feat: add ring-buffer scaffold for memory profiling#428
Open
rcoh wants to merge 2 commits into
Open
Conversation
Adds `pub(crate) mod memory_profiling` (gated on the new `memory-profiling` cargo feature) with: - RawAlloc<MAX_FRAMES> and RawFree fixed-size POD records. - RingBuffers<MAX_FRAMES> wrapping two `crossbeam_queue::ArrayQueue`s — alloc queue (default 4096 slots × 1056 B = ~4 MiB at 128 frames) and free queue (default 32K slots × 32 B = ~1 MiB; 8× the alloc slot count per design §9 because frees are pushed unconditionally and allocs only when sampled). - MemoryProfileSource<MAX_FRAMES> implementing the existing Source trait, draining both queues each flush cycle and emitting AllocEvent / FreeEvent into the trace via the consolidator's ThreadLocalEncoder. Drain is timestamp-ordered: at each step we emit whichever queue head has the older timestamp. Naive 'all allocs, then all frees' would corrupt the liveset under address reuse within a single flush cycle (alloc → free → alloc-with-same-addr); see the `address_reuse_within_flush_cycle_preserves_liveset` test. Liveset tracking (HashMap<u64, LivesetEntry>) is wired in but only activates when MemoryProfileSource is constructed with `MemoryProfileSource::new(rings, true)`. No allocator hook yet (later commit). Tests push synthetic RawAllocs and RawFrees directly into the queues and verify AllocEvents and FreeEvents come out the other end.
- Add compile-time assertion that MAX_FRAMES <= u8::MAX (frame_count field) - Replace frames().to_vec() with direct slice of destructured array to avoid a heap allocation per drain in handle_alloc
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.
Adds
pub(crate) mod memory_profiling(gated on the newmemory-profilingcargo feature) with:crossbeam_queue::ArrayQueues — alloc queue (default 4096 slots × 1056 B = ~4 MiB at 128 frames) and free queue (default 32K slots × 32 B = ~1 MiB; 8× the alloc slot count per design §9 because frees are pushed unconditionally and allocs only when sampled).Drain is timestamp-ordered: at each step we emit whichever queue head has the older timestamp. Naive 'all allocs, then all frees' would corrupt the liveset under address reuse within a single flush cycle (alloc → free → alloc-with-same-addr); see the
address_reuse_within_flush_cycle_preserves_livesettest.Liveset tracking (HashMap<u64, LivesetEntry>) is wired in but only activates when MemoryProfileSource is constructed with
MemoryProfileSource::new(rings, true).No allocator hook yet (later commit). Tests push synthetic RawAllocs and RawFrees directly into the queues and verify AllocEvents and FreeEvents come out the other end.