Watch and notifications¶
Follow registered destinations from the command line:
watch samples job state, prints changes as JSON, and backs off after failures.
The first sample establishes a baseline; it does not replay historical
completions. Use molq job history hpc to inspect earlier jobs.
Defaults¶
Add a watcher configuration to your configuration file:
Pass --cluster repeatedly to watch several destinations. If no clusters are
selected, the watcher uses registered destinations.
Notifications¶
Set client.watch.nerve_endpoint to your receiver URL. The receiver gets the
same JSON envelopes that the watcher prints. Cadence and notification routing
belong to the watcher configuration, not to a cluster definition.
Delivery failures do not block observations. Slow receivers may miss notifications; stdout remains the record of observed changes. A repeated scheduler error is printed once until the destination succeeds again.
Use a shared service when other clients should receive the same observed events.