feat: add --track-allocators toggle for memory mode

[not-matthias]

What

Adds a --track-allocators flag (default on, env CODSPEED_TRACK_ALLOCATORS) on run/exec and the memtrack track subcommand. When disabled (--track-allocators false), memtrack skips the per-allocation uprobe machinery and only emits coarse mmap/munmap/brk events.

Why

Allocation-heavy programs (e.g. a Rust build) generate an overwhelming number of malloc/free events, and the per-allocation uprobes slow the target significantly. This flag lets users trade allocation granularity for lower overhead while still collecting RSS-relevant memory events.

How

• memtrack: the mmap/munmap/brk syscall tracepoints (already compiled into the BPF program but never attached) are now always attached in every memory run. The flag gates only the expensive allocator uprobes (exec-mapping watcher + attach worker).
• Tracker::new(track_allocators: bool) conditionally starts the exec watcher / AttachWorker; spawn/finish tolerate a None worker.
• runner: flag lives on ExecAndRunSharedArgs, threaded through OrchestratorConfig → ExecutorConfig (mirroring exclude_allocations), and passed to the memtrack subprocess as an explicit --track-allocators <bool> arg (not via env inheritance, to survive glibc secure-execution env stripping).

Default runs are unchanged in allocation coverage and now additionally emit mmap/brk events.

Testing

• cargo check --workspace --tests, cargo test --release -p codspeed-runner — green.
• CLI surface verified: run/exec/memtrack track --help show the flag (default true); CODSPEED_TRACK_ALLOCATORS=false parses.
• New CI-gated integration test (test_track_allocators_disabled_skips_allocations) asserts that with allocators disabled, the trace contains Mmap events and zero allocation-kind events.

Fixes COD-3231.

[greptile-apps]

Greptile Summary

Adds configurable allocator tracking alongside expanded RSS accounting.

• Adds --track-allocators support and conditionally disables allocator uprobes while retaining coarse memory events.
• Adds eBPF RSS, reverse-mapping, and process-lifecycle event collection.
• Extends shared memtrack event schemas and reconstruction logic for the new event types.
• Adds architecture-matrix CI coverage and RSS integration fixtures and snapshots.

Confidence Score: 5/5

The PR appears safe to merge because no blocking failures remain within the follow-up review scope.

No blocking failures remain.

Important Files Changed

Filename	Overview
crates/memtrack/src/ebpf/tracker.rs	Makes allocator attachment optional while retaining the common tracker lifecycle and tracepoint setup.
crates/memtrack/src/ebpf/c/rss.bpf.h	Introduces RSS, reverse-mapping, fork, exec, and exit event collection with process ownership tracking.
crates/memtrack/src/ebpf/memtrack/tracking.rs	Attaches syscall, RSS, lifecycle, and supported reverse-mapping probes.
crates/memtrack/src/main.rs	Exposes the allocator-tracking toggle and passes it into tracker construction.
crates/runner-shared/src/artifacts/memtrack/mod.rs	Extends the shared memtrack artifact schema for RSS and process-lifecycle events.
.github/workflows/ci.yml	Expands eBPF integration coverage across x86-64 and ARM runners and adds RSS tests.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
    CLI[run / exec / memtrack CLI] --> CFG[track_allocators configuration]
    CFG --> TRACKER[Tracker]
    TRACKER --> TRACE[Always-attached syscall and lifecycle tracepoints]
    CFG -->|true| UPROBES[Allocator watcher and uprobes]
    CFG -->|false| SKIP[Skip allocator uprobes]
    TRACE --> EVENTS[Memtrack events]
    UPROBES --> EVENTS
    EVENTS --> ARTIFACT[runner-shared artifact encoding]

Loading

_{Reviews (3): Last reviewed commit: "feat: add --track-allocators toggle for ..." | Re-trigger Greptile}

[codspeed-hq]

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.

✅ 17 untouched benchmarks

---

_{Comparing cod-3231-support-disabling-allocator-tracking (131f8a3) with main (26fb4b5)}