state == SessionState::Stopped,
            "refusing deferred cleanup for active session {session_id}"
        );
        let Some(locator) = previous.target.as_ref() else {
            return Ok(());
        };
        let backend = backend_locator(locator, &previous, &self.config)?;
        ensure!(
            targets::quiesce_plan(&backend, session_id)?.is_some(),
            "session {session_id} retained a non-Podman target after stopping"
        );
        if let Err(error) = execute_target_cleanup(&backend, session_id, executor) {
            let record = self.state.sessions.get_mut(session_id).unwrap();
            record.updated_at = now();
            record.last_error = Some(format!("deferred target cleanup failed: {error:#}"));
            let persisted = persist_session_record_transition_or_restore(
                &mut self.state,
                session_id,
                &previous,
                "persist deferred target cleanup failure",
                &persist,
            );
            return match persisted {
                Ok(()) => Err(error),
                Err(persist_error) => Err(error.context(format!(
                    "also failed to persist deferred target cleanup failure: {persist_error:#}"
                ))),
            };
        }
        let record = self.state.sessions.get_mut(session_id).unwrap();
        record.target = None;
        record.updated_at = now();
        record.last_error = None;
        persist_session_record_transition_or_restore(
            &mut self.state,
            session_id,
            &previous,
            "persist completion of deferred Podman target cleanup",
            &persist,
        )
    }

    /// Tear down the current target without taking a fresh checkpoint, then
    /// leave the logical session resumable from its latest verified archive.
    pub fn force_stop(
        &mut self,
        session_id: &str,
        executor: &impl CommandExecutor,
    ) -> Result<bool> {
        self.force_stop_with(
            session_id,
            executor,
            crate::database::save_lifecycle_session,
        )
    }

    fn force_stop_with(
        &mut self,
        session_id: &str,
        executor: &impl CommandExecutor,
        persist: impl Fn(&SessionRecord) -> Result<()>,
    ) -> Result<bool> {
        let session = self
            .state
            .sessions
            .get(session_id)
            .with_context(|| format!("unknown session {session_id}"))?
            .clone();
        ensure!(
            session.state.is_active(),
            "session {session_id} is already inactive"
        );
        let checkpoint = session
            .checkpoint
            .as_ref()
            .context("force stop requires an existing recovery archive")?;
        // Force stop skips a new checkpoint, never the checksum gate on the
        // archive that makes the logical session resumable afterwards.
        {
            let _verifying = ProvisionStageGuard::new(executor, ProvisionStage::Verifying);
            verify_installed_checkpoint_gate(session_id, checkpoint)
                .context("verify the recovery archive before force stopping")?;
        }
        let mut deferred = false;
        if let Some(locator) = &session.target {
            let backend = backend_locator(locator, &session, &self.config)?;
            if let Some(plan) = targets::quiesce_plan(&backend, session_id)? {
                plan.execute(executor)?;
                deferred = true;
            } else {
                execute_target_cleanup(&backend, session_id, executor)?;
            }
        }
        if let Some(worktree) = &session.managed_worktree {
            retire_managed_worktree(executor, worktree)
                .context("retire managed raw-session worktree after force stop")?;
        }
        let record = self.state.sessions.get_mut(session_id).unwrap();
        record.state = SessionState::Stopped;
        if !deferred {
            record.target = None;
        }
        record.updated_at = now();
        record.last_error = None;
        record.last_checkpoint_error = None;
        persist_session_record_transition_or_restore(
            &mut self.state,
            session_id,
            &session,
            "persist stopped state after force stopping the current target",
            &persist,
        )?;
        Ok(deferred)
    }

    /// Permanently destroy an inactive session and every artifact Hel owns for it.
    /// External cleanup happens before the durable record is dropped so failures
    /// remain visible and retryable.
    pub fn destroy_session_controlled(
        &mut self,
        session_id: &str,
        executor: &impl CommandExecutor,
    ) -> Result<()> {
        self.destroy_session_controlled_with(session_id, executor, BranchDisposition::Delete)
    }

    /// The same, with a say in what happens to the managed worktree's branch.
    ///
    /// A session that is already `Stopped` has had its checkout removed by
    /// [`retire_managed_worktree`], so only the branch is left for
    /// [`cleanup_managed_worktree`] to take. [`BranchDisposition::Keep`]
    /// therefore just skips that call, which is what archiving wants: the
    /// record, the checkpoint, and the attachments go, and the branch stays.
    pub fn destroy_session_controlled_with(
        &mut self,
        session_id: &str,
        executor: &impl CommandExecutor,
        branch: BranchDisposition,
    ) -> Result<()> {
        let session = self
            .state
            .sessions
            .get(session_id)
            .with_context(|| format!("unknown session {session_id}"))?
            .clone();
        if session.state.is_active() {
            bail!("refusing to destroy active session {session_id}");
        }
        if branch == BranchDisposition::Delete
            && let Some(worktree) = &session.managed_worktree
        {
            cleanup_managed_worktree(executor, worktree)
                .context("remove managed raw-session worktree")?;
        }
        if let Some(checkpoint) = &session.checkpoint
            && let Err(error) = std::fs::remove_file(&checkpoint.archive_path)
            && error.kind() != std::io::ErrorKind::NotFound
        {
            return Err(error).with_context(|| {
                format!(
                    "remove session recovery archive {}",
                    checkpoint.archive_path.display()
                )
            });
        }
        mj_core::attachment::AttachmentStore::controller(session_id)?
            .remove_session_data()
            .context("remove session image attachments")?;
        crate::database::delete_session(session_id)
            .context("destroy stopped session in database")?;
        self.state.subagents.remove(session_id);
        self.state.destroy_stopped_session(session_id)?;
        Ok(())
    }

    /// Permanently destroy a session from any state, without checkpointing
    /// and without requiring a recovery archive.
    ///
    /// Unlike [`Controller::destroy_session_controlled`], this accepts active
    /// states: it tears the live target down with the same close plan a
    /// verified close uses, so the owning process group dies before any files
    /// go. External cleanup happens before the durable record is dropped so
    /// failures stay visible and retryable; the recovery archive is removed,
    /// which is what makes the destruction irreversible.
    pub fn force_destroy_session(
        &mut self,
        session_id: &str,
        executor: &impl CommandExecutor,
    ) -> Result<()> {
        self.force_destroy_session_with(session_id, executor, crate::database::delete_session)
    }

    fn force_destroy_session_with(
        &mut self,
        session_id: &str,
        executor: &impl CommandExecutor,
        delete: impl Fn(&str) -> Result<()>,
    ) -> Result<()> {
        let session = self
            .state
            .sessions
            .get(session_id)
            .with_context(|| format!("unknown session {session_id}"))?
            .clone();
        // A session destroyed for good keeps nothing, including a broker an
        // earlier failure left running; retiring it first also stops a live
        // writer from recreating files under the teardown below.
        if let Some(locator) = &session.target {
            let backend = backend_locator(locator, &session, &self.config)?;
            if self.state.subagents.contains_key(session_id) {
                targets::borrowed_worker_cleanup_plan(&backend, session_id)?.execute(executor)?;
            } else {
                execute_target_cleanup(&backend, session_id, executor)?;
            }
        }
        if let Some(worktree) = &session.managed_worktree {
            cleanup_managed_worktree(executor, worktree)
                .context("remove managed raw-session worktree")?;
        }
        if let Some(checkpoint) = &session.checkpoint
            && let Err(error) = std::fs::remove_file(&checkpoint.archive_path)
            && error.kind() != std::io::ErrorKind::NotFound
        {
            return Err(error).with_context(|| {
                format!(
                    "remove session recovery archive {}",
                    checkpoint.archive_path.display()
                )
            });
        }
        mj_core::attachment::AttachmentStore::controller(session_id)?
            .remove_session_data()
            .context("remove session image attachments")?;
        delete(session_id).context("force destroy session in database")?;
        self.state.subagents.remove(session_id);
        self.state.destroy_session_force(session_id)?;
        Ok(())
    }
}

fn execute_target_cleanup(
    backend: &targets::TargetLocator,
    session_id: &str,
    executor: &impl CommandExecutor,
) -> Result<()> {
    if let Err(cleanup_error) = targets::close_plan(backend, session_id)?.execute(executor) {
        match targets::cleanup_target_is_confirmed_absent(backend, session_id, executor) {
            Ok(true) => {
                tracing::warn!(
                    session_id,
                    error = format!("{cleanup_error:#}"),
                    "target cleanup command failed, but the target was confirmed absent"
                );
            }
            Ok(false) => {
                tracing::error!(
                    session_id,
                    error = format!("{cleanup_error:#}"),
                    "target cleanup failed and the target is still present"
                );
                return Err(cleanup_error);
            }
            Err(probe_error) => {
                tracing::error!(
                    session_id,
                    cleanup_error = format!("{cleanup_error:#}"),
                    probe_error = format!("{probe_error:#}"),
                    "target cleanup failed and exact absence could not be confirmed"
                );
                return Err(cleanup_error.context(format!(
                    "target cleanup failed and exact absence could not be confirmed: {probe_error:#}"
                )));
            }
        }
    }
    Ok(())
}

fn apply_close_checkpoint_started(record: &mut SessionRecord, updated_at: String) {
    record.state = SessionState::Closing;
    record.updated_at = updated_at;
    record.last_checkpoint_error = None;
}

/// Record a close whose checkpoint failed.
///
/// An ordinary failure is non-destructive: the session returns to the state it
/// had. A restart that left no live worker cannot return to Running, because
/// nothing is listening there any more; it records `Error` so the session stops
/// being polled, and keeps its target for a later resume or forced close.
fn apply_close_checkpoint_failure(
    record: &mut SessionRecord,
    previous: &SessionRecord,
    error: &anyhow::Error,
    updated_at: String,
) {
    if WorkerRestartLeftNoWorker::marks(error) {
        record.state = SessionState::Error;
        record.last_error = Some(format!(
            "close failed and left the session without a live worker; retry the close, \
             resume from its checkpoint, or close it with --force: {error:#}"
        ));
    } else {
        record.state = previous.state;
    }
    record.last_checkpoint_error = Some(format!("{error:#}"));
    record.updated_at = updated_at;
}

fn apply_interrupted_close_error(
    record: &mut SessionRecord,
    error: &anyhow::Error,
    updated_at: &str,
) {
    let destroying = record.state == SessionState::Destroying;
    if !destroying {
        record.state = SessionState::Closing;
    }
    record.updated_at = updated_at.to_owned();
    record.last_error = Some(if destroying {
        format!("target cleanup is safely retryable from its verified checkpoint: {error:#}")
    } else {
        format!("close is safely resumable from its verified checkpoint: {error:#}")
    });
}

#[cfg(test)]
mod tests;