Reconnect-to-resumed exceeds the 900 ms D130 gate at 300 ms RTT #61

Open
opened 2026-10-07 08:05:50 +00:00 by kayg · 5 comments
Owner

Finding from the D130 latency run on branch latency-gate at base cecc7ecd7b. Reproduce with TRACER_TEST_PORT=4621 bun x playwright test test/terminal.spec.ts; the tracer production build ran through the stream latency proxy at configured RTT 300 ms and jitter ±30 ms. The proxy self-test measured p50 299.05 ms and p99 301.81 ms over 5 samples.

METRIC name=reconnect_resumed source=tracer-production-build n=1 p50_ms=1038.70 p99_ms=1038.70 budget_ms=900 result=FAIL

Budget is RTT × 2 + 300 ms. The probe measures forced socket drop until the first resumed marker paints. Product changes are out of scope for #55; this issue tracks investigation and a regression test.

Finding from the D130 latency run on branch latency-gate at base cecc7ecd7bb452229fcf292bebde0917c0568522. Reproduce with TRACER_TEST_PORT=4621 bun x playwright test test/terminal.spec.ts; the tracer production build ran through the stream latency proxy at configured RTT 300 ms and jitter ±30 ms. The proxy self-test measured p50 299.05 ms and p99 301.81 ms over 5 samples. METRIC name=reconnect_resumed source=tracer-production-build n=1 p50_ms=1038.70 p99_ms=1038.70 budget_ms=900 result=FAIL Budget is RTT × 2 + 300 ms. The probe measures forced socket drop until the first resumed marker paints. Product changes are out of scope for #55; this issue tracks investigation and a regression test.
Author
Owner

Confirmation run on the corrected per-flow jitter proxy: reconnect→first resumed marker painted p99 1041.40 ms, still above 900 ms. This was the same production tracer probe at 300 ms configured RTT.

Confirmation run on the corrected per-flow jitter proxy: reconnect→first resumed marker painted p99 1041.40 ms, still above 900 ms. This was the same production tracer probe at 300 ms configured RTT.
Author
Owner

Latest run using the finalized 200 ms burst-jitter relay measured reconnect→first resumed marker at p99 996.70 ms, still above 900 ms. Machine load was 7.00/7.52/7.43 on 8 cores.

Latest run using the finalized 200 ms burst-jitter relay measured reconnect→first resumed marker at p99 996.70 ms, still above 900 ms. Machine load was 7.00/7.52/7.43 on 8 cores.
Author
Owner

Updated #55 production tracer measurement on infraternal-dev (x86_64, 8 cores; tracer-browser load 14.31/16.38/15.76), configured RTT 300 ms ±30 ms: reconnect-to-resumed n=1 p50/p99 995.20 ms versus the 900 ms budget, FAIL. Measured by scripts/verify latency with TRACER_TEST_PORT=4621; metric is emitted by apps/tracer/test/terminal.spec.ts. Run log: latency-profile.log in the latency-gate scratch directory.

Updated #55 production tracer measurement on infraternal-dev (x86_64, 8 cores; tracer-browser load 14.31/16.38/15.76), configured RTT 300 ms ±30 ms: reconnect-to-resumed n=1 p50/p99 995.20 ms versus the 900 ms budget, FAIL. Measured by scripts/verify latency with TRACER_TEST_PORT=4621; metric is emitted by apps/tracer/test/terminal.spec.ts. Run log: latency-profile.log in the latency-gate scratch directory.
Author
Owner

Latest production-build profile on infraternal-dev (x86_64, 8 cores; tracer-browser phase load 17.73 10.48 10.96): reconnect→resumed n=1, p50/p99=1012.70 ms, budget 900 ms, FAIL (112.70 ms over). See #55 for the full run.

Latest production-build profile on infraternal-dev (x86_64, 8 cores; tracer-browser phase load 17.73 10.48 10.96): reconnect→resumed n=1, p50/p99=1012.70 ms, budget 900 ms, FAIL (112.70 ms over). See #55 for the full run.
Author
Owner

Latest production reconnect measurement on infraternal-dev (x86_64, 8 cores; tracer-browser load 8.84/9.29/7.19), 300 ms RTT ±30 ms: METRIC name=reconnect_resumed source=tracer-production-build n=20 p50_ms=808.00 p99_ms=1009.00 budget_ms=900 result=FAIL. The 900 ms ceiling is RTT × 2 + 300 ms; the p99 misses it by 109 ms. Full profile log was $HOME/.cache/codex-scratch/latency-gate/latency-final.log.

Latest production reconnect measurement on infraternal-dev (x86_64, 8 cores; tracer-browser load 8.84/9.29/7.19), 300 ms RTT ±30 ms: `METRIC name=reconnect_resumed source=tracer-production-build n=20 p50_ms=808.00 p99_ms=1009.00 budget_ms=900 result=FAIL`. The 900 ms ceiling is RTT × 2 + 300 ms; the p99 misses it by 109 ms. Full profile log was `$HOME/.cache/codex-scratch/latency-gate/latency-final.log`.
Sign in to join this conversation.
No labels
launch-qa
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
kayg/infraternal#61
No description provided.