hpcflow.sdk.monitoring.status.observe_activity#
- hpcflow.sdk.monitoring.status.observe_activity(app, records, progress=None, *, min_interval=0, ignore_scheduler_interval=False, no_update=False)#
Update records using one snapshot per compatible scheduler query target.
Submission directives and submit/cancel commands do not split status queries. Deduplicate references across submissions and reuse each query’s observation for all matching jobs. Different effective show commands or targets remain isolated. Malformed references fail individually instead of blocking the batch. Batch saved references across submission parts; known inactive records do not need queries. A live submitting process also keeps a record active. Scheduler failures produce unknown activity unless activity is independently known to be active, and are recorded without aborting observations of other contexts. Foreign jobs without a shared configured machine remain remotely unavailable; archives use saved observations only. Neither case inspects local PIDs or makes the workflow unreadable.
progressreceives transient query messages. Update submission records in memory; the caller controls registry persistence. Rate-limit snapshots are saved separately unlessno_updateis True.min_intervalenables persistent rate limiting per query target. Cached snapshots retain their observation time and do not supply fresh queued/terminal parent evidence. New references absent from a cached snapshot remain unknown.