You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
test(integration): make the llm-log plan and aging tests chunk-boundary proof (Thursday 00:00 UTC hypertable chunk) - #138
Fixes#137. Test-only: three integration tests fail deterministically after the Thursday 00:00 UTC chunk boundary of the context_llm_log hypertable (7-day chunks), because the newest chunk is almost empty then. Root 52a35d0b, #134 and #136 are red in the heavy and rest shards for exactly this reason; #135 is not involved (its head was green at 23:51 UTC with identical test files).
What changes
TestDistillSpendGuard/WindowAndBackoffExpiryLetItRunAgain (internal/events/distill_spend_integration_test.go): aging the llm log by UPDATE created_at moved rows across the chunk boundary, which TimescaleDB rejects (constraint_1, SQLSTATE 23514). Rows are now deleted and re-inserted 3 h earlier in one transaction via a temp table. The rollup trigger counts them twice on the way; the guard reads the raw log (distill_spend.go:313), so the checked invariant is unchanged.
TestMigration158OnPopulatedHypertable (internal/llmlog/consumer_attribution_integration_test.go): seed and window are aligned with the chunk grid via time_bucket('7 days', now(), origin => 'epoch') (without origin the bucket starts on Monday, not Thursday — checked against the live DB). The test asserts that the newest chunk ends exactly there. Scan counting is unchanged; no enable_seqscan.
:411 … and 1 seq scan(s), want idx_llm_log_consumer only 0/3
3/3 (5 index scans on idx_llm_log_consumer each)
Quota
6 findings per run :365 a raw scan reads "_hyper_1_6_chunk_context_llm_log_created_at_idx" 0/3
3/3; rollup numbers identical red and green (rows 3/5/5, ceiling 6, 1441 tenant rows, 10 lookups)
Package suites: events 263 pass / 0 fail / 2 skip, llmlog 44/0/4, backends 102/0/0. golangci-lint with and without the integration tag 0 issues, go vet -tags=integration ./... rc 0, both deadcode gates ok. Not run: the repo-wide go test -short -race ./... (no non-test file changed; CI covers it).
Expanded integration coverage for quota spending and consumer attribution using time-based data aligned to storage intervals.
Updated distiller log test data handling to support shifting records across storage intervals.
Added checks for recent data ranges and chunk boundaries in quota and consumer attribution scenarios. These changes are limited to test coverage and do not describe changes to end-user functionality.
…ry proof (Thursday 00:00 UTC hypertable chunk)
context_llm_log is a TimescaleDB hypertable with 7-day chunks laid on the
Unix-epoch grid, so every chunk starts on a Thursday 00:00 UTC. Right after
that boundary the newest chunk holds only the hours since midnight. Three
integration tests seeded relative to now() and broke deterministically in
that window (local 3/3 each on 52a35d0, CI root 41944930718 and PRs #134,
#136 identical):
1. TestDistillSpendGuard/WindowAndBackoffExpiryLetItRunAgain aged the log
with UPDATE created_at = created_at - 3 h. A row moved across a chunk
boundary is refused by the chunk's range CHECK, not re-routed:
"violates check constraint constraint_1 (SQLSTATE 23514)", 00:00-03:00
UTC every Thursday, latent since the test landed.
Fix: dsAgePipeline copies the pipeline's rows into a temp table, deletes
them and re-inserts them aged, in one transaction, so every row lands in
the chunk of its new created_at. The guard reads the raw log, not the
rollup, so the window it sees is the one the UPDATE meant to produce;
the asserted story (trip, both clocks run out, next tick is ok with
budget 1) is unchanged.
2. TestMigration158OnPopulatedHypertable/per-consumer window: the seed ended
at now(), so the newest chunk in the 30-day window was one heap page
(22 rows) and a Seq Scan is the cheaper plan there: "... and 1 seq
scan(s), want idx_llm_log_consumer only".
Fix: seed, consumer rows and window end at anchor, the start of the
chunk holding now() (time_bucket over the hypertable's own chunk
interval with origin => epoch; the default origin 2000-01-03 is a
Monday, three days off the chunk grid). The test asserts that the
newest chunk ends exactly at anchor. Every chunk the window touches now
carries the full seed density at every clock; the gate is unchanged:
one index scan on idx_llm_log_consumer per chunk, no seq scan, no
enable_seqscan, no loosened count.
3. TestQuotaSpendPlan: the relief lookups take their created_at bound at
run time, so every chunk is planned; on the Thursday-morning chunk (four
rows per key) the planner reads the chunk's created_at index - correct
for a chunk of a few pages, meaningless for the access path at scale:
"a raw scan reads _hyper_1_6_chunk_context_llm_log_created_at_idx, not
idx_llm_log_apikey" (6 of 6 plans).
Fix: the accountant runs on a test clock (SetClockForTest, existing
seam) one to two hours before anchor, keeping the wall clock's minute
of the hour so the edges keep every width (#129); the seed and the
30-day tenant count end at that clock. The newest chunk holds almost
seven days, the oldest about three, at every wall clock. All checks stay
hard and unchanged: two-sided created_at bound, raw rows <= ceiling 6,
rollup via idx_llm_usage_key, exact created_at MCV statistics on every
chunk, idx_llm_log_apikey on every raw scan of every chunk. No rows lie
after the clock's next full hour, so the rollup is the same seed moved in
time: the run logs the same figures as the red run (raw rows 3/5/5,
ceiling 6, 1441 tenant rows in 30 days, 10 executed relief lookups).
Red -> green, -count=3 each, Thursday 2026-11-05 01:42-01:50 UTC:
TestDistillSpendGuard 0/3 -> 3/3 (red: SQLSTATE 23514)
TestMigration158OnPopulatedHypertable 0/3 -> 3/3 (red: 1 seq scan)
TestQuotaSpendPlan 0/3 -> 3/3 (red: 6 created_at_idx plans)
Negative probe: the fixed quota test with its clock moved to Thursday
01:xx (anchor + 1 h) is red again with the same 6 findings; the plans show
the chunk-6 created_at_idx scans are the relief-lookup subplans.
Packages with -tags=integration -p 1 -count=1 (top-level pass/fail/skip):
events 263/0/2, llmlog 44/0/4, backends 102/0/0.
golangci-lint (with and without the integration tag) 0 issues,
go vet -tags=integration ./... rc 0, both deadcode gates ok.
Test files only; no production code, migration or config.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
No actionable comments were generated in the recent review.
Walkthrough
The integration tests now align seeded data and query windows with hypertable chunk boundaries. The distiller test ages rows by deleting and reinserting them through a temporary table in a transaction.
Changes: Chunk-aligned integration tests
Layer / File(s)
Summary
Quota plan seed and clock alignment go/internal/backends/quota_plan_integration_test.go
The test seeds rows and evaluates the 30-day tenant-row count using a test clock aligned with the current chunk anchor. It checks that the newest chunk ends at the anchor.
The test replaces a direct timestamp update with a transactional temporary-table workflow that deletes and reinserts the selected pipeline’s rows at shifted timestamps.
Consumer attribution seed and window alignment go/internal/llmlog/consumer_attribution_integration_test.go
The test can seed through a supplied timestamp, derives the current chunk start, and uses that anchor for seeded rows and the 30-day query window. It checks that the newest chunk ends at the anchor.
GottZ/ctx#74 — Adds the test clock seam used by TestQuotaSpendPlan.
GottZ/ctx#135 — Updates the same test’s planner assertions; this change also adjusts its chunk layout and test clock.
GottZ/ctx#73 — Adds the rollup trigger relevant to reinserting context_llm_log rows in the distiller test.
Merge Risk: Minimal · up to 87a70
The tests use chunk-aligned timestamps and transactional row aging. No merge-blocking risk was identified.
Pre-merge checks: 5
Check name
Status
Explanation
Description Check
Check skipped - CodeRabbit’s high-level summary is enabled.
Title check
The title clearly identifies integration-test changes that address hypertable chunk-boundary failures. It accurately describes the main purpose, although it does not name every affected test.
Linked Issues check
Issue #137 requires test-only fixes for three boundary failures. The diff changes only the three named integration tests. TestDistillSpendGuard now ages rows by transactional delete and re-insert. `…
Out of Scope Changes check
The changed paths are go/internal/events/distill_spend_integration_test.go, go/internal/llmlog/consumer_attribution_integration_test.go, and go/internal/backends/quota_plan_integration_test.go. …
Docstring Coverage
Docstring coverage is 80.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff with cyclomatic complexity of at least 2. Documentati…
Full details: Linked Issues check
Explanation
Issue #137 requires test-only fixes for three boundary failures. The diff changes only the three named integration tests. TestDistillSpendGuard now ages rows by transactional delete and re-insert. TestMigration158OnPopulatedHypertable aligns seed data and the query window to an epoch-origin chunk boundary and checks the newest chunk end. Its per-chunk index and no-sequential-scan gate remains. TestQuotaSpendPlan uses SetClockForTest with a boundary-safe test clock, retains the wall-clock minute, and keeps the row-ceiling, MCV, and hard index checks. The issue's minimum-size condition is a proposed fix; the boundary-safe clock is a valid alternative that preserves the required checks.
Full details: Out of Scope Changes check
Explanation
The changed paths are go/internal/events/distill_spend_integration_test.go, go/internal/llmlog/consumer_attribution_integration_test.go, and go/internal/backends/quota_plan_integration_test.go. Each change supports one of the three failures named in issue #137. No production code or unrelated behavior changed.
Full details: Docstring Coverage
Explanation
Docstring coverage is 80.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff with cyclomatic complexity of at least 2. Documentation-only edits and deletions within surviving functions are included; unchanged diff context and fully deleted functions are excluded. Analyzed 5 functions across 3 files.
Finishing Touches
Generate docstrings
Commit to this branch
Create a new PR
Generate unit tests (beta)
Commit to this branch
Create a new PR
Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Recent review info
Run configuration
Source mode: Auto
Repository: Unavailable
Branch: Unavailable
File: No configuration file from this repository applied
Commit: Unavailable
Layers: defaults
Review profile: CHILL
Plan: Advanced
Run ID: 77532d78-90e0-466f-9a67-1d2f3b081164
Commits
Reviewing files that changed from the base of the PR and between 52a35d0 and 87a7094.
Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 8 remain after this review.
The reason will be displayed to describe this comment to others. Learn more.
🟢 Approval recommended
No active review issues remain for this test-only change addressing chunk-boundary sensitivity in three integration tests.
No open issues 🎉
What changed in this PR
This test-only PR addresses #137 by making three integration tests independent of the Thursday 00:00 UTC TimescaleDB chunk boundary. Production code and existing plan assertions remain unchanged.
Changes:
Ages distill log rows through transactional delete and reinsert, avoiding cross-chunk update failures.
Aligns migration-test seed data and query windows with the hypertable chunk grid.
Uses a chunk-relative quota test clock while preserving wall-clock minute variation and strict index checks.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #137. Test-only: three integration tests fail deterministically after the Thursday 00:00 UTC chunk boundary of the
context_llm_loghypertable (7-day chunks), because the newest chunk is almost empty then. Root52a35d0b, #134 and #136 are red in theheavyandrestshards for exactly this reason; #135 is not involved (its head was green at 23:51 UTC with identical test files).What changes
TestDistillSpendGuard/WindowAndBackoffExpiryLetItRunAgain(internal/events/distill_spend_integration_test.go): aging the llm log byUPDATE created_atmoved rows across the chunk boundary, which TimescaleDB rejects (constraint_1, SQLSTATE 23514). Rows are now deleted and re-inserted 3 h earlier in one transaction via a temp table. The rollup trigger counts them twice on the way; the guard reads the raw log (distill_spend.go:313), so the checked invariant is unchanged.TestMigration158OnPopulatedHypertable(internal/llmlog/consumer_attribution_integration_test.go): seed and window are aligned with the chunk grid viatime_bucket('7 days', now(), origin => 'epoch')(withoutoriginthe bucket starts on Monday, not Thursday — checked against the live DB). The test asserts that the newest chunk ends exactly there. Scan counting is unchanged; noenable_seqscan.TestQuotaSpendPlan(internal/backends/quota_plan_integration_test.go): the test now runs on a test clock 1–2 h before the chunk boundary through the existingSetClockForTestseam, with the minute still taken from the wall clock (so test(backends): TestQuotaSpendPlan is planner-dependent in CI — custom plan touches 102 raw rows via the created_at chunk index instead of idx_llm_log_apikey (2 of 3 rest-shard runs on Go-untouched PRs) #129 stays covered). Every assertion stays hard, including the index check on every chunk. Rejected alternatives: a page-count threshold for the index assertion (the Thursday chunk already has ~400 rows at 01:40 UTC, so the threshold would have been a guess that weakens the assertion), and filling the current chunk with seed rows (writes into the future and inflates the rollup).Verification (
-tags=integration -p 1, inside the red window 01:42–01:50 UTC)52a35d0b:252 … violates check constraint "constraint_1" (SQLSTATE 23514)0/3:411 … and 1 seq scan(s), want idx_llm_log_consumer only0/3idx_llm_log_consumereach):365 a raw scan reads "_hyper_1_6_chunk_context_llm_log_created_at_idx"0/3Package suites: events 263 pass / 0 fail / 2 skip, llmlog 44/0/4, backends 102/0/0.
golangci-lintwith and without the integration tag 0 issues,go vet -tags=integration ./...rc 0, both deadcode gates ok. Not run: the repo-widego test -short -race ./...(no non-test file changed; CI covers it).Report:
.project/plan-openai-api-2026-11-03/reports/bau/fix-ci-thursday-chunk-boundary.md.Summary by CodeRabbit