Daily supervision
This page describes the routine check of a running processing. It assumes the scenarios are already configured and running unattended; the aim is to confirm that a run is progressing normally and to notice problems early.
Contents
The routine check
Open the Monitor screen. By default it shows the currently active processings.
- Check the executor status. The status button in the filter bar should read Running. If it reads Paused, Stopping or Stopped, nothing new will start — see Executor control.
- Check for executor messages. The warning button next to the status appears only when messages exist. Errors reported here are produced by the executor itself, not by individual tasks, and often explain a whole processing failing to start.
- Review the Active processings panel. For each processing check the status, the percentage complete and the estimated finish.
- Look at the counters. A non-zero Error counter needs attention; see Handling failed tasks. A Warning counter means the tasks completed but reported something worth reading.
- Confirm progress. A processing with a Running counter of zero and a Waiting counter above zero is not progressing. See Investigating a stalled processing.
Reading the estimate
For each running processing the panel shows a percentage complete and an estimated finish time, calculated from the recorded average runtime of the tasks involved.
Treat the estimate as an indication, not a commitment. It is only as good as the run history behind it:
- a processing running for the first time, or after a large change to the task list, has little history to work from
- a run that is behind because of a failure will show an estimate that assumes normal execution from that point on
Where no history is available, the panel says so rather than showing a misleading number.
Deciding whether a run is late
Compare the estimated finish against the time the downstream consumers expect the data. If the run will not make it:
- check whether the delay comes from a failed task that has been retrying — see Handling failed tasks
- check whether tasks are waiting on resource limits rather than dependencies — see Resource management
- if the delay is structural rather than a one-off, analyse it properly instead of intervening every morning; see Analysing processing performance
Narrowing the view
On an installation with many scenarios the default view is too broad for a quick check. Useful narrowing steps:
- use the Processings active today button to restrict the list to the current day
- add a scenario or group filter chip to follow one part of the workload
- use the tag filter to follow the tasks that feed one consumer — see Tag
- type into the text box to jump to a task by name or ID
Filters combine, and each is shown as a chip that can be removed individually.
Following a dependency chain
When a specific output is late, select the task that produces it and open the Backward tab. It lists what the task is waiting for, with each entry's current status. Clicking an entry moves the panel to that task, so the chain can be walked back to whatever is actually blocking it.
The Forward tab does the reverse, which is the quicker way to answer "if this task fails, what else is affected?".
What not to do during a run
- Do not change dependencies while a processing is running. Dafne checks for dependency cycles when a processing is planned, not while it runs, so a cycle introduced mid-run will not be detected and will have to be resolved manually. See Dependencies.
- Do not use Set complete to move a run along unless the work the task represents has genuinely been done. Dependent tasks will start on the assumption that its output exists.
