CLI

@dudousxd/nestjs-durable-cli gives you durable inspect and durable cancel — a quick look at runs and their step timelines, and a way to stop one, without opening the dashboard.

pnpm add -D @dudousxd/nestjs-durable-cli

Point it at your store

The CLI reads from a durable.config.{mjs,js,cjs} in your project. Export your already-configured store as store, or as the default export:

import { MikroOrmStateStore } from '@dudousxd/nestjs-durable-store-mikro-orm';
import { buildOrm } from './src/orm';

export const store = new MikroOrmStateStore(await buildOrm());

A default export works too — either the store itself, or an object with a store property (e.g. export default { store }). The CLI checks store, then default.store, then default, in that order, and errors if none of them is a StateStore.

Use it

durable inspect                  # list recent runs
durable inspect wrun_8Kb2        # show one run's step timeline
durable inspect --status failed  # filter the list

RUN        WORKFLOW  STATUS     UPDATED
wrun_8Kb2  checkout  running    3s ago
wrun_Qz71  onboarding failed    2m ago

checkout  [completed]  v1
wrun_8Kb2

  0  reserveStock          (local)              completed  150ms
  1  payments.charge-card  (remote @payments)   completed  750ms  ×2
  2  sleep                 (sleep)              completed  3.0s
  3  shipping.dispatch     (remote @logistics)  completed  800ms

Reading the output:

• ×2 after a step's duration is an attempt count — it only shows up once a step has retried (attempts > 1); a step that succeeded on its first try has no suffix.
• @payments / @logistics is the worker group the step ran on, shown only for steps that recorded one.
• Timestamps in the runs list (UPDATED) are relative — seconds/minutes/hours/days ago, rounded to the coarsest unit that fits (3s ago, 2m ago, 4h ago, 1d ago).

Cancel a run

durable cancel wrun_8Kb2

This marks the run cancelled in the store — a suspended run won't resume, and the recovery/timer poller skips it going forward. --config applies here too, same as inspect.

It's a soft cancel, not a hard kill: if a step is already in flight on some worker, that worker finishes its current attempt before it next checks the run's status, so cancellation is observed at the next checkpoint rather than interrupting execution mid-step. Cancelling a run that's already completed, failed, or cancelled is a no-op — the CLI tells you it's already terminal and leaves it alone.

Usage and exit codes

Running durable with no subcommand, or an unrecognized one, prints the usage text instead of doing anything:

durable            # prints usage, exits 0
durable frobnicate  # prints usage, exits 1 (unknown subcommand)