I’m seeing SQLite-backed @effect/workflow / @effect/cluster runner storage become unhealthy when a long-lived runner process and multiple short-lived control-plane clients access the same workflow store concurrently.
Reproduction
https://github.com/schickling-repros/2026-04-effect-workflow-sqlite-lock
npm install
npm run repro
The repro uses:
- one long-lived runner process backed by
SingleRunner.layer({ runnerStorage: "sql" })
- SQLite via
@effect/sql-sqlite-node
- many one-shot control-plane processes that concurrently call workflow start / poll operations against the same SQLite file
Expected
Concurrent control-plane access should stay healthy when talking to the same SQLite-backed workflow store as the runner.
Actual
The setup becomes unhealthy under contention. In the minimized repro, many control-plane operations time out. In the larger integration where I first hit this, the same family of failures also surfaced as PersistenceError and SqliteError: database is locked.
Versions
@effect/cluster: 0.58.0@effect/workflow: 0.18.0@effect/sql-sqlite-node: 0.52.0effect: 3.21.0- Node:
v24.14.0
- OS:
Darwin arm64
If useful, I can try a second reduced repro that focuses specifically on the PersistenceError path rather than the timeout/lockup symptom.
Filed by an AI assistant on behalf of @schickling
I’m seeing SQLite-backed
@effect/workflow/@effect/clusterrunner storage become unhealthy when a long-lived runner process and multiple short-lived control-plane clients access the same workflow store concurrently.Reproduction
https://github.com/schickling-repros/2026-04-effect-workflow-sqlite-lock
The repro uses:
SingleRunner.layer({ runnerStorage: "sql" })@effect/sql-sqlite-nodeExpected
Concurrent control-plane access should stay healthy when talking to the same SQLite-backed workflow store as the runner.
Actual
The setup becomes unhealthy under contention. In the minimized repro, many control-plane operations time out. In the larger integration where I first hit this, the same family of failures also surfaced as
PersistenceErrorandSqliteError: database is locked.Versions
@effect/cluster:0.58.0@effect/workflow:0.18.0@effect/sql-sqlite-node:0.52.0effect:3.21.0v24.14.0Darwin arm64If useful, I can try a second reduced repro that focuses specifically on the
PersistenceErrorpath rather than the timeout/lockup symptom.Filed by an AI assistant on behalf of @schickling