Filed while implementing #13688 (its A2.2 instruction: sweep the package for other
live-DDL tests still on vitest's inherited 5000ms default, widen the fix only within
the one target file, file a finding for the rest). Unassigned: recording only.
Two it(...) blocks in sql-driver-13056-orphan-shadow-column-cleanup.test.ts drove
2-3 full new SqlDriver(...) -> initObjects(...) -> disconnect() cycles against
live MySQL with no explicit timeout, so they inherited vitest's 5000ms default. That
budget was never chosen — four sibling files in the same package already carry
explicit per-test budgets (60_000 / 120_000 / 40_000). The timeout (not an
assertion failure, no MySQL error in the logs) reddened unrelated PRs, e.g. #13685
(a plugin-auth-only diff) and ejected #13687/#13761 from the merge queue (#13767).
The sweep
A script walked every it(...) nested inside a declareDialectCell(...) callback
in packages/drivers/driver-sql/src/*.test.ts and checked whether the third argument
is an explicit numeric timeout. Same defect shape, still on the default, in 9 more
files, 37 more it(...) blocks:
| File |
Cells |
it() blocks w/o budget |
sql-driver-11627-hash-shadow-key.test.ts |
MySQL, PG |
9 |
sql-driver-11565-row-byte-budget.test.ts |
MySQL |
6 |
sql-driver-keyed-text-mysql.test.ts |
MySQL |
5 |
sql-driver-12998-shadow-null-safe-key.test.ts |
MySQL |
4 |
sql-driver-13015-shadow-carried-index-drift.test.ts |
MySQL |
4 |
sql-driver-11794-richtext-text-family.test.ts |
MySQL, PG |
3 |
sql-driver-string-maxlength-varchar.test.ts |
MySQL, PG |
3 |
sql-driver-backend-fault-envelope.test.ts |
PG |
2 |
sql-driver-11389-date-tz-skew.test.ts |
MySQL, PG |
1 |
Most of these (all but sql-driver-backend-fault-envelope.test.ts) follow the exact
#13688 shape: each it(...) itself constructs a fresh new SqlDriver(...), calls
initObjects(...) (schema-sync DDL), and reads information_schema back — one or
more full connect cycles per test, inline in the it() body. Spot-checked
sql-driver-11627-hash-shadow-key.test.ts:135 as representative; the pattern repeats
across the other 7.
sql-driver-backend-fault-envelope.test.ts's two it() blocks are lower-risk: they
reuse a single driver connected once in a beforeAll and each issue 1-2 queries
against an already-open connection, not a fresh connect/DDL/disconnect cycle. Listed
for completeness since they are still on the inherited default, but the cost profile
is not the same as the other 8 files.
What this is and is not
This is a structural sweep (does an explicit budget exist), not a confirmed list
of CI failures — I have not run these against live MySQL/PG here (no live containers
in this environment; see #13688's own A2.1 for the same constraint). The evidence
that the shape is a real defect, not merely a theoretical one, is #13688 itself:
timeout-not-assertion, no DB error, and the repo's own convention (4 sibling files
budgeted) treating "no timeout" as an oversight rather than a choice.
Recommendation
Per #13688's dispatch order (STOP condition 3): "A2.2 finds the same shape across
many files in the package -> report before sweeping; a package-wide re-budget is a
different, larger decision" -- so #13688 fixed only its own two tests and this issue
records the rest for a deliberate decision, not a silent drive-by. The repo's existing
convention is a per-test budget (never a package-wide testTimeout, which would
silently re-budget the 150+ tests that legitimately want the fast default) -- same
shape as #13688's fix, applied file by file.
Related
Filed while implementing #13688 (its A2.2 instruction: sweep the package for other
live-DDL tests still on vitest's inherited 5000ms default, widen the fix only within
the one target file, file a finding for the rest). Unassigned: recording only.
What #13688 fixed
Two
it(...)blocks insql-driver-13056-orphan-shadow-column-cleanup.test.tsdrove2-3 full
new SqlDriver(...)->initObjects(...)->disconnect()cycles againstlive MySQL with no explicit timeout, so they inherited vitest's 5000ms default. That
budget was never chosen — four sibling files in the same package already carry
explicit per-test budgets (
60_000/120_000/40_000). The timeout (not anassertion failure, no MySQL error in the logs) reddened unrelated PRs, e.g. #13685
(a
plugin-auth-only diff) and ejected #13687/#13761 from the merge queue (#13767).The sweep
A script walked every
it(...)nested inside adeclareDialectCell(...)callbackin
packages/drivers/driver-sql/src/*.test.tsand checked whether the third argumentis an explicit numeric timeout. Same defect shape, still on the default, in 9 more
files, 37 more
it(...)blocks:it()blocks w/o budgetsql-driver-11627-hash-shadow-key.test.tssql-driver-11565-row-byte-budget.test.tssql-driver-keyed-text-mysql.test.tssql-driver-12998-shadow-null-safe-key.test.tssql-driver-13015-shadow-carried-index-drift.test.tssql-driver-11794-richtext-text-family.test.tssql-driver-string-maxlength-varchar.test.tssql-driver-backend-fault-envelope.test.tssql-driver-11389-date-tz-skew.test.tsMost of these (all but
sql-driver-backend-fault-envelope.test.ts) follow the exact#13688shape: eachit(...)itself constructs a freshnew SqlDriver(...), callsinitObjects(...)(schema-sync DDL), and readsinformation_schemaback — one ormore full connect cycles per test, inline in the
it()body. Spot-checkedsql-driver-11627-hash-shadow-key.test.ts:135as representative; the pattern repeatsacross the other 7.
sql-driver-backend-fault-envelope.test.ts's twoit()blocks are lower-risk: theyreuse a single driver connected once in a
beforeAlland each issue 1-2 queriesagainst an already-open connection, not a fresh connect/DDL/disconnect cycle. Listed
for completeness since they are still on the inherited default, but the cost profile
is not the same as the other 8 files.
What this is and is not
This is a structural sweep (does an explicit budget exist), not a confirmed list
of CI failures — I have not run these against live MySQL/PG here (no live containers
in this environment; see #13688's own A2.1 for the same constraint). The evidence
that the shape is a real defect, not merely a theoretical one, is #13688 itself:
timeout-not-assertion, no DB error, and the repo's own convention (4 sibling files
budgeted) treating "no timeout" as an oversight rather than a choice.
Recommendation
Per #13688's dispatch order (STOP condition 3): "A2.2 finds the same shape across
many files in the package -> report before sweeping; a package-wide re-budget is a
different, larger decision" -- so #13688 fixed only its own two tests and this issue
records the rest for a deliberate decision, not a silent drive-by. The repo's existing
convention is a per-test budget (never a package-wide
testTimeout, which wouldsilently re-budget the 150+ tests that legitimately want the fast default) -- same
shape as #13688's fix, applied file by file.
Related
sql-driver-13056-orphan-shadow-column-cleanup.test.tsspecifically; driver-sql (live MySQL): the two #13056 orphan-shadow-cleanup tests run 3–4 connect→DDL→disconnect round trips under vitest's DEFAULT 5000ms, so
Temporal Conformancereddens unrelated PRs #13688 addresses its root cause.AssertionErrorvs timeout heuristic has a counterexample class — the discriminator is what the assertion MEASURES, not what kind of failure it is #13830 -- merge-queue triage guidance on the timeout-vs-assertion heuristic thiswhole class turns on.