FlexMeasures / flexmeasures / 39314076735

87%

push

github

Fix: pin retried run start (#2661)

* data/services/automations: pin a retried forecast run's start

Context:
- Issue #2537, reported by Felix in review of PR 2457: an automation which leaves its start to the run time resolves a new one on every dispatch attempt,
  so a run resumed hours later queued its remaining jobs for a later window than the ones already queued,
  and the wrap-up job reported on a different window than some of the cycles it waited for.

Change:
- Store the start the first attempt resolves on the run's parameters, so later attempts read it back.
- Pin only the start: the end follows from it and the automation's duration, and giving both alongside a duration is rejected as contradictory.
- This does not settle the belief time: without a start, the schema saves forecasts as believed at the time of dispatch,
  while with the pinned start it saves them as believed at that start,
  so the jobs a resumed run queues are believed up to one sensor resolution earlier than the ones its first attempt queued.
- Schedule automations default their start the same way in 'prepare_schedule_trigger_message', which matters once schedule runs are retried (#2510).

Signed-off-by: Mohamed Belhsan Hmida <mohamedbelhsanhmida@gmail.com>

* data/tests: cover a forecast run resumed hours after its first attempt

Context:
- Issue #2537: the jobs of one resumed run could forecast from different starts.

Change:
- Queue one job at 01:00 and fail, resume at 03:00, and require every job of the run to start at 01:00.
- Without the pin, cycle-002 and the wrap-up job start at 03:00 and the test fails.

Signed-off-by: Mohamed Belhsan Hmida <mohamedbelhsanhmida@gmail.com>

* docs/automations: a resumed run keeps the window its first attempt started on

Context:
- The retry section described issue #2537 as current behaviour: a resumed run's jobs could cover a later window.

Change:
- Say that a forecast start left to the run time is resolved on the first attempt and st... (continued)

20 of 21 new or added lines in 2 files covered. (95.24%)

22123 of 25491 relevant lines covered (86.79%)

0.87 hits per line