|
| 1 | +# Plan: Fix Orphaned Messages on Run Failure |
| 2 | + |
| 3 | +## Problem Statement |
| 4 | + |
| 5 | +When a run fails, messages for pending tasks remain in the queue indefinitely, causing: |
| 6 | +1. **Resource waste**: Workers continuously poll orphaned messages |
| 7 | +2. **Performance degradation**: Queue operations slow down over time |
| 8 | +3. **Map step issues**: Failing one map task leaves N-1 sibling messages orphaned |
| 9 | +4. **Type violations**: Deterministic errors retry unnecessarily |
| 10 | + |
| 11 | +## Current Behavior |
| 12 | + |
| 13 | +### When fail_task is called |
| 14 | +```sql |
| 15 | +-- Only archives the single failing task's message |
| 16 | +SELECT pgmq.archive('pgflow_tasks_queue', fail_task.msg_id); |
| 17 | +-- Leaves all other queued messages orphaned |
| 18 | +``` |
| 19 | + |
| 20 | +### When type constraint violation occurs |
| 21 | +```sql |
| 22 | +-- Raises exception, causes retries |
| 23 | +RAISE EXCEPTION 'Map step % expects array input...'; |
| 24 | +-- Transaction rolls back, but retries will hit same error |
| 25 | +``` |
| 26 | + |
| 27 | +## Implementation Plan |
| 28 | + |
| 29 | +### 1. Update fail_task Function |
| 30 | +**File**: `pkgs/core/schemas/0100_function_fail_task.sql` |
| 31 | + |
| 32 | +Add after marking run as failed (around line 47): |
| 33 | +```sql |
| 34 | +-- Archive all pending messages for this run |
| 35 | +WITH tasks_to_archive AS ( |
| 36 | + SELECT t.msg_id |
| 37 | + FROM pgflow.step_tasks t |
| 38 | + WHERE t.run_id = fail_task.run_id |
| 39 | + AND t.status = 'pending' |
| 40 | + AND t.msg_id IS NOT NULL |
| 41 | +) |
| 42 | +SELECT pgmq.archive('pgflow_tasks_queue', msg_id) |
| 43 | +FROM tasks_to_archive; |
| 44 | +``` |
| 45 | + |
| 46 | +### 2. Update complete_task for Type Violations |
| 47 | +**File**: `pkgs/core/schemas/0100_function_complete_task.sql` |
| 48 | + |
| 49 | +Replace the current RAISE EXCEPTION block (lines 115-120) with: |
| 50 | +```sql |
| 51 | +IF v_dependent_map_slug IS NOT NULL THEN |
| 52 | + -- Mark run as failed immediately (no retries for type violations) |
| 53 | + UPDATE pgflow.runs |
| 54 | + SET status = 'failed', |
| 55 | + failed_at = now(), |
| 56 | + error = format('Type contract violation: Map step %s expects array input but dependency %s produced %s (output: %s)', |
| 57 | + v_dependent_map_slug, |
| 58 | + complete_task.step_slug, |
| 59 | + CASE WHEN complete_task.output IS NULL THEN 'null' |
| 60 | + ELSE jsonb_typeof(complete_task.output) END, |
| 61 | + complete_task.output) |
| 62 | + WHERE run_id = complete_task.run_id; |
| 63 | + |
| 64 | + -- Archive ALL pending messages for this run |
| 65 | + PERFORM pgmq.archive('pgflow_tasks_queue', t.msg_id) |
| 66 | + FROM pgflow.step_tasks t |
| 67 | + WHERE t.run_id = complete_task.run_id |
| 68 | + AND t.status = 'pending' |
| 69 | + AND t.msg_id IS NOT NULL; |
| 70 | + |
| 71 | + -- Mark the current task as failed (not completed) |
| 72 | + UPDATE pgflow.step_tasks |
| 73 | + SET status = 'failed', |
| 74 | + failed_at = now(), |
| 75 | + error_message = format('Type contract violation: produced %s instead of array', |
| 76 | + CASE WHEN complete_task.output IS NULL THEN 'null' |
| 77 | + ELSE jsonb_typeof(complete_task.output) END) |
| 78 | + WHERE run_id = complete_task.run_id |
| 79 | + AND step_slug = complete_task.step_slug |
| 80 | + AND task_index = complete_task.task_index; |
| 81 | + |
| 82 | + -- Return empty result set (task not completed) |
| 83 | + RETURN QUERY SELECT * FROM pgflow.step_tasks WHERE false; |
| 84 | + RETURN; |
| 85 | +END IF; |
| 86 | +``` |
| 87 | + |
| 88 | +### 3. Add Supporting Index |
| 89 | +**File**: New migration or add to existing |
| 90 | + |
| 91 | +```sql |
| 92 | +-- Speed up the archiving query |
| 93 | +CREATE INDEX IF NOT EXISTS idx_step_tasks_pending_with_msg |
| 94 | +ON pgflow.step_tasks(run_id, status) |
| 95 | +WHERE status = 'pending' AND msg_id IS NOT NULL; |
| 96 | +``` |
| 97 | + |
| 98 | +## Testing |
| 99 | + |
| 100 | +### Tests Already Written (Stashed) |
| 101 | + |
| 102 | +1. **`supabase/tests/fail_task/archive_sibling_map_tasks.test.sql`** |
| 103 | + - Verifies all map task messages are archived when one fails |
| 104 | + - Tests: 8 assertions about message archiving and status |
| 105 | + |
| 106 | +2. **`supabase/tests/initial_tasks_null/archive_messages_on_type_constraint_failure.test.sql`** |
| 107 | + - Verifies type violations archive all pending messages |
| 108 | + - Tests: 8 assertions about queue cleanup and run status |
| 109 | + |
| 110 | +### How to Run Tests |
| 111 | +```bash |
| 112 | +# After unstashing and implementing the fixes: |
| 113 | +pnpm nx test:pgtap core -- supabase/tests/fail_task/archive_sibling_map_tasks.test.sql |
| 114 | +pnpm nx test:pgtap core -- supabase/tests/initial_tasks_null/archive_messages_on_type_constraint_failure.test.sql |
| 115 | +``` |
| 116 | + |
| 117 | +## Migration Considerations |
| 118 | + |
| 119 | +### Backward Compatibility |
| 120 | +- New behavior only affects failed runs (safe) |
| 121 | +- Archiving preserves messages (can be recovered if needed) |
| 122 | +- No schema changes to existing tables |
| 123 | + |
| 124 | +### Performance Impact |
| 125 | +- One-time cost during failure (acceptable) |
| 126 | +- Prevents ongoing performance degradation (improvement) |
| 127 | +- Index ensures archiving query is efficient |
| 128 | + |
| 129 | +### Rollback Plan |
| 130 | +If issues arise: |
| 131 | +1. Remove the archiving logic |
| 132 | +2. Messages remain in queue (old behavior) |
| 133 | +3. No data loss since we archive, not delete |
| 134 | + |
| 135 | +## Edge Cases to Consider |
| 136 | + |
| 137 | +### 1. Concurrent Task Completion |
| 138 | +If multiple tasks complete/fail simultaneously: |
| 139 | +- PostgreSQL row locks ensure consistency |
| 140 | +- Each failure archives all pending messages |
| 141 | +- Idempotent: archiving already-archived messages is safe |
| 142 | + |
| 143 | +### 2. Very Large Map Steps |
| 144 | +For maps with 1000+ tasks: |
| 145 | +- Archiving might take several seconds |
| 146 | +- Consider batching if performance issues arise |
| 147 | +- Current approach should handle up to ~10k tasks reasonably |
| 148 | + |
| 149 | +### 3. Mixed Step Types |
| 150 | +When run has both map and single steps: |
| 151 | +- Archive logic handles all pending tasks regardless of type |
| 152 | +- Correctly archives both map siblings and unrelated pending tasks |
| 153 | + |
| 154 | +## Future Enhancements (Not for this PR) |
| 155 | + |
| 156 | +1. **Selective Archiving**: Only archive tasks that can't proceed |
| 157 | +2. **Batch Operations**: Archive in chunks for very large runs |
| 158 | +3. **Recovery Mechanism**: Function to unarchive and retry |
| 159 | +4. **Monitoring**: Track archived message counts for alerting |
| 160 | + |
| 161 | +## Success Criteria |
| 162 | + |
| 163 | +- [ ] All tests pass (both new test files) |
| 164 | +- [ ] No orphaned messages after run failure |
| 165 | +- [ ] Type violations don't retry |
| 166 | +- [ ] Performance acceptable for maps with 100+ tasks |
| 167 | +- [ ] No impact on successful run performance |
| 168 | + |
| 169 | +## Implementation Checklist |
| 170 | + |
| 171 | +- [ ] Update `fail_task` function |
| 172 | +- [ ] Update `complete_task` function |
| 173 | +- [ ] Add database index |
| 174 | +- [ ] Unstash and run tests |
| 175 | +- [ ] Test with large map steps (100+ tasks) |
| 176 | +- [ ] Update migration file |
| 177 | +- [ ] Document behavior change in function comments |
| 178 | + |
| 179 | +## Notes |
| 180 | + |
| 181 | +- This fix is **critical for production** - without it, queue performance will degrade over time |
| 182 | +- Type violations are **deterministic** - retrying them is always wasteful |
| 183 | +- Archiving (vs deleting) preserves debugging capability |
| 184 | +- The fix is relatively simple (~30 lines of SQL) but high impact |
0 commit comments