runtime: TestMemoryLimitNoGCPercent failures

[gopherbot]

#!watchflakes
default <- pkg == "runtime" && test == "TestMemoryLimitNoGCPercent"

Issue created automatically to collect these failures.

Example (log):

=== RUN   TestMemoryLimitNoGCPercent
    gc_test.go:789: /Users/swarming/.swarming/w/ir/x/t/go-build568260939/testprog.exe GCMemoryLimitNoGCPercent: exit status 2
    gc_test.go:792: expected "OK\n", but got "retained=285218120 limit=268435456 bound=285212672\npanic: exceeded memory limit by more than bound allows\n\ngoroutine 1 [running]:\nmain.gcMemoryLimit(0xffffffffffffffff)\n\t/Users/swarming/.swarming/w/ir/x/w/goroot/src/runtime/testdata/testprog/gc.go:404 +0x465\nmain.GCMemoryLimitNoGCPercent()\n\t/Users/swarming/.swarming/w/ir/x/w/goroot/src/runtime/testdata/testprog/gc.go:317 +0x1a\nmain.main()\n\t/Users/swarming/.swarming/w/ir/x/w/goroot/src/runtime/testdata/testprog/main.go:34 +0x133\n"
--- FAIL: TestMemoryLimitNoGCPercent (3.55s)

— watchflakes

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
default <- pkg == "runtime" && test == "TestMemoryLimitNoGCPercent"

2025-04-01 22:09 gotip-darwin-amd64-longtest go@dceb77a3 runtime.TestMemoryLimitNoGCPercent (log)

=== RUN   TestMemoryLimitNoGCPercent
    gc_test.go:789: /Users/swarming/.swarming/w/ir/x/t/go-build568260939/testprog.exe GCMemoryLimitNoGCPercent: exit status 2
    gc_test.go:792: expected "OK\n", but got "retained=285218120 limit=268435456 bound=285212672\npanic: exceeded memory limit by more than bound allows\n\ngoroutine 1 [running]:\nmain.gcMemoryLimit(0xffffffffffffffff)\n\t/Users/swarming/.swarming/w/ir/x/w/goroot/src/runtime/testdata/testprog/gc.go:404 +0x465\nmain.GCMemoryLimitNoGCPercent()\n\t/Users/swarming/.swarming/w/ir/x/w/goroot/src/runtime/testdata/testprog/gc.go:317 +0x1a\nmain.main()\n\t/Users/swarming/.swarming/w/ir/x/w/goroot/src/runtime/testdata/testprog/main.go:34 +0x133\n"
--- FAIL: TestMemoryLimitNoGCPercent (3.55s)

— watchflakes

[randall77]

This is just over the limit allowed. Maybe we need to loosen the condition a bit?

[mknyszek]

The condition is already somewhat loose (a couple percent IIRC). This is in theory always possible since the limit is soft, and the test is really allocating quite quickly. Maybe we retry? IIRC this has happened before, and off the top of my head maybe about once per year. @dmitshur notes that #66893 is the only other instance.