ECS Executor - add support to adopt orphaned tasks. #36803
Conversation
boring-cyborg
bot
added
area:providers
provider:amazon-aws
| """ | ||
| Try to adopt running task instances if adopt_task_instances option is set to True. | ||
|
|
||
| These tasks instances should have an ECS process which can be adopted by the unique task ARN. |
There was a problem hiding this comment.
Is it expected that there is no difference for the adoption process between a completed/terminated state to a running state of the instance?
There was a problem hiding this comment.
I think this is right. For comparison,. k8s executor filters out "status == successful", but local just passes them all, and celery appears to print the state for each but isn't filtering.
| @@ -377,6 +383,9 @@ def attempt_task_runs(self): | |||
| else: | |||
| task = run_task_response["tasks"][0] | |||
| self.active_workers.add_task(task, task_key, queue, cmd, exec_config, attempt_number) | |||
| # Add Fargate task ARN to executor event buffer, which gets saved | |||
| # in TaskInstance.external_executor_id. | |||
| self.event_buffer[task_key] = (State.QUEUED, task.task_arn) | |||
There was a problem hiding this comment.
We have self.success and self.fail (e.g. link) which handles the other state changes. These are provided by the base executor. I wonder if we should add a new method for putting a task in queued state 🤔 but this might cause weird issues if a provider which contains an executor is installed alongside an older version of airflow... (also, not required for this PR, just thinking out loud)
airflow/jobs/scheduler_job_runner.py
Outdated
| @@ -721,6 +721,7 @@ def _process_executor_events(self, session: Session) -> int: | |||
| if state == TaskInstanceState.QUEUED: | |||
| ti.external_executor_id = info | |||
| self.log.info("Setting external_id for %s to %s", ti, info) | |||
| session.commit() | |||
There was a problem hiding this comment.
I'm not a DB/SQLA expert, but do we want to commit here, our outside of the for loop? Here is going to do a commit for each queued task individually that we got from the query. Which could be many? Interested to hear others weigh in.
There was a problem hiding this comment.
A valid point, and I don't know the right answer. I'll see what others think. It may be safest to move it outside the loop
There was a problem hiding this comment.
For posterity, I have started a Slack discussion [here] for this to try to get more attention. I'll drop a comment here with whatever decision is reached there.
ferruzzi
force-pushed
the
ferruzzi/ecs-executor/adoption-agency
branch
2 times, most recently
from
883e672
to
47a7e05
Compare
|
Accidentally did a rebase and ingested the bug fixed by #36958, re-rebased so CI should clear now. EDIT: ... or maybe not... I'll look into it. |
ferruzzi
force-pushed
the
ferruzzi/ecs-executor/adoption-agency
branch
from
47a7e05
to
06fff7a
Compare
Co-authored-by: o-nikolas
Co-authored-by: Niko Oliveira <[email protected]>
ferruzzi
force-pushed
the
ferruzzi/ecs-executor/adoption-agency
branch
from
2ba8803
to
63b10c5
Compare
|
Unless anyone here feels strongly otherwise, I think I'm going to close this one and open a new PR. It's been a while and it needs a rebase, plus there are now a few commits which can be squashed, and it looks like it is going to be dependent on a bugfix that I'll submit in a separate PR. No real point leaving this open and then doing a rebase. |
Adds support for "adopt orphaned tasks" to the ECS Executor.
A task can become orphaned, for example, when the scheduler container is re-deployed or terminated or other cases where the SchedulerJob got interrupted.
try_adopt_task_instancesis defined in the BaseExecutor (this PR implements it in the ECS Executor) and is called by the Scheduler here.Based heavily on the work here by Matt Ellis (who I can't tag for some reason) .
closes: #35491
^ Add meaningful description above
Read the Pull Request Guidelines for more information.
In case of fundamental code changes, an Airflow Improvement Proposal (AIP) is needed.
In case of a new dependency, check compliance with the ASF 3rd Party License Policy.
In case of backwards incompatible changes please leave a note in a newsfragment file, named
{pr_number}.significant.rstor{issue_number}.significant.rst, in newsfragments.