Skip to main content

Deterministic Testing

Kommander includes a non-production test toolchain for deterministic and timing-sensitive scenarios.

Simulation Runtime​

The source tree includes a deterministic simulation runtime with:

  • seeded randomness
  • replay logs
  • virtual time
  • reproducible failure scenarios.

This is useful when a timing-sensitive failure is hard to reproduce in ordinary multi-process tests.

Several timing controls help reproducibility directly in the runtime:

  • ElectionTimeoutSeed can make partition election timeouts deterministic
  • WaitForLeaderStableAsync lets tests wait for a leader that stays stable for a minimum duration.

The election tests include targeted pre-vote coverage:

  • state-machine tests verify that pre-vote stays side-effect free until quorum is reached
  • transport tests verify the PreVote RPC flag survives gRPC serialization
  • multi-node tests cover the stale follower rejoin case that pre-vote is meant to stabilize.

Focused Test Areas​

The current test tree includes several useful slices:

  • Kommander.Tests.Simulation: simulation and replay.
  • Kommander.Tests.Scheduler: partition executor, fair schedulers, timer service, transport batching, and system coordinator behavior.
  • Kommander.Tests.RaftSafety: election safety, commit monotonicity, stale completion handling, log matching, and system restore behavior.
  • Kommander.Tests.WAL: RocksDB, SQLite, and automatic compaction coverage.

Useful Commands​

dotnet test Kommander.Tests/Kommander.Tests.csproj --filter FullyQualifiedName~Kommander.Tests.Simulation
dotnet test Kommander.Tests/Kommander.Tests.csproj --filter FullyQualifiedName~Kommander.Tests.Scheduler
dotnet test Kommander.Tests/Kommander.Tests.csproj --filter FullyQualifiedName~Kommander.Tests.RaftSafety
dotnet test Kommander.Tests/Kommander.Tests.csproj --filter FullyQualifiedName~Kommander.Tests.WAL

Why It Matters​

This tooling makes it easier to verify invariants, reproduce leadership churn, test delayed I/O, validate pre-vote behavior under isolation, and replay a failure with the same random seed instead of trying to rediscover the bug by chance.