Repository navigation
Migrate to TGB Format for cache entries - #525
Open
tinder-maxwellelliott wants to merge 3 commits into
Open
tinder-maxwellelliott wants to merge 3 commits into
tinder-maxwellelliott wants to merge 3 commits into
Conversation
Record the investigation into porting Tango's TGB format (columnar blobs, raw fixed-stride hash column, block-aligned bisecting compare) to the serve cache. Measured on 1M-entry synthetic hash files: JSON is 212 B per node and a columnar entry would be 66 B (3.2x, not Tango's 10x, because 97% of it is two SHA-256 columns); the compare drops from 748 ms to 22 ms, and the 6.3 s per-request JSON parse disappears entirely. Deferred until serve entries or compare latency become a problem; the note lists the thresholds, the cheaper interim steps (in-memory parsed-entry LRU, direct parse, allocation-free compare, compressed shared tier) and a column layout and compare algorithm fitted to bazel-diff's hash files. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Gs1vjHt61po6uHgkYuh68N
…ing, cache parsed entries, gzip the shared tier The four interim steps from docs/tgb-binary-cache.md, none of which changes the hash-file format: - model: `HashFileData` deserializes with a serde visitor straight into the final map instead of building a `serde_json::Value` and re-inserting every label. Hashes are a new `Digest` type that keeps a SHA-256 as 32 raw bytes (one table-driven hex pass; anything that is not 64 lowercase hex digits stays text and round-trips unchanged). `TargetHash::same_as` replaces the two formatted `identity()` strings per node, and `filter_and_sort_labels` ranks each label once instead of cloning it inside the sort. - serve: parsed entries are kept in an in-memory LRU bounded by `--cacheMemoryMax` (default 512MB of serialized size, 0 disables), so a request whose revisions were already parsed skips the disk read and the JSON parse. A memory hit still refreshes the on-disk entry's mtime, which `touch` now sets directly instead of rewriting the whole file. `/metrics` reports the tier under `cache.memory`. - serve: entries are published to S3 or Nexus as fast-level gzip under `<key>.json.gz`, 2-3x smaller. A read tries that name first and falls back to the plain `<key>.json` an older release wrote; a read failure stays a one-request miss. The local-disk tier is unchanged. Measured on a synthetic 1M-entry pair, one `get-impacted-targets` (two parses plus the compare) goes from 4.8s and 1.46GB peak RSS to 2.5s and 0.96GB; the compare alone from 0.75s to 0.30s. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Gs1vjHt61po6uHgkYuh68N
Resolves the src/server.rs conflicts with Tinder#524, which replaced the State literal with State::new. The in-memory tier is now constructed there, and the remote-cache tests keep their compressed and plain fallback responses on top of the shared ready_state helper. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Gs1vjHt61po6uHgkYuh68N
This branch has not been deployed
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
2 participants
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.
In alignment with Tango approach