Thank you for reading this SO question, it may seem long, but I'll try to get as most information as possible in it to help to get the answer.
Summary
We are currently experiencing a scheduling issue with our Flink cluster.
The symptoms are that some/most/all (it depends, the symptoms are not always the same) of our tasks are shown as SCHEDULED but fail after a timeout. The jobs are then shown as RUNNING.
The failing exception is the following one:
Caused by: java.util.concurrent.CompletionException: org.apache.flink.runtime.jobmanager.scheduler.NoResourceAvailableException: Slot request bulk is not fulfillable! Could not allocate the required slot within slot request timeout
After analysis, we assume (we cannot prove it, as there are not that much logs for that part of the code) that the failure is due to a deadlock/race condition that is happening when several jobs are being submitted at the same time to the Flink cluster, even though we have enough slots available in the cluster.
We actually have the error with 52 available task slots, and have 12 jobs that are not scheduled.
Additional information
Flink version: 1.13.1 commit a7f3192
Flink cluster in session mode
2 Job managers using k8s HA mode (resource requests: 2 CPU, 4Gb Ram, limits sets on memory to 4Gb)
50 task managers with 2 slots each (resource requests: 2 CPUs, 2GB Ram. No limits set).
Our Flink cluster is shut down every night, and restarted every morning. The error seems to occur when a lot of jobs needs to be scheduled. The jobs are configured to restore their state, and we do not see any issues for jobs that are being scheduled and run correctly, it seems to really be related to a scheduling issue.
Questions
May it be that the issue described in FLINK-23409 is actually the same, but occurs only when there is a race condition when scheduling several jobs?
Is there any way to increase logging in the scheduler to debug this issue?
Is it a known issue? If yes, is there any workaround/solution to resolve it?
P.S: a while ago, I asked more or less the same question on the ML, but dropped it, I'm sorry if this is considered as cross-asking, it's not intended t. We are just opening a new thread as we have more information and the issue re-occur.
Related
Are there any existing out of the box job queue framework? basic idea is
someone to enqueue a job with job status New
(multiple) workers get a job and work on it, mark the job as Taken. One job can only be running on at most one worker
something will monitor the worker status, if the running jobs exceed predefined timeout, will be re-queued with status New, could be worker health issue
Once a worker completes a task, it marks the task as Completed in the queue.
something keeps cleaning up completed tasks. Or at step #4 when worker completes a task, the worker simply dequeues the task.
From my investigation, things like Kafka (pub/sub) or MQ (push/pull & pub/sub) or cache (Redis, Memcached) are mostly sufficient for this work. However, they all require some sort of development around its core functionality to become a fully functional job queue.
Also looked into relational DB, the ones supports "SELECT * FOR UPDATE SKIP LOCKED" syntax is also a good candidate, this again requires a daemon between the DB and worker, which means extra effort.
Also looked into the cloud solutions, Azure Queue storage, etc. similar assessment.
So my question is, is there any out of the box solution for job queue, that are tailored and dedicated for one thing, job queuing, without much effort to set up?
Thanks
Take a look at Python Celery. https://docs.celeryproject.org/en/stable/getting-started/introduction.html
The default mode uses RabbitMQ as the message broker, but other options are available. Results can be stored in a DB if needed.
I am using an Azure batch account to run sqlpackage.exe in order to move databases from a server to another. A task that has started 6 days ago has suddenly been restarted and started from the beginning after 4 days of running (extremely large databases). The task run uninterruptedly up until then and should have continued to run for about 1-2 days.
The PowerShell script that contains all the logic handles all the exceptions that could occur during the execution. Also, the retry count for the task was set to 0 in case it fails.
Unfortunately, I did not have diagnostics settings configured and I could only look at the metrics and there was a short period when there wasn't any node.
What can be the causes for this behavior? Restarting while the node is still running
Thanks
Unfortunately, there is no way to give a definitive answer to this question. You will need to dig into the compute node (interactively log in) and check system logs to give you details on why the node restarted. There is no guarantee that a compute node will have 100% uptime as there may be hardware faults or other service interruptions.
In general, it's best practice to have long running tasks checkpoint progress combined with a retry policy. Programs that can reload state can pick up at the time of the checkpoint when the Batch service automatically reschedules the task execution. Please see the Batch best practices guide for more information.
Mongock looks very promising. We want to use it inside a kubernetes service that has multiple replicas that run in parallel.
We are hoping that when our service is deployed, the first replica will acquire the mongockLock and all of its ChangeLogs/ChangeSets will be completed before the other replicas attempt to run them.
We have a single instance of mongodb running in our kubernetes environment, and we want the mongock ChangeLogs/ChangeSets to execute only once.
Will the mongockLock guarantee that only one replica will run the ChangeLogs/ChangeSets to completion?
Or do I need to enable transactions (or some other configuration)?
I am going to provide the short answer first and then the long one. I suggest you to read the long one too in order to understand it properly.
Short answer
By default, Mongock guarantees that the ChangeLogs/changeSets will be run only by one pod at a time. The one owning the lock.
Long answer
What really happens behind the scenes(if it's not configured otherwise) is that when a pod takes the lock, the others will try to acquire it too, but they can't, so they are forced to wait for a while(configurable, but 4 mins by default)as many times as the lock is configured(3 times by default). After this, if i's not able to acquire it and there is still pending changes to apply, Mongock will throw an MongockException, which should mean the JVM startup fail(what happens by default in Spring).
This is fine in Kubernetes, because it ensures it will restart the pods.
So now, assuming the pods start again and changeLogs/changeSets are already applied, the pods start successfully because they don't even need to acquire the lock as there aren't pending changes to apply.
Potential problem with MongoDB without transaction support and Frameworks like Spring
Now, assuming the lock and the mutual exclusion is clear, I'd like to point out a potential issue that needs to be mitigated by the the changeLog/changeSet design.
This issue applies if you are in an environment such as Kubernetes, which has a pod initialisation time, your migration take longer than that initialisation time an the Mongock process is executed before the pod becomes ready/health(and it's a condition for it). This last condition is highly desired as it ensures the application runs with the right version of the data.
In this situation imagine the Pod starts the Mongock process. After the Kubernetes initialisation time, the process is still not finished, but Kubernetes stops the JVM abruptly. This means that some changeSets were successfully executed, some other not even started(no problem, they will be processed in the next attempt), but one changeSet was partially executed and marked as not done. This is the potential issue. The next time Mongock runs, it will see the changeSet as pending and it will execute it from the beginning. If you haven't designed your changeLogs/changeSets accordingly, you may experience some unexpected results because some part of the data process covered by that changeSet has already taken place and it will happen again.
This, somehow needs to be mitigated. Either with the help of mechanisms like transactions, with a changeLog/changeSet design that takes this into account or both.
Mongock currently provides transactions with “all or nothing”, but it doesn’t really help much as it will retry every time from scratch and will probably end up in an infinite loop. The next version 5 will provide transactions per ChangeLogs and changeSets, which together with good organisation, is the right solution for this.
Meanwhile this issue can be addressed by following this design suggestions.
Just to follow up... Mongock's locking mechanism works fine with replicas. To solve the "long-running script" problem, we will run our Mongock scripts from Kubernetes initContainer. K8s will wait for the initContainers to finish before it starts the pod's main service containers.
For transactions, we will follow the advice above of making our scripts idempotent.
According to both of these Link1 and Link2, my Airflow DAG run is returning the error INFO - Task exited with return code -9 due to an out-of-memory issue. My DAG run has 10 tasks/operators, and each task simply:
makes a query to get one of my BigQuery tables, and
writes the results to a collection in my Mongo database.
The size of the 10 BigQuery tables range from 1MB to 400MB, and the total size of all 10 tables is ~1GB. My docker container has default 2GB of memory and I've increased this to 4GB, however I am still receiving this error from a few of the tasks. I am confused about this, as 4GB should be plenty of memory for this. I am also concerned because, in the future, these tables may become larger (a single table query could be 1-2GB), and I'd like to avoid these return code -9 errors at that time.
I'm not quite sure how to handle this issue, since the point of the DAG is to transfer data from BigQuery to Mongo daily, and the queries / data in-memory for the DAG's tasks is necessarily fairly large then, based on the size of the tables.
As you said, the error message you get corresponds to an out of memory issue.
Referring to the official documentation:
DAG execution is RAM limited. Each task execution starts with two
Airflow processes: task execution and monitoring. Currently, each node
can take up to 6 concurrent tasks. More memory can be consumed,
depending on the size of the DAG.
High memory pressure in any of the GKE nodes will lead the Kubernetes scheduler to evict pods from nodes in an attempt to relieve that pressure. While many different Airflow components are running within GKE, most don't tend to use much memory, so the case that happens most frequently is that a user uploaded a resource-intensive DAG. The Airflow workers run those DAGs, run out of resources, and then get evicted.
You can check it with following steps:
In the Cloud Console, navigate to Kubernetes Engine -> Workloads
Click on airflow-worker, and look under Managed pods
If there are pods that show Evicted, click each evicted pod and look for the The node was low on resource: memory message at the top of the window.
What are the possible ways to fix OOM issue?
Create a new Cloud Composer environment with a larger machine type than the current machine type.
Ensure that the tasks in the DAG are idempotent, which means that the result of running the same DAG run multiple times should be the same as the result of running it once.
Configure task retries by setting the number of retries on the task - this way when your task gets -9'ed by the scheduler it will go to up_for_retry instead of failed
Additionally you can check the behavior of CPU:
In the Cloud Console, navigate to Kubernetes Engine -> Clusters
Locate Node Pools at the bottom of the page, and expand the default-pool section
Click the link listed under Instance groups
Switch to the Monitoring tab, where you can find CPU utilization
Ideally, the GCE instances shouldn't be running over 70% CPU at all times, or the Composer environment may become unstable during resource usage.
I hope you find the above pieces of information useful.
I am going to chunk the data so that less is loaded into any 1 task at any given time. I'm not sure yet whether I will need to use GCS/S3 for intermediary storage.
I am currently running a Flink session cluster (Kubernetes, 1 JobManager, 1 TaskManager, Zookeeper, S3) in which multiple jobs run.
As we are working on adding more jobs, we are looking to improve our deployment and cluster management strategies. We are considering migrating to using job clusters, however there is reservation about the number of containers which will be spawned. One container per job is not an issue, but two containers (1 JM and 1 TM) per job raises concerns about memory consumption. Several of the jobs need high-availability and the ability to use checkpoints and restore from/take savepoints as they aggregate events over a window.
From my reading of the documentation and spending time on Google, I haven't found anything that seems to state whether or not what is being considered is really possible.
Is it possible to do any of these three things:
run both the JobManager and TaskManager as separate processes in the same container and have that serve as the Flink cluster, or
run the JobManager and TaskManager as literally the same process, or
run the job as a standalone JAR with the ability to recover from/take checkpoints and the ability to take a savepoint and restore from that savepoint?
(If anyone has any better ideas, I'm all ears.)
One of the responsibilities of the job manager is to monitor the task manager(s), and initiate restarts when failures have occurred. That works nicely in containerized environments when the JM and TMs are in separate containers; otherwise it seems like you're asking for trouble. Keeping the TMs separate also makes sense if you are ever going to scale up, though that may moot in your case.
What might be workable, though, would be to run the job using a LocalExecutionEnvironment (so that everything is in one process -- this is sometimes called a Flink minicluster). This path strikes me as feasible, if you're willing to work at it, but I can't recommend it. You'll have to somehow keep track of the checkpoints, and arrange for the container to be restarted from a checkpoint when things fail. And there are other things that may not work very well -- see this question for details. The LocalExecutionEnvironment wasn't designed with production deployments in mind.
What I'd suggest you explore instead is to see how far you can go toward making the standard, separate container solution affordable. For starters, you should be able to run the JM with minimal resources, since it doesn't have much to do.
Check this operator which automates the lifecycle of deploying and managing Flink in Kubernetes. The project is in beta but you can still get some idea about how to do it or directly use this operator if it fits your requirement. Here Job Manager and Task manager is separate kubernetes deployment.