Expand description
Testing tools for schedulers and processors built on grommet.
The runtime can be proven correct on its own, but your processor is where the interesting failures live: a lost acknowledgement, a retry that applies twice, a cache that outlives the value it caches. This crate provides the machinery to find those, generic over your fault labels and your workload.
FaultPlaninjects failures at points you label, either in a fixed order, at one chosen position, or steered by a fuzzer.conformance::single_fault_campaignruns your workload once per reachable failure position and insists every one of them recovers.conformance::assert_idempotentreplays one request and insists only the first application changed anything.conformance::scheduler_sweepdrives your scheduler configuration through a mixed workload, checking every structural invariant and measuring the starvation bound.Deterministicplusconformance::assert_deterministicturn “this is reproducible” from an unchecked claim into a test.
Nothing here is a marker trait that grants correctness by being implemented. Every check runs your code and reports what it observed.
Re-exports§
pub use conformance::CampaignReport;pub use conformance::CaseOutcome;pub use conformance::SweepReport;pub use conformance::SweepSpec;pub use determinism::Deterministic;pub use fault::FaultPlan;pub use fault::FaultPoint;
Modules§
- conformance
- Checks you can run against your own processor and configuration.
- determinism
- Declaring, and then actually checking, that a processor is reproducible.
- fault
- Deterministic fault injection for the adapters a processor depends on.
Structs§
- Inline
Offload - Runs the closure inline on the calling shard.
- Manual
Clock - A clock that only moves when a test moves it. Clones share one timeline, so a test can hold one handle while the runtime holds another.