Repository navigation
Commit c643ca1
test: de-flake transform_decisions retention cap via injectable cap
The retention-cap test wrote TRANSFORM_DECISIONS_RETENTION+5 (2005) rows, and
each writeRow opens a fresh DB connection (the off-hot-path telemetry writer's
design). Under CI load that loop exceeded the 5s test timeout and flaked the
plugin check intermittently (unrelated to the change under test). The prune SQL
is cap-agnostic (LIMIT ?), so the behavior is identical at any cap: add a
test-only retention override and exercise the prune with a cap of 10 instead of
2000. Runtime drops from ~10s (timeout) to ~0.25s. A guard test pins the real
TRANSFORM_DECISIONS_RETENTION constant so the production cap can't silently
drift.
Co-authored-by: Alfonso [Magic Context] <288211368+alfonso-magic-context@users.noreply.github.com>1 parent aa93c58 commit c643ca1
2 files changed
Lines changed: 31 additions & 6 deletions
File tree
- packages/plugin/src/features/magic-context
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
55 | 55 | | |
56 | 56 | | |
57 | 57 | | |
58 | | - | |
| 58 | + | |
| 59 | + | |
| 60 | + | |
| 61 | + | |
| 62 | + | |
| 63 | + | |
| 64 | + | |
| 65 | + | |
| 66 | + | |
| 67 | + | |
| 68 | + | |
59 | 69 | | |
60 | | - | |
| 70 | + | |
61 | 71 | | |
62 | 72 | | |
63 | 73 | | |
64 | | - | |
| 74 | + | |
65 | 75 | | |
66 | 76 | | |
67 | 77 | | |
| |||
75 | 85 | | |
76 | 86 | | |
77 | 87 | | |
78 | | - | |
| 88 | + | |
79 | 89 | | |
80 | 90 | | |
81 | | - | |
| 91 | + | |
| 92 | + | |
| 93 | + | |
| 94 | + | |
| 95 | + | |
| 96 | + | |
82 | 97 | | |
83 | 98 | | |
84 | 99 | | |
| |||
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
82 | 82 | | |
83 | 83 | | |
84 | 84 | | |
| 85 | + | |
| 86 | + | |
| 87 | + | |
| 88 | + | |
| 89 | + | |
| 90 | + | |
85 | 91 | | |
86 | 92 | | |
87 | 93 | | |
| |||
353 | 359 | | |
354 | 360 | | |
355 | 361 | | |
356 | | - | |
| 362 | + | |
357 | 363 | | |
358 | 364 | | |
359 | 365 | | |
| |||
373 | 379 | | |
374 | 380 | | |
375 | 381 | | |
| 382 | + | |
376 | 383 | | |
377 | 384 | | |
378 | 385 | | |
379 | 386 | | |
| 387 | + | |
| 388 | + | |
| 389 | + | |
380 | 390 | | |
381 | 391 | | |
382 | 392 | | |
| |||
0 commit comments