Repository navigation
Fix/known gaps - #3
Merged
Merged
Conversation
Every document change appended a row to yjs_updates and nothing ever removed one, so replay cost and storage grew for the life of a board. Snapshots were already being written; they just were not being used to retire the history they contain. Storing a snapshot now prunes in the same transaction. The ordering is what makes it safe: read the update log's high-water mark BEFORE encoding, then delete only at or below that mark. Hocuspocus applies an update to the document before onChange persists it, so any row at or below the mark is provably inside the snapshot encoded afterwards, while rows appended during the encode land above it and survive. Replaying a redundant update is a no-op; losing a live one is not, so the bias is deliberate. Snapshots are capped too, keeping the newest few as a manual fallback. They are the larger leak of the two, each being the whole document rather than one delta. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
PostgresPersistence had no tests. These pin the ordering rule compaction depends on, including the case that matters most: an update appended while the snapshot is being encoded must survive. Writing them caught a real defect — snapshot retention was skipped whenever the update log happened to be empty, so an idle board still accumulated snapshots without bound. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The build step's comment claimed it enforced the budget. Nothing did — it ran `next build` and checked only that the build succeeded, so the board route could have crossed the limit silently. Next reports First Load JS gzipped (verified: a chunk reported as 54.2 kB measures 53.0 KB gzipped and 169 KB raw), so its own figures compare directly against the budget. The gate parses them rather than recomputing from .next/, which keeps the number CI enforces identical to the number the table shows. `shell: bash` on the build step is load-bearing: it brings `-eo pipefail`, so piping through tee cannot mask a failed build. Current state: 15 routes, worst is /b/[boardId] at 245.0 kB — 5 kB of headroom. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two entries were no longer true and one is now fixed: rate limiting shipped in 667b1ac, the e2e suite runs (18/18, repeatedly), and the Yjs update log is compacted as of this branch. A gaps list that understates the product is as misleading as one that overstates it. Replaced with the gaps that are actually open, including several surfaced by recent QA and DX passes but never written down: ws as a single point of failure, the shared rate-limit bucket when no proxy header is present, the absent Y.Doc migration path, and agent-protocol being unpublished with no version on the wire. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Caught by reading the live ws log against the row counts in Postgres: "prunedUpdates: 0, prunedSnapshots: 0" on every write, while the tables showed the log emptied and snapshots capped. The deletes were correct; the telemetry was lying. postgres-js returns an Array SUBCLASS whose real figure is on `.count` and whose `.length` is 0 without a RETURNING clause, so the Array.isArray branch matched first and reported zero every time. Checking count/rowCount before the array fallback fixes it — now the same run logs prunedUpdates: 1 alongside a table that agrees. The unit tests missed this because the in-memory fake returns plain values and never modelled the driver's result shape, so the regression test pins the shape itself. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
What changed and why
How it was verified
Checklist
pnpm typecheck && pnpm lint && pnpm testall passpnpm test:e2epasses, or this change cannot affect itpackages/sharedorlib/consensushas a unit testNeeds a maintainer's sign-off before merge
packages/agent-protocol)doc-guard