Interface inventory: the text here is the text on screen. For the step-by-step investigation, see Track a run.
Where it is: menu ETL/ELT → Runs (formerly called DAGs).
Find a run in the list (/dags)#
Title Runs, subtitle "See when each pipeline ran, whether it worked and when it runs again." There's no create button here: pipelines are created in Pipelines. With no pipelines: No pipeline yet + the Create pipeline button.
The four indicators, and why there are four
| Indicator | Scope |
|---|---|
| running | now |
| failed on last run | current state |
| pipelines with failures in period | history (counts pipelines; how many runs failed appears next to it) |
| paused schedules | current state |
Failing now and failing in the history are different questions
failed on last run is what needs action today. failures in period is how many times it failed in the chosen window — a pipeline can have ten failures in the period and be healthy now. The two filter chips repeat this split: Failed on last run and Failures in period.
Toolbar
- Search: "Search pipeline…"
- Chips: Failed on last run · Failures in period
- Schedule: all · active (time-based) · on file arrival · paused · manual
- Period: Last 7 days · Last 15 days · Last 30 days · Last 90 days · All
- Clear filters
Columns: Pipeline · Last run · Schedule · Duration · Next run · Actions. In Next run, when the trigger is an event, the screen writes "on file arrival" instead of a time, "schedule paused" when it's turned off — and — when there's no schedule. Dates and times are in Brasília time.
Each row's actions
Open and, as icons: "Open with the last run already selected", "Run now (asks for confirmation)", "Try again" (when the last run failed — it opens that run's recovery) or "Refresh this row", "View runs without leaving the list", and the schedule switch. Running from the list asks for confirmation (scope whole pipeline / current version) and only then Confirm; if the run is refused (balance, permission…), the reason appears below the buttons.
A Manual pipeline has no schedule switch: in its place, Set time appears, which opens the pipeline's Scheduling stage. The pipeline's technical code is in the expanded panel (Pipeline code).
The schedule switch says exactly what it does: "Pause the schedule — stops new scheduled runs and does NOT cancel the run in progress" and "Activate the schedule — does not trigger a load now".
Run, investigate, and recover a run (/dags/[id])#
Go back with Back to runs. At the top, the Extraction or Gold report badge, and the split Run now button, whose menu offers Run now and Run with parameters.
Four tabs: Run · Graph · Logs · Definition. The first becomes Task when you select a task — that's how you drill down from run to task, from task to attempt, and only then to that attempt's log.
The three actions on a run
- Recover… — opens the recovery panel
- Mark success — you declare the run resolved without running it again
- Mark failed — you declare the run failed
Mark success doesn't process anything
Mark success and Mark failed change the record, not the data. Use them when the problem was solved outside the product and the history needs to reflect it. To actually reprocess, use Recover….
The Recover run panel
Three steps: Choose scope → Review impact → Track result. And four modes, which are not the same thing:
| Mode | What it does |
|---|---|
| New task attempt | just that task, again |
| Rerun task and dependents | that task and everything that depends on it |
| Run the pipeline again | the whole run |
| Reprocess period | a date range |
The confirmation button describes the operation instead of saying "OK": Confirm rerun of {N} task(s), Confirm new run of {N} task(s), or Confirm reprocessing of {N} day(s).
The state vocabulary#
Run: queued · running · success · partial (with failures) · failed · stopped (DLQ) · skipped · unknown state.
Task: the same, plus not run.
"partial (with failures)" is a state, not a half-success
A partial (with failures) run finished with tasks that failed. It's not a success: part of the data didn't reach the target. Check before treating the result as good. A partial run is not retried automatically and does not go to the DLQ: fix the task and use Recover… › New task attempt.
Back to the app#
- Runs: ETL/ELT → Runs