Please see cmd/
for enqueuer and worker demo.
- queue backend abstraction
- redis is still the default, but the new design allows custom queue implementation.
- simplify the keyspace design of redis queue backend
- The new design uses 1 redis hash per job, and 1 redis sorted set for queue.
- Interesting read
- modular
- The core only catches panics, retries on failure, and waits if a queue is empty.
- All other functionalities are either removed or moved to separate middlewares.
- support binary payload/args with message pack.
- replace built-in UI with prometheus metrics (use grafana if you want dashboard).
- additional optimizations (alloc + bulk queue ops)
BenchmarkWorkerRunJob/work_v1_1-8 3000 515957 ns/op BenchmarkWorkerRunJob/work_v2_1-8 5000 284516 ns/op BenchmarkWorkerRunJob/work_v1_10-8 1000 2136546 ns/op BenchmarkWorkerRunJob/work_v2_10-8 5000 367997 ns/op BenchmarkWorkerRunJob/work_v1_100-8 100 18234023 ns/op BenchmarkWorkerRunJob/work_v2_100-8 1000 1759186 ns/op BenchmarkWorkerRunJob/work_v1_1000-8 10 162110100 ns/op BenchmarkWorkerRunJob/work_v2_1000-8 100 12646080 ns/op BenchmarkWorkerRunJob/work_v1_10000-8 1 1691287122 ns/op BenchmarkWorkerRunJob/work_v2_10000-8 10 144923087 ns/op BenchmarkWorkerRunJob/work_v1_100000-8 1 17515722574 ns/op BenchmarkWorkerRunJob/work_v2_100000-8 1 1502468637 ns/op PASS ok github.com/taylorchu/work 87.901s
- http server
- delete job
- create job
- get job status
- get queue metrics (kubernetes autoscaler integration with keda metrics api scaler)
- OpenAPI spec