feat: speedup ebpf uprobe attaching by not-matthias · Pull Request #440 · CodSpeedHQ/codspeed

not-matthias · 2026-07-07T10:15:10Z

No description provided.

greptile-apps · 2026-07-07T10:17:52Z

Greptile Summary

This PR speeds up memtrack eBPF allocator attachment. The main changes are:

Adds offset resolution for allocator symbols before attaching probes.
Falls back to symbol-name attachment when a defined symbol has no file offset.
Resolves allocator libraries in parallel with Rayon.
Adds a CodSpeed benchmark for serial versus parallel offset resolution.
Extends CI benchmark coverage to include memtrack.

Confidence Score: 5/5

This looks safe to merge.

No blocking issues found in the changed code.

Important Files Changed

Filename	Overview
crates/memtrack/src/ebpf/memtrack.rs	Adds pre-resolved symbol offsets, by-name fallback for no-offset symbols, and parallel allocator-library resolution.
crates/memtrack/src/ebpf/tracker.rs	Delegates bulk allocator attachment to the new `MemtrackBpf::attach_allocators` flow.
crates/memtrack/benches/attach.rs	Adds benchmarks comparing serial and parallel allocator symbol offset resolution.
crates/memtrack/Cargo.toml	Adds Rayon and registers the new attach benchmark.
.github/workflows/ci.yml	Installs libbpf build dependencies and includes memtrack in benchmark build and run steps.

_{Reviews (3): Last reviewed commit: "ci(memtrack): run offset resolution benc..." | Re-trigger Greptile}

codspeed-hq · 2026-07-07T10:18:56Z

Merging this PR will not alter performance

⚠️

Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

✅ 7 untouched benchmarks
🆕 2 new benchmarks

Performance Changes

	Mode	Benchmark	`BASE`	`HEAD`	Efficiency
🆕	Simulation	`resolve_offsets_parallel`	N/A	10.5 ms	N/A
🆕	Simulation	`resolve_offsets_serial`	N/A	10 ms	N/A

_{Comparing cod-3070-speedup-allocator-attaching-on-nixos (9f32601) with main (eac7766)}

…f func_name Resolve every defined symbol's libbpf-compatible file offset once per library (st_value - sh_addr + sh_offset) and attach uprobes by offset, replacing the per-probe func_name lookup that reparsed the ELF each time.

Symbol resolution parses each allocator ELF independently, so fan it out across cores with rayon before the serial attach step. Cuts offset resolution for all discovered allocators from ~190ms to ~29ms locally.

Divan bench comparing serial vs parallel symbol offset resolution over the discovered allocator libraries. Runs without sudo since it only parses ELFs, no eBPF attach.

GuillaumeLagrange

small changes, olgtm

GuillaumeLagrange · 2026-07-10T12:40:17Z

        paste! {
-            fn [<try_ $name>](&mut self, lib_path: &Path, symbol: &str) {
-                let result = self.$name(lib_path, symbol);
+            fn [<$name _by_name>](&mut self, lib_path: &Path, symbol: &str) -> Result<()> {


since we are resolving symbol offsets, do we still need the by_name variant?

GuillaumeLagrange · 2026-07-10T12:40:35Z

        paste! {
-            fn [<try_ $name>](&mut self, lib_path: &Path, symbol: &str) {
-                let result = self.$name(lib_path, symbol);
+            fn [<$name _by_name>](&mut self, lib_path: &Path, symbol: &str) -> Result<()> {


same remark here

GuillaumeLagrange · 2026-07-10T12:43:01Z

+                let result = if let Some(offset) = symbols.offset(symbol) {
+                    self.$name(lib_path, offset)
+                } else if symbols.is_defined_without_offset(symbol) {
+                    self.[<$name _by_name>](lib_path, symbol)
+                } else {
+                    return;
+                };


Basically my above comment is: since we are resolving symbols ahead of time, is it possible that we failed to resolve the symbol earlier on but still manage to attach by name here?

If that's possible, it warrants a small explanation of why it can happen

else just remove the function

GuillaumeLagrange · 2026-07-10T12:45:16Z

+                Ok(())
+            }
+
+            fn [<try_ $name>](


in rust, try_xxx infers a Result. It seems we have them mixed up here

Seems like the attach functions should be named try_attach_xxx, and this function should be named something else, like attach_if_found or sthing

GuillaumeLagrange · 2026-07-10T12:47:09Z

@@ -0,0 +1,36 @@
+use memtrack::{AllocatorLib, resolve_symbol_offsets};


shouldnt these run on vendored allocator libs as part of a testdata folder instead of relying on the host machine's allocator?

GuillaumeLagrange · 2026-07-10T12:49:57Z

          tool: cargo-codspeed

+      - name: Install libbpf build deps
+        run: sudo apt-get update && sudo apt-get install -y build-essential pkgconf zlib1g-dev libbpf-dev autopoint bison flex


there is quite a lot of duplication to install build time dependencies of memtrack across the whole workflow, cnat we create a install-bpf-deps local action?

Dont bundle biuld-essential with it, first check if it's required here, and if it's required here, add it as either a separate step, or an additional-packages action input

greptile-apps Bot reviewed Jul 7, 2026

View reviewed changes

Comment thread crates/memtrack/src/ebpf/memtrack.rs Outdated

Comment thread crates/memtrack/src/ebpf/memtrack.rs Outdated

Comment thread crates/memtrack/src/ebpf/memtrack.rs Outdated

not-matthias force-pushed the cod-3070-speedup-allocator-attaching-on-nixos branch from 0e8688a to 3a322ae Compare July 7, 2026 17:07

greptile-apps Bot reviewed Jul 7, 2026

View reviewed changes

Comment thread crates/memtrack/src/ebpf/memtrack.rs Outdated

not-matthias marked this pull request as ready for review July 7, 2026 17:47

not-matthias added 4 commits July 8, 2026 19:13

perf(memtrack): resolve allocator symbol offsets in parallel

a56a0dc

Symbol resolution parses each allocator ELF independently, so fan it out across cores with rayon before the serial attach step. Cuts offset resolution for all discovered allocators from ~190ms to ~29ms locally.

bench(memtrack): add offset resolution benchmark

9a995ca

Divan bench comparing serial vs parallel symbol offset resolution over the discovered allocator libraries. Runs without sudo since it only parses ELFs, no eBPF attach.

ci(memtrack): run offset resolution benchmarks

9f32601

not-matthias force-pushed the cod-3070-speedup-allocator-attaching-on-nixos branch from 3a322ae to 9f32601 Compare July 8, 2026 17:18

not-matthias requested a review from GuillaumeLagrange July 8, 2026 17:21

GuillaumeLagrange requested changes Jul 10, 2026

View reviewed changes

		@@ -0,0 +1,36 @@
		use memtrack::{AllocatorLib, resolve_symbol_offsets};

Uh oh!

Conversation

not-matthias commented Jul 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

greptile-apps Bot commented Jul 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 5/5

Important Files Changed

Uh oh!

Uh oh!

Uh oh!

Uh oh!

codspeed-hq Bot commented Jul 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will not alter performance

Performance Changes

Uh oh!

Uh oh!

GuillaumeLagrange left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

not-matthias commented Jul 7, 2026 •

edited

Loading

greptile-apps Bot commented Jul 7, 2026 •

edited

Loading

codspeed-hq Bot commented Jul 7, 2026 •

edited

Loading