feat(memtrack): pause producers under ring pressure

[not-matthias]

Adds opt-in, best-effort backpressure for memtrack's BPF ring buffers.

When a ring crosses its watermark, BPF latches the episode and stops the winning tracked producer. A userspace coordinator then stops the owned process tree using pidfds, drains the event, stack, mapping, and attach pipelines, clears the latch, and resumes the tree. Finite timeouts terminate the tracked tree and fail the capture; 0 remains the default and preserves drop detection, while inf waits until recovery. Detected ring loss still fails the capture.

The change also:

• lowers ring polling latency from 10 ms to 1 ms
• keeps attach/startup resume ownership coordinated with pressure pauses
• propagates the timeout through the runner's experimental memory-mode flag
• avoids x86 per-CPU private BPF stack corruption during nested uprobes by keeping hash scratch in the unpublished ring record and reducing BPF stack use

This PR is stacked on #522.

Verification

• cargo fmt --all -- --check
• cargo clippy --release -p memtrack -p codspeed-runner -- -D warnings
• Runtime tests are deferred to CI.

[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.

✅ 33 untouched benchmarks
⏩ 4 skipped benchmarks¹

---

_{Comparing feat/memtrack-pause-worker (4953b6c) with cod-3222-add-ebpf-based-dwarffp-unwinding (2152fe2)}

Footnotes

1. 4 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[greptile-apps]

Confidence Score: 4/5

The PR is not yet safe to merge because a pressure-release race can leave a tracked producer permanently stopped.

Findings

1.  Pressure Stop Can Be Lost ▶

Fix with agent prompt

### Issue 1
crates/memtrack/src/ebpf/c/utils/pressure.bpf.h:34-37
After the BPF code records a pressure hold, the independent poller can observe an empty ring, delete that record, and send `SIGCONT` before `bpf_send_signal(SIGSTOP)` runs. The later `SIGSTOP` then suspends the producer with no remaining record that can resume it, causing the tracked workload and capture to hang.

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

Summary

This PR adds stack-ring backpressure to memtrack, coordinates pressure and attach stop ownership, bounds synchronous draining by a producer-position snapshot, moves stack-hash scratch into unpublished ring records, and adds high-pressure integration fixtures.

• Stops tracked stack-producing processes when the stack ring crosses its watermark.
• Coordinates pressure and attach holds through shared BPF maps.
• Reduces poll latency and validates stack-ring loss as capture loss.
• Adds multithreaded and multiprocess pressure tests.
• Preserves teardown work when waiting for the tracked command fails.

Diagram

sequenceDiagram
    participant P as Tracked producer
    participant B as BPF stack probe
    participant R as Stack ring
    participant U as Userspace poller
    participant S as Stop coordinator

    P->>B: Allocation probe
    B->>R: Submit stack record
    B->>S: Record pressure hold
    B-->>P: Request SIGSTOP
    U->>R: Drain snapshot
    U->>S: Release pressure holds
    S-->>P: SIGCONT when no attach hold remains

Loading

_{Reviews (5) · Last reviewed commit: "fixup! feat(memtrack): pause producers u..."}

[not-matthias]

@greptileai

[greptile-apps]

Pressure Stop Can Be Lost

After the BPF code records a pressure hold, the independent poller can observe an empty ring, delete that record, and send SIGCONT before bpf_send_signal(SIGSTOP) runs. The later SIGSTOP then suspends the producer with no remaining record that can resume it, causing the tracked workload and capture to hang.

Knowledge Base Used: eBPF memory tracker

Prompt To Fix With AI

This is a comment left during a code review.
Path: crates/memtrack/src/ebpf/c/utils/pressure.bpf.h
Line: 34-37

Comment:
**Pressure Stop Can Be Lost**

After the BPF code records a pressure hold, the independent poller can observe an empty ring, delete that record, and send `SIGCONT` before `bpf_send_signal(SIGSTOP)` runs. The later `SIGSTOP` then suspends the producer with no remaining record that can resume it, causing the tracked workload and capture to hang.

**Knowledge Base Used:** [eBPF memory tracker](https://app.greptile.com/codspeed/-/custom-context/knowledge-base/codspeedhq/codspeed/-/docs/ebpf-memory-tracker.md)

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.