perf(ecstore): replace mmap+spawn_blocking shard reads with pread by rdiperri-wasabi · Pull Request #27 · wasabi/rustfs

rdiperri-wasabi · 2026-06-22T20:09:26Z

Fix 1 — consolidate spawn_blocking in read_all_data_with_dmtime: Previously each EC shard read dispatched two thread-pool round-trips (one for the pread/mmap setup, one for the actual read). This patch collapses both into a single blocking closure, eliminating one cross-thread handoff per shard per GET request.

Fix 2 — replace mmap with pread in batch_shard_pread: EC shard reads for object data were using mmap, which causes kernel page-cache pressure for large objects and adds VM fault overhead on each access. Replaced with O_DIRECT-friendly pread calls that read directly into caller-supplied buffers, removing the mmap lifecycle (map, access, unmap) from the hot GET path.

Measured on smc1 (single-node, SAS SSDs, EC:1) vs wasabi baseline at 50 threads, 5-minute warm-cache GET runs:

Small objects (0–1500K, loopback):
4d: 1,562 → 2,109 MB/s (+35%), 22 → 16ms
6d: 906 → 2,437 MB/s (+169%), 40 → 14ms
8d: 662 → 3,072 MB/s (+364%), 54 → 12ms
10d: 472 → 2,458 MB/s (+420%), 76 → 14ms

Large objects (1000–1500K, 100 GbE from smc2):
4d: 1,987 → 2,949 MB/s (+48%), 30 → 19ms
6d: 1,577 → 2,949 MB/s (+87%), 38 → 19ms
8d: 1,270 → 2,867 MB/s (+126%), 48 → 20ms
10d: 824 → 2,785 MB/s (+238%), 73 → 21ms

Also includes perf harness improvements: analyze.py adds per-interface NIC reporting; loadgen-run.sh and run-perf-test.sh add LG_OBJ_SIZE, remote loadgen support (LOADGEN_HOST), and related flags.

Related Issues

Summary of Changes

Verification

Impact

Additional Notes

Thank you for your contribution! Please ensure your PR follows the community standards (CODE_OF_CONDUCT.md). If this is your first contribution, review the CLA document and sign it by commenting I have read and agree to the CLA. on the PR.

Copilot

Pull request overview

This PR optimizes EC shard reads on the GET hot path by replacing mmap-based reads with pread-style reads and reducing spawn_blocking round-trips, while also extending the perf harness to support more workload mixes and richer reporting.

Changes:

Switch local “zero-copy” shard reads from mmap to FileExt::read_at and add batched shard reads (batch_shard_pread) to reduce per-shard scheduling overhead.
Add detailed GET-path tracing spans around fanout, locks, decode, and pipe stages.
Improve perf harness ergonomics (env overrides for op mix/object size, summary detection) and enhance analysis output (per-interface NIC, disk read stats).

Reviewed changes

Copilot reviewed 12 out of 13 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
scripts/perf/run-perf-test.sh	Makes loadgen completion detection op-agnostic; enables GET tracing recipe in TRACE mode.
scripts/perf/lib/loadgen-run.sh	Adds env-driven workload controls (op mix, object size, resume/save/preload flags).
scripts/perf/analyze/analyze.py	Generalizes loadgen parsing across op types; adds disk read metrics and updates report tables.
rustfs/src/app/object_usecase.rs	Adds a pre-dispatch GET tracing span around request-context preparation.
crates/ecstore/src/set_disk/read.rs	Introduces a batched local-shard read fast-path and instruments shard setup/fanout timing.
crates/ecstore/src/set_disk.rs	Adds GET-path spans around lock acquisition, xlmeta read, and spawned read task.
crates/ecstore/src/erasure_coding/decode.rs	Instruments shard read, EC decode, and pipe write stages for GET tracing.
crates/ecstore/src/disk/mod.rs	Adds a helper to obtain local filesystem paths only for local disks.
crates/ecstore/src/disk/disk_store.rs	Exposes local object-path retrieval on the local disk wrapper.
crates/ecstore/src/disk/local.rs	Replaces mmap with pread in `read_file_zero_copy`, consolidates `spawn_blocking`, adds `batch_shard_pread`, and adds tests.
crates/ecstore/src/bitrot.rs	Extracts shard offset/length adjustment into a helper and updates docs/tests.
crates/ecstore/Cargo.toml	Removes `memmap2` dependency from ecstore.
Cargo.lock	Drops `memmap2` from the lockfile for ecstore.

Copilot

Pull request overview

Copilot reviewed 12 out of 13 changed files in this pull request and generated 3 comments.

- set_disk/read.rs: wrap unix batch fast-path in #[cfg(unix)] and use early return so non-unix targets (Windows) compile without batch_shard_pread and adjust_shard_read_params, which are unix-only. Sequential fallback is now unconditional. - disk/local.rs: move should_reclaim_file_cache_after_read() call inside the #[cfg(target_os = "linux")] fadvise block, eliminating the unused- variable warning on macOS/non-linux unix targets. - bitrot.rs: tighten adjust_shard_read_params visibility from pub to pub(crate); it is only used within this crate. - analyze.py: fix empty-device placeholder row from 9 columns to 8 to match the disk table header, preventing zip-truncation in _md_table(). Co-authored-by: Cursor <cursoragent@cursor.com>

Fix 1 — consolidate spawn_blocking in read_all_data_with_dmtime: Previously each EC shard read dispatched two thread-pool round-trips (one for the pread/mmap setup, one for the actual read). This patch collapses both into a single blocking closure, eliminating one cross-thread handoff per shard per GET request. Fix 2 — replace mmap with pread in batch_shard_pread: EC shard reads for object data were using mmap, which causes kernel page-cache pressure for large objects and adds VM fault overhead on each access. Replaced with O_DIRECT-friendly pread calls that read directly into caller-supplied buffers, removing the mmap lifecycle (map, access, unmap) from the hot GET path. Measured on smc1 (single-node, SAS SSDs, EC:1) vs wasabi baseline at 50 threads, 5-minute warm-cache GET runs: Small objects (0–1500K, loopback): 4d: 1,562 → 2,109 MB/s (+35%), 22 → 16ms 6d: 906 → 2,437 MB/s (+169%), 40 → 14ms 8d: 662 → 3,072 MB/s (+364%), 54 → 12ms 10d: 472 → 2,458 MB/s (+420%), 76 → 14ms Large objects (1000–1500K, 100 GbE from smc2): 4d: 1,987 → 2,949 MB/s (+48%), 30 → 19ms 6d: 1,577 → 2,949 MB/s (+87%), 38 → 19ms 8d: 1,270 → 2,867 MB/s (+126%), 48 → 20ms 10d: 824 → 2,785 MB/s (+238%), 73 → 21ms Also includes perf harness improvements: analyze.py adds per-interface NIC reporting; loadgen-run.sh and run-perf-test.sh add LG_OBJ_SIZE, remote loadgen support (LOADGEN_HOST), and related flags. Co-authored-by: Cursor <cursoragent@cursor.com>

- set_disk/read.rs: wrap unix batch fast-path in #[cfg(unix)] and use early return so non-unix targets (Windows) compile without batch_shard_pread and adjust_shard_read_params, which are unix-only. Sequential fallback is now unconditional. - disk/local.rs: move should_reclaim_file_cache_after_read() call inside the #[cfg(target_os = "linux")] fadvise block, eliminating the unused- variable warning on macOS/non-linux unix targets. - bitrot.rs: tighten adjust_shard_read_params visibility from pub to pub(crate); it is only used within this crate. - analyze.py: fix empty-device placeholder row from 9 columns to 8 to match the disk table header, preventing zip-truncation in _md_table(). Co-authored-by: Cursor <cursoragent@cursor.com>

github-actions · 2026-06-26T01:57:38Z

Dependency Review

✅ No vulnerabilities or license issues or OpenSSF Scorecard issues found.

Scanned Files

None

Copilot

Pull request overview

Copilot reviewed 11 out of 12 changed files in this pull request and generated 1 comment.

Co-authored-by: Cursor <cursoragent@cursor.com>

Copilot AI review requested due to automatic review settings June 22, 2026 20:09

Copilot started reviewing on behalf of rdiperri-wasabi June 22, 2026 20:10 View session

Copilot AI reviewed Jun 22, 2026

View reviewed changes

Comment thread scripts/perf/analyze/analyze.py

Comment thread crates/ecstore/src/disk/local.rs

Comment thread crates/ecstore/src/disk/local.rs

Comment thread crates/ecstore/src/set_disk/read.rs Outdated

rdiperri-wasabi force-pushed the perf/productionize-get-path-fixes branch 2 times, most recently from b3e8d49 to e9b8f40 Compare June 22, 2026 20:37

Copilot AI review requested due to automatic review settings June 22, 2026 20:37

Copilot started reviewing on behalf of rdiperri-wasabi June 22, 2026 20:38 View session

Copilot AI reviewed Jun 22, 2026

View reviewed changes

Comment thread crates/ecstore/src/set_disk/read.rs Outdated

Comment thread scripts/perf/analyze/analyze.py

Comment thread crates/ecstore/src/bitrot.rs

rdiperri-wasabi force-pushed the perf/productionize-get-path-fixes branch from 3e5c5f4 to e4b135b Compare June 22, 2026 21:42

rdiperri-wasabi enabled auto-merge June 22, 2026 21:43

rdiperri-wasabi closed this Jun 22, 2026

auto-merge was automatically disabled June 22, 2026 22:53
Pull request was closed

rdiperri-wasabi reopened this Jun 22, 2026

rdiperri-wasabi and others added 2 commits June 25, 2026 18:57

rdiperri-wasabi force-pushed the perf/productionize-get-path-fixes branch from e4b135b to b044968 Compare June 26, 2026 01:57

Copilot AI review requested due to automatic review settings June 26, 2026 01:57

Copilot started reviewing on behalf of rdiperri-wasabi June 26, 2026 01:57 View session

Copilot AI reviewed Jun 26, 2026

View reviewed changes

Comment thread crates/ecstore/src/set_disk.rs

style(ecstore): fix rustfmt line-length violation in set_disk/read.rs

e6ed46c

Co-authored-by: Cursor <cursoragent@cursor.com>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

perf(ecstore): replace mmap+spawn_blocking shard reads with pread#27

perf(ecstore): replace mmap+spawn_blocking shard reads with pread#27
rdiperri-wasabi wants to merge 3 commits into
mainfrom
perf/productionize-get-path-fixes

rdiperri-wasabi commented Jun 22, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

github-actions Bot commented Jun 26, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

rdiperri-wasabi commented Jun 22, 2026

Related Issues

Summary of Changes

Verification

Impact

Additional Notes

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

github-actions Bot commented Jun 26, 2026

Dependency Review

Scanned Files

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants