We're currently implementing a workflow in Amazon SWF where we submit jobs/workflow executions from our web application. Everything was fairly quick and painless to get set up using the Ruby Flow framework. As long as the deciders/activity workers don't crash we seem to be able to handle most issues/exceptions gracefully.
My question is, what is common practice for the scenario where the decider process crashes midway through a workflow execution? If the task fails in that way, is it possible to push an SNS notification (I've seen no examples) or something to indicate to another process that there's been an unexpected failure/crash?
There are various types of "decider" failures.
Workflow worker crashes while processing a decision. The decision task is automatically rescheduled after specified timeout. Make sure that workflow type defaultTaskStartToCloseTimeout is not set too high. If this crash is not related to code correctness then rescheduled task is processed and workflow execution continues normally.
Workflow worker doesn't crash but workflow execution itself fails. In this case you can use ListClosedWorkflowExecutions to count such failed workflows.
Workflow worker doesn't crash but a decision task cannot complete as RespondDecisionTaskCompleted fails due to a bug in the Flow framework. As from SWF point of view task is never completed it at some point is marked as timed out and rescheduled. As bug is still present a new task is again never completes and rescheduled, and so on. The workflow execution that is experiencing such issue has a history with a tail that consists from repeated "decision task scheduled, decision task timed out" events. If your workflow has a known execution time limit then the best way to catch this issue is to set reasonable executionStartToCloseTimeout and look for timed out workflow executions. If the decision task timeout is set too low such workflows can also hit the limit on history size before the execution timeout.
All swf metrics are not published to cloud watch. So all completed and failed workflows will send the metrics to cloudwatch where you can create alarms to send you notifications when any workflow fails.
Related
The documentation of GitHub actions says:
You can use jobs.<job_id>.concurrency to ensure that only a single job or workflow using the same concurrency group will run at a time.
...
When a concurrent job or workflow is queued, if another job or workflow using the same concurrency group in the repository is in progress, the queued job or workflow will be pending. Any previously pending job or workflow in the concurrency group will be canceled.
It is annoying that previously pending jobs get cancelled. Evidently the orchestration logic can only maintain a tiny "queue" of one (1) pending job.
I would like to be able to have multiple jobs enqueued. I.e., if I trigger 5 jobs in rapid succession, and they all belong to the same concurrency group, then the first one starts to run immediately (when a runner is availble) and the next 4 get enqueued and wait for their turn to run, one at a time.
Is there any way to achieve this? Or will I need to request this as a feature from GitHub?
We have a flow where if some actions are not done by a certain time period, we want to fail the workflow, to use alarming mechanisms.
For failing the workflow, I was initially thinking of just returning an exception from the code. But after reading sources online, it seems like the exception in decider flow will not let the host return the result and some other host will just pick the pending decision task after some time.
Wanted to know if there's a programmatic way to terminate the workflow in between via workflow code and marking the SWF workflow execution as failed.
To fail a workflow using the Flow Framework throw an Exception or its subclass from the workflow code.
Don't throw an Error from the workflow code. This indeed is going to fail a decision task which will lead to the workflow getting blocked in a retry loop.
I am using AWS SWF and flow framework. I wanted to make my activities idempotent so that a workflow can be restarted from the beginning after any failure. Many of the activities are manual tasks (#ManualActivityCompletion) which need to be completed asynchronously.
Is there a way to finish manual tasks like normal tasks if I know that it is already complete? This way a new manual task will not be scheduled everytime the workflow is retried.
Or, is there a way to retry a workflow so that it starts from the point it failed?
Currently there is no way to override activity completion behavior at runtime. The work around is to complete activity using ManualActivityCompletionClient from within activity implementation.
There is no supported way to retry workflow to start from the point of failure.
The maximum amount of time the pollForActivityTask method stays open polling for requests is 60 seconds. I am currently scheduling a cron job every minute to call my activity worker file so that my activity worker machine is constantly polling for jobs.
Is this the correct way to have continuous queue coverage?
The way that the Java Flow SDK does it and the way that you create an ActivityWorker, give it a tasklist, domain, activity implementations, and a few other settings. You set both the setPollThreadCount and setTaskExecutorSize. The polling threads long poll and then hand over work to the executor threads to avoid blocking further polling. You call start on the ActivityWorker to boot it up and when wanting to shutdown the workers, you can call one of the shutdown methods (usually best to call shutdownAndAwaitTermination).
Essentially your workers are long lived and need to deal with a few factors:
New versions of Activities
Various tasklists
Scaling independently on tasklist, activity implementations, workflow workers, host sizes, etc.
Handle error cases and deal with polling
Handle shutdowns (in case of deployments and new versions)
I ended using a solution where I had another script file that is called by a cron job every minute. This file checks whether an activity worker is already running in the background (if so, I assume a workflow execution is already being processed on the current server).
If no activity worker is there, then the previous long poll has completed and we launch the activity worker script again. If there is an activity worker already present, then the previous poll found a workflow execution and started processing so we refrain from launching another activity worker.
Is there a way to kill / re-start a long running task in AWS SWF? Sometimes some of our tasks run for a longer duration and we would like to manually kill a certain task (either via UI or programmatically) and re-start the task if possible. How to achieve this?
Console is option to manually kill workflow.
You can also set timeouts to whole workflow execution time or to individual activities. This can be set when you register your activity or when you start your activity (defaultTaskStartToCloseTimeoutSecond).
It's not clear what language you're using.
If you're using java, then you should look into Exponential Retry in Flow Framework. This make SDK restart your activity if it fails.
Long running activity is expected to heartbeat using RecordActivityTaskHeartbeat. It leads to timeout failure after short hearbeat interval instead of long task execution timeout if the activity process hangs or crashes.
The workflow code (decider) can always request activity cancellation through RequestCancelActivityTask decision. The cancellation request is returned as output of the RecordActivityTaskHeartbeat call. Activity implementation should cancel itself and report back to the service using RespondActivityTaskCanceled API call.
See Error Handling section of AWS Flow Framework Developer Guide for the AWS Flow Framework way of cancelling activities.
Sometimes activity implementation cannot support heartbeating and self cancellation. The solution is to execute another kill activity that terminates the first activity execution. For example under Unix such kill activity could emit "kill -9" command for the process that implements the first one.