fix(bench): warm every app before judging, and relay the daemon's output live

[proggeramlug]

Two harness defects, both found while reproducing perry#8546 and both of which cost real conclusions.

1. The warm phase stopped at the first non-200

warm_every_app asserted per app in index order, so a run reported bench-000: 500 and nothing about the rest. The question under #8546 is which apps fail, and the answer turned out to be: every application except the one whose init finished last, regardless of dispatch order (serial 3 apps: 000 and 001 fail, 002 serves; concurrent: whichever finished init last serves). That rule was invisible while the harness stopped at the first failure.

Now every app is warmed, each status and a body excerpt are printed, and the assertion carries the full failure list.

2. The daemon's output was discarded on the failure paths that need it

• stderr went into a bounded sink that only the ready-timeout branches printed. A daemon that dies mid-request — the exact outcome PERRY_GC_PROTECT_FROMSPACE=1 produces on a hit — surfaced as reqwest::Error(IncompleteMessage) with the fault report thrown away.
• stdout was read only until HTTP listener ready. After that the deployment dispatch failed … handler promise rejected: <reason> line — the one fix(worker): never report an empty handler rejection reason #14 exists to make useful — was never read, and a chatty daemon would eventually block on a full pipe.

COOP_BENCH_TRACE_DAEMON=1 relays both streams live; the stdout reader keeps draining regardless. Default behaviour is unchanged.

Verified

With this in place, on Perry 0.5.1516, the from-space fault report from the pre-#15 worker printed in full (obj_type=5, backtrace in invoke_buffer_handler), and post-#15 the per-app table above was produced in four configurations. Results are on perry#8546.

https://claude.ai/code/session_01UZJbhb2FTuakurTHPAKQgd

Summary by CodeRabbit

• Enhancements
  • Added optional live daemon output streaming during benchmark runs for easier diagnostics.
  • Improved startup failure reporting with retained error output.
  • Benchmark warm-up now runs across all applications instead of stopping after the first failure.
  • Warm-up results include request statuses and response excerpts, with failures reported together for easier review.

[coderabbitai]

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Free

Run ID: 2e880b5e-c148-40ba-af86-a27cdc5334c8

📥 Commits

Reviewing files that changed from the base of the PR and between 3e7180e and 00a80e4.

📒 Files selected for processing (1)

• crates/coop-daemon/tests/resource_benchmark.rs

Included review availability: Your plan provides up to 1 included review per hour; 0 remain after this review.

---

📝 Walkthrough

Walkthrough

The resource benchmark can stream daemon output when tracing is enabled. It continues draining daemon output after readiness. Warm-up requests now run for every app and report all failures together.

Changes

Resource benchmark

Layer / File(s)	Summary
Daemon output handling crates/coop-daemon/tests/resource_benchmark.rs	When COOP_BENCH_TRACE_DAEMON is set, the benchmark forwards daemon stderr and stdout live. It retains bounded stderr capture and drains stdout after readiness.
Warm-up failure reporting crates/coop-daemon/tests/resource_benchmark.rs	The benchmark logs each warm-up response, collects status and expected-body failures for every app, and reports the complete failure list after all requests finish.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk: ⚪ Minimal · up to 00a80

This PR improves benchmark warm-up reporting and preserves live daemon diagnostics without changing product behavior; no actionable merge-blocking risk remains beyond normal checks and review.

---

Note

🎁 Summarized by CodeRabbit Free

Your organization is on the Free plan. CodeRabbit will generate a high-level summary and a walkthrough for each pull request. For a comprehensive line-by-line review, please upgrade your subscription to CodeRabbit Pro by visiting https://app.coderabbit.ai/settings/billing.

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