Fix descending fetch gap-range bounds and tail-mark#155
Closed
afrind wants to merge 3 commits into
Closed
Conversation
Summary: Add a portable LocationIntervalSet class that stores disjoint intervals of AbsoluteLocation. This is an alternative to quic::IntervalSet that works with AbsoluteLocation (two uint64_t values) without requiring uint128_t, making it portable to platforms without __int128 support (MSVC, 32-bit). Key features: - insert(start, end): Add an interval with automatic merging of overlapping/adjacent - contains(loc): O(log n) check if a location is within any interval - findIntervalEnd(loc): Find the end of the interval containing a location - next()/prev(): Static helpers for location successor/predecessor This will be used to efficiently track known gaps in MoQCache, enabling O(log n) skipping over large gaps (e.g., from group 1 to group 2^60). Reviewed By: sandarsh Differential Revision: D91961632
Summary: Refactors MoQCache to use LocationIntervalSet for tracking known-nonexistent object/group ranges (gaps) instead of storing individual OBJECT_NOT_EXIST and GROUP_NOT_EXIST cache entries. Key changes: - Add `gaps` field (LocationIntervalSet) to CacheTrack for O(1) gap lookups - Process Prior Object ID Gap (0x3E) and Prior Group ID Gap (0x3C) extensions to populate the gaps set via processGapExtensions() - getCachedObjectMaybe() now returns optional<CacheEntry*>: - nullopt = cache miss (need upstream fetch) - nullptr = known nonexistent (in gaps set) - valid ptr = cache hit - Fetch loop skips gaps in O(1) using findInterval() and skipPastGap() - Simplified endOfFetch() logic: call when !(lastObject && servedOneObject) - Added helper functions: getGapRanges(), isGroupNonExistent(), skipPastGap() This enables efficient handling of large gaps (e.g., group 1 to 2^60) without O(n) iteration or memory overhead. Differential Revision: D91961635
Summary: Three bugs in the descending-fetch path of FetchWriteback / getGapRanges, found by review of the LocationIntervalSet conversion: 1. getGapRanges DESC range #1 extended to {startGroup, MAX} regardless of the user's fetch end, marking positions above fetchEnd as nonexistent. Likewise range #3 started at {endGroup, 0} regardless of fetchStart. Either could shadow positions outside the requested range and suppress later upstream fetches. Fixed by adding fetchStart/fetchEnd parameters and clamping ranges #1 and #3 against them when start.group / end.group is the fetch boundary. 2. cacheImpl streaming objectPayload finFetch path called markNonExistentTo(end_) without first advancing the iterator past the just-completed object, so the gap range overlapped it. Fixed by stepping fetchRangeIt_.next() before tail-marking. 3. Both finFetch tail-marks called markNonExistentTo(end_), where end_ is the user's highest endpoint. In DESC iteration, the iteration end is the lowest position, so the wrong endpoint was used. Fixed by switching to fetchRangeIt_.end(), which returns the order-aware iteration end. Adds three regression tests covering each bug. Differential Revision: D103210059
meta-cla
Bot
added
the
CLA Signed
Contributor
|
@afrind has exported this pull request. If you are a Meta employee, you can view the originating Diff in D103210059. |
Contributor
Author
|
This was merged as part of another commit |
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:
Three bugs in the descending-fetch path of FetchWriteback / getGapRanges, found by review of the LocationIntervalSet conversion:
getGapRanges DESC range Re-sync with internal repository #1 extended to {startGroup, MAX} regardless of the user's fetch end, marking positions above fetchEnd as nonexistent. Likewise range Convert directory fbcode/opensource to use the Ruff Formatter #3 started at {endGroup, 0} regardless of fetchStart. Either could shadow positions outside the requested range and suppress later upstream fetches. Fixed by adding fetchStart/fetchEnd parameters and clamping ranges Re-sync with internal repository #1 and Convert directory fbcode/opensource to use the Ruff Formatter #3 against them when start.group / end.group is the fetch boundary.
cacheImpl streaming objectPayload finFetch path called markNonExistentTo(end_) without first advancing the iterator past the just-completed object, so the gap range overlapped it. Fixed by stepping fetchRangeIt_.next() before tail-marking.
Both finFetch tail-marks called markNonExistentTo(end_), where end_ is the user's highest endpoint. In DESC iteration, the iteration end is the lowest position, so the wrong endpoint was used. Fixed by switching to fetchRangeIt_.end(), which returns the order-aware iteration end.
Adds three regression tests covering each bug.
Differential Revision: D103210059