ci: add memtrack walltime benchmarks to CI

[not-matthias]

Adds a memtrack-benchmarks CI job that runs codspeed-memtrack track against a few representative workloads (ls, dd, tar), each with RSS-only and RSS+rmap variants, to catch regressions in memtrack's own overhead.

Runs both walltime (execution time) and memory (memtrack's own peak RSS/allocations while tracking each workload) modes against the same crates/memtrack/codspeed.yml config.

[codspeed]

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
🆕 10 new benchmarks

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
🆕	WallTime	memtrack track tar	N/A	9.7 s	N/A
🆕	WallTime	memtrack track tar (with physical)	N/A	9.8 s	N/A
🆕	WallTime	memtrack track dd	N/A	1.3 s	N/A
🆕	WallTime	memtrack track dd (with physical)	N/A	1.5 s	N/A
🆕	WallTime	memtrack track ls	N/A	1.2 s	N/A
🆕	WallTime	memtrack track ls (with physical)	N/A	1.2 s	N/A
🆕	Memory	memtrack track ls	N/A	50.9 MB	N/A
🆕	Memory	memtrack track ls (with physical)	N/A	51 MB	N/A
🆕	Memory	memtrack track dd	N/A	51 MB	N/A
🆕	Memory	memtrack track dd (with physical)	N/A	51.1 MB	N/A

---

_{Comparing cod-3093-memtrack-data-format-improvements (2c36a70) with main (5af022c)}

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge, with the previously reported runner-noise and partial-batch coverage concerns addressed.

Summary

Adds CI coverage for memtrack overhead and improves its event-delivery pipeline.

• Runs walltime benchmarks on the dedicated codspeed-macro runner and memory benchmarks on GitHub-hosted runners.
• Splits the ls, dd, and tar workloads into independent matrix shards with RSS-only and physical-memory variants.
• Batches ring-buffer events while explicitly flushing partial batches on timeout, drain, and shutdown.
• Adds lifecycle tests covering each partial-batch flush path.
• Adds restore-only cache support so each runner architecture has one cache writer.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    M[CI matrix: mode × workload] --> W{Mode}
    W -->|walltime| R1[codspeed-macro]
    W -->|memory| R2[ubuntu-latest]
    R1 --> C[Install memtrack and runner]
    R2 --> C
    C --> S[Grant eBPF capabilities]
    S --> B[Run ls, dd, or tar config]
    B --> P[Ring-buffer poller]
    P --> Q[Batch events]
    Q -->|full batch| E[Encoding pipeline]
    Q -->|timeout, drain, or shutdown| F[Flush partial batch]
    F --> E

Loading

_{Reviews (3) · Last reviewed commit: "fixup! ci: let the runner action handle ..."}

[GuillaumeLagrange]

olgtm, good idea!

[GuillaumeLagrange]

Memtrack walltime benchmarks took 19m to run, maybe we can shard something before introducing a 20m delay in the CI pipeline? 😇