fix: bound raw MCP query worker lifecycle - #256
Closed
lzehrung wants to merge 3 commits into
Closed
Conversation
Impact streaming's internal chunk queue buffered every produced item in an unbounded array whenever the consumer was not actively reading, and the background analyzeImpact() producer kept running to completion even after a consumer abandoned the stream (broke out of iteration, or the generator was otherwise returned early), retaining index snapshots, sets, and closures for work nobody would ever read. Add an internal AbortController to analyzeImpactStreaming(); the generator's finally block aborts it on every exit path, including the async-generator return protocol triggered by early consumer cancellation. The onImpactItem producer callback checks the signal and throws once it fires, unwinding analyzeImpact's in-progress work so no further symbol batches or transitive passes start. This needs no public API change: yield* delegation already forwards a caller's early return into the inner generator, so session.ts's analyzeImpactStream benefits without any code change there. Cap the internal queue at a bounded number of unread chunks (DEFAULT_MAX_IMPACT_STREAM_QUEUED_CHUNKS, overridable via the internal ImpactStreamingContext.maxQueuedChunks test seam). True backpressure would require making the onImpactItem emission callback awaitable at every synchronous call site in direct.ts/transitive.ts, which is outside this module. On overflow the stream surfaces an explicit ImpactStreamOverflowError as a terminal error chunk instead of silently dropping items or completing as if nothing were missed.
Raw query_sqlite reads ran a synchronous DatabaseSync iteration with row and byte caps but no time budget or cancellation: a non-recursive but expensive statement (a large join, ORDER BY random() with no matching index, ...) could hold the host event loop for as long as SQLite took to produce a row, and a client disconnect did not stop it. Run the query in a dedicated Piscina worker thread (same pattern as the existing query-index worker pool) with a hard deadline. On expiry the worker thread is force-terminated and the call rejects with SqliteQueryDeadlineExceededError immediately; the host event loop is never blocked regardless of how long the underlying query actually runs, and a subsequent query against the same file succeeds right away since concurrent read-only SQLite connections do not block each other. Pool teardown is fire-and-forget on deadline expiry rather than awaited, so the caller is never delayed by an orphaned worker thread still finishing a single already-in-flight synchronous native call (worker termination cannot preempt one in-progress call the same way it can prevent further JS from running) -- documented in rawQueryWorkerPool.ts, verified directly against a 200M-row recursive CTE. If the compiled worker asset cannot be located (a corrupted or partial install), the query falls back to running in-process under a per-row elapsed-time budget instead of refusing outright. That fallback is strictly weaker and is documented as such: the budget is only checked between rows the native iterator has already produced, so a statement that is slow to produce its very first row is not bounded by it. Preserves the existing row/byte cap contracts and normalizeSqliteRowLimit reuse in this file.
Owner
Author
|
Superseded by #264, which consolidates the MCP session, transport, SQLite, and streaming fixes. |
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.
Summary
Verification