Skip to content

Speed up sessions/bridge-resilience e2e test to fix Windows timeout#267

Merged
jancurn merged 2 commits into
mainfrom
claude/quirky-ramanujan-qb8qzf
Jun 8, 2026
Merged

Speed up sessions/bridge-resilience e2e test to fix Windows timeout#267
jancurn merged 2 commits into
mainfrom
claude/quirky-ramanujan-qb8qzf

Conversation

@jancurn

@jancurn jancurn commented Jun 8, 2026

Copy link
Copy Markdown
Member

Fixes the sessions/bridge-resilience e2e test timing out on Windows (exit 137). It ran nearly every check through run_xmcpc, which spawns 5 mcpc processes each (the command plus 4 --json/--verbose invariant variants) — ~90 spawns total, making this the slowest e2e test at ~180s, right at the per-test watchdog and over it on Windows.

  • Use single-spawn run_mcpc for the resilience checks (this test's actual purpose)
  • Keep two run_xmcpc calls — one success, one error path — so live-bridge output invariants stay covered (also covered by basic/output-invariants)
  • Spawns ~90 → ~30; runtime ~180s → ~30s locally, well under the watchdog on Windows; all 10 cases pass

https://claude.ai/code/session_01TFGKjwkFNC1oupR2jEg2C8

claude added 2 commits June 8, 2026 17:17
This test ran nearly every check through run_xmcpc, which spawns 5 mcpc
processes (the command plus 4 --json/--verbose invariant-check variants).
With ~17 such calls on live-bridge session commands it did ~90 process
spawns and took ~180s on Linux — right at the per-test watchdog limit, and
over it on Windows, where the slower spawns made test 10 time out (exit 137).

Switch the checks to run_mcpc (one spawn each). This test is about bridge
resilience, not output-format invariants, which are already covered by
basic/output-invariants (the invariant machinery) and basic/errors /
basic/auth-errors (the --json error path). Runtime drops from ~180s to ~19s
with no loss of resilience coverage.

https://claude.ai/code/session_01TFGKjwkFNC1oupR2jEg2C8
The previous commit converted every check to run_mcpc, which dropped this
suite's output-invariant coverage on live-bridge MCP commands. Restore two
run_xmcpc calls — one success path (tools-list) and one error path
(tools-call on an unknown tool) — so the --json/--verbose invariants stay
covered on a real bridge, while the rest stays on single-spawn run_mcpc.

Still fast: ~30 process spawns (was ~90), ~30s locally and well under the
180s per-test watchdog on Windows. All 10 cases pass.

https://claude.ai/code/session_01TFGKjwkFNC1oupR2jEg2C8
@jancurn jancurn changed the title Optimize bridge-resilience test performance with selective run_xmcpc Speed up sessions/bridge-resilience e2e test to fix Windows timeout Jun 8, 2026
@jancurn jancurn merged commit 0931516 into main Jun 8, 2026
6 checks passed
@jancurn jancurn deleted the claude/quirky-ramanujan-qb8qzf branch June 8, 2026 20:44
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants