Repository navigation
[BUG] Intermittent SIGBUS (exit 135) in MTP test hosts on Linux with dynamic instrumentation (dotnet-coverage 18.11.2) #253
Description
- Linux: MTP test hosts intermittently die at startup with SIGBUS (exit 135) under
dotnet-coverage collectwith default dynamic instrumentation;dotnet testreportsZero tests ran, module varies - Static instrumentation (
EnableDynamicManagedInstrumentation=false,EnableStaticManagedInstrumentation=true,--include-files) or no coverage: no crashes - Env: GitHub Actions ubuntu-24.04 (8 vCPU), .NET 10.0.12 (also seen on 10.0.11), dotnet-coverage 18.11.2, TUnit 1.70.1 / MTP 2.4.1,
--report-trx - Repro: 8 MTP test apps started in parallel under one
dotnet-coverage collectsession, 40 iterations (320 host starts) per mode
| Mode | Exit 135 | Exit 7 (zero tests) |
|---|---|---|
| No coverage | 0 | 0 |
| Dynamic (default) | 6 | 10 |
| Static | 0 | 0 |
Linux counterpart of #249 (this is the isolation run suggested there); possibly related to #232 / #238.
Reacted by Amaury Levé and Alan Deutscher
Activity
adeutscher commentedon Oct 25, 2026
I'm running into Exit Code 135 problems as well, and I think my problems are also MTP-based.
dotnet coverage collect --session-id <id> "dotnet test <sln> --no-build -c Release --report-xunit-xml --results-directory artifacts/ci" -f xml -o coverage.cobertura.xmlI'm testing a band-aid for this by setting a limit on parallel jobs by adding
--max-parallel-test-modules 1to the innerdotnet testcall. Not super-ideal, but it beats all the crashes. Early results look good (zero crashes after a bunch of consecutive runs of the pipeline), but on the other hand this fix is new this afternoon. Time will tell.build.sh Cleanup (#88)