diff --git a/.github/workflows/arch-qemu-nightly.yml b/.github/workflows/arch-qemu-nightly.yml index 638147fc..df5c22ff 100644 --- a/.github/workflows/arch-qemu-nightly.yml +++ b/.github/workflows/arch-qemu-nightly.yml @@ -46,9 +46,18 @@ jobs: go-version: stable - run: sudo apt-get update && sudo apt-get install -y qemu-user-static - name: Test under qemu (${{ matrix.goarch }}) - # qemu-user emulation is ~50-100x slower than native; the heaviest lane - # (s390x, the slowest target) can exceed Go's default 10m test-binary - # timeout on the full vm suite. Give the emulated lanes generous headroom - # so a slow-but-correct run isn't killed as a false failure. (On real - # s390x silicon the same suite runs natively in seconds.) - run: GOARCH=${{ matrix.goarch }} go test -timeout=30m -exec=${{ matrix.qemu }} ./... + # qemu-user emulation is ~50-100x slower than native, so the wall here is + # not a hang detector: it is the only thing that decides whether a + # slow-but-correct run counts. It has to sit well clear of what the run + # actually takes, and 30m did not. Measured on internal/vm, the heaviest + # package, in run 32661057000 (all four green): + # + # loong64 990s riscv64 1204s + # ppc64le 1284s s390x 1754s <- 46s under a 1800s wall + # + # The very next run of the same tree died at 1801.594s with + # "panic: test timed out after 30m0s" and no failing test — a 2.6% + # margin, which ordinary runner variance crosses. 60m puts s390x at + # about half the wall, which is what "generous headroom" was meant to + # say. (On real s390x silicon the same suite runs natively in seconds.) + run: GOARCH=${{ matrix.goarch }} go test -timeout=60m -exec=${{ matrix.qemu }} ./...