fix(bench): drop cgroup page cache between certify subprocesses (#904)

[DecisionNerd]

Problem

Fly S18 re-qualification with canonical ordered-LIMIT queries completed every GraphForge certification phase (status: passed, max per-phase anonymous RSS 555 MiB), yet BenchExec terminated the run with OUT OF MEMORY at cgroup memory 4000067584 B after ~3.5 h.

Root cause: BenchExec's 4 GiB cgroup memlimit counts file-backed page cache from sequential gf subprocess invocations. Each phase stays bounded in anonymous RSS, but cache from multi-TB logical reads accumulates across the full S18 lifecycle (ingest → query → export → reopen_proof).

Fix

After each public-command child completes in graphforge-benchmark-certify, call sync and write 3 to /proc/sys/vm/drop_caches on Linux to release page-cache pressure before the next subprocess starts.

Evidence

Downloaded from Fly machine 89122db6197348 run #2 (commit c640c49):

• graphforge-certify-evidence.json: all 10 phases passed, max peak_rss_bytes = 555175936
• BenchExec XML: terminationreason=memory, memory=4000067584B, returnvalue=0

Testing

• cargo test -p graphforge-benchmark-certify (16 passed)
• cargo clippy -p graphforge-benchmark-certify -- -D warnings

Closes #904 (partial — Fly S18 re-run required after merge to confirm full lifecycle under 4 GiB).

  

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

Note

Drop cgroup page cache between certify subprocesses in execute_process

• Adds a Linux-only release_cgroup_page_cache helper that runs sync and writes "3" to /proc/sys/vm/drop_caches, best-effort with all errors ignored; non-Linux gets a no-op stub.
• Calls the helper on every return path in execute_process in lib.rs after the child process exits, so page cache is evicted between subprocess runs.
• Behavioral Change: benchmarks now evict page cache after each subprocess, which may reduce cross-run cache interference and slow down individual runs that previously benefited from warm cache.

^{Macroscope summarized 720a140.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Essentials

Run ID: 04c69a23-9f23-485a-b384-d75ed8f143f9

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

Warning

Billing warning: we have not been able to collect payment for this subscription for more than 72 hours. Please update the payment method or pay any pending invoices in Billing to avoid service interruption.

---

_{Comment @coderabbitai help to get the list of available commands.}