Repository navigation
Conversation
StoreBlock clears the SQL response cache only after its commit, so a "not found" cached before the commit can be served, for the moment between commit and reset, for a block that is already stored. Block validation sends a child whose parent reads as missing to catchup, so even that short window is not worth the cache. Only a found result is cached now; a not-found lookup is one indexed query.
Contributor
|
🤖 Claude Code Review Status: Complete Current Review: The change is small and correct. Non-blocking tradeoff: every lookup of an unknown hash (e.g. peer announcements of new blocks) now queries the DB instead of hitting the cache. The query is a single indexed lookup on |
|
Contributor
Benchmark Comparison ReportBaseline: Current: Summary
All benchmark results (sec/op)
Threshold: >10% with p < 0.05 | Generated: 2026-10-26 14:45 UTC |
oskarszoon
approved these changes
Oct 26, 2026
ordishs
approved these changes
Oct 27, 2026
freemans13
approved these changes
Oct 27, 2026
freemans13
left a comment
freemans13
left a comment
Collaborator
There was a problem hiding this comment.
Approved. No actionable findings at 118e316.
Removing negative caching closes the window between a block's commit and the response-cache reset. Positive caching retains its generation checks, and the blockchain clients and block-validation callers preserve the same boolean/error contract.
Validation:
go test -race -tags testtxmetacache -count=2 ./stores/blockchain/...passed.- Ran
TestSQLGetBlockExists_NeverCachesNotFoundagainst the pre-fix implementation through a Go overlay. It failed on the cached-false assertion in both repetitions, confirming that the test detects the removed behavior.
Repeated lookups of the same absent hash now perform an indexed database query. This is the intended tradeoff; I did not benchmark its cost.
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
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.
Problem
SQL.GetBlockExistscached a not-found result in the response cache.StoreBlockresets that cache only after its commit, so a "not found" cached before the commit can still be served, in the window between commit and reset, for a block that is already stored. The generation counter stops aSet(false)that races the reset, but not a read of afalsecached earlier.Block validation decides from
GetBlockExists(parent)whether a child goes to catchup (queueFetchedBlock), and a needless catchup is expensive: on the scaling env on 2026-10-26 one meant moving to CATCHINGBLOCKS and re-downloading a 274 GB block. That incident was a different race (#1997 covers it, #2053 limits the damage). This closes the store-side window for the same decision.Change
Only a found result is cached. A not-found lookup is one indexed query on
blocks.hash.BlockValidation.blockExistsCachealready caches only positives, so both layers now agree.Cost
Every caller looks up a block per block or per announcement (block validation parent checks, catchup entry checks, legacy netsync). The asset handler only reaches it from the admin invalidate/revalidate routes. Lookups of distinct unknown hashes already missed the cache. Only repeated lookups of the same absent hash now go to the database.
Tests
TestSQLGetBlockExists_NeverCachesNotFound: a not-found result leaves no cache entry, and a found result is still cached. It fails with the oldSet(false).go test -race ./stores/blockchain/...passes, andmake lintreports 0 issues.Follow-ups (not in this PR)
GetBlockHeaderscaches an empty result for an unknown start hash, inchainWalkCache(10 min TTL), which a plain extend does not clear.GetBlockHeaderIDsalready guards against this.LocateBlockHeadersdiscards theGetBlockExistsbool and always takeslocator[0].