mirror of
https://github.com/jmcorgan/fips.git
synced 2026-10-05 19:18:25 +00:00
The job builds and smoke-installs inside a FreeBSD VM on a Linux runner and normally finishes in eight to eleven minutes. Five consecutive runs instead sat in the VM step for 70, 190, 360, 360 and 360 minutes and ended cancelled, the last three at GitHub's default six-hour job ceiling. The 70 and 190 minute runs are not explained by that ceiling. The job set no timeout-minutes, so nothing of ours bounded them. Thirty minutes leaves a slow run room to finish and turns a stall into a half-hour failure rather than a six-hour one. This does not diagnose the stall; it stops it costing a runner day. The boundary is sharp and the same on both trunks: every run carrying the native datagram API commits has stalled, and every run before them passed. That is a correlation with a candidate mechanism and not a cause, and it is recorded in the tracker rather than asserted here.