retcon

’26 · go · sqlite

open source

g go esc index

retcon run · /tmp/retcon.dbtty

deploy_pipeline10 actions · db: /tmp/retcon.dbdeploy_staging         httpstaging_healthcheck    httpdeploy_production      http

run · status · history · replay

A workflow runtime on SQLite. After a crash it continues from where it stopped.

A job dies halfway. Restart it blind and you charge twice, skip a step, or lose what already ran.

Retcon keeps an append-only history in SQLite WAL. A script is a YAML or JSON DAG. A timeline is one run. Actions are http, exec, sleep, webhook, grpc, transform, assert, and noop. The scheduler re-claims work when a lease expires. Execution is at-least-once, so HTTP actions send an Idempotency-Key.

replay rebuilds state from history and does not call the network. There is no Redis, Postgres, or queue to stand up. One binary and one database file.

It is still an MVP. There is no server mode and no cancellation. Retry backoff is in-process. If the process restarts during backoff, the timer starts over.

example

make build, then run the bundled example: ./retcon run examples/deploy_pipeline.yaml --db /tmp/retcon.db Ten actions. HTTP steps hit httpbin.org, not your cluster. The run prints a timeline id. status <id> and history <id> read that same file. replay <id> rebuilds the run and does not call the network. Kill the process mid-run if you want to see it continue from the SQLite file.

Use

github.com/doganaydindev/retcon ↗

nextferris-pulse