Thanks, Till. There are many reasons to reduce the heartbeat interval and timeout. But I am not sure what values are suitable. In our cases, the GC time and big job can be related factors. Since most flink jobs are pipeline and a total failover can cost some time, we should tolerate some stop-world situations. Also, I think that the FLINK-23216 should be solved to detect lost container fast and react to it. For my side, I suggest reducing the values gradually.
Till Rohrmann <trohrm...@apache.org> 于2021年7月22日周四 下午5:33写道： > Thanks for your inputs Gen and Arnaud. > > I do agree with you, Gen, that we need better guidance for our users on > when to change the heartbeat configuration. I think this should happen in > any case. I am, however, not so sure whether we can give hard threshold > like 5000 tasks, for example, because as Arnaud said it strongly depends on > the workload. Maybe we can explain it based on symptoms a user might > experience and what to do then. > > Concerning your workloads, Arnaud, I'd be interested to learn a bit more. > The user code runs in its own thread. This means that its operation won't > block the main thread/heartbeat. The only thing that can happen is that the > user code starves the heartbeat in terms of CPU cycles or causes a lot of > GC pauses. If you are observing the former problem, then we might think > about changing the priorities of the respective threads. This should then > improve Flink's stability for these workloads and a shorter heartbeat > timeout should be possible. > > Also for the RAM-cached repositories, what exactly is causing the heartbeat > to time out? Is it because you have a lot of GC or that the heartbeat > thread does not get enough CPU cycles? > > Cheers, > Till > > On Thu, Jul 22, 2021 at 9:16 AM LINZ, Arnaud <al...@bouyguestelecom.fr> > wrote: > > > Hello, > > > > > > > > From a user perspective: we have some (rare) use cases where we use > > “coarse grain” datasets, with big beans and tasks that do lengthy > operation > > (such as ML training). In these cases we had to increase the time out to > > huge values (heartbeat.timeout: 500000) so that our app is not killed. > > > > I’m aware this is not the way Flink was meant to be used, but it’s a > > convenient way to distribute our workload on datanodes without having to > > use another concurrency framework (such as M/R) that would require the > > recoding of sources and sinks. > > > > > > > > In some other (most common) cases, our tasks do some R/W accesses to > > RAM-cached repositories backed by a key-value storage such as Kudu (or > > Hbase). If most of those calls are very fast, sometimes when the system > is > > under heavy load they may block more than a few seconds, and having our > app > > killed because of a short timeout is not an option. > > > > > > > > That’s why I’m not in favor of very short timeouts… Because in my > > experience it really depends on what user code does in the tasks. (I > > understand that normally, as user code is not a JVM-blocking activity > such > > as a GC, it should have no impact on heartbeats, but from experience, it > > really does) > > > > > > > > Cheers, > > > > Arnaud > > > > > > > > > > > > *De :* Gen Luo <luogen...@gmail.com> > > *Envoyé :* jeudi 22 juillet 2021 05:46 > > *À :* Till Rohrmann <trohrm...@apache.org> > > *Cc :* Yang Wang <danrtsey...@gmail.com>; dev <dev@flink.apache.org>; > > user <u...@flink.apache.org> > > *Objet :* Re: [DISCUSS] FLIP-185: Shorter heartbeat timeout and interval > > default values > > > > > > > > Hi, > > > > Thanks for driving this @Till Rohrmann <trohrm...@apache.org> . I would > > give +1 on reducing the heartbeat timeout and interval, though I'm not > sure > > if 15s and 3s would be enough either. > > > > > > > > IMO, except for the standalone cluster, where the heartbeat mechanism in > > Flink is totally relied, reducing the heartbeat can also help JM to find > > out faster TaskExecutors in abnormal conditions that can not respond to > the > > heartbeat requests, e.g., continuously Full GC, though the process of > > TaskExecutor is alive and may not be known by the deployment system. > Since > > there are cases that can benefit from this change, I think it could be > done > > if it won't break the experience in other scenarios. > > > > > > > > If we can address what will block the main threads from processing > > heartbeats, or enlarge the GC costs, we can try to get rid of them to > have > > a more predictable response time of heartbeat, or give some advices to > > users if their jobs may encounter these issues. For example, as far as I > > know JM of a large scale job will be more busy and may not able to > process > > heartbeats in time, then we can give a advice that users working with job > > large than 5000 tasks should enlarge there heartbeat interval to 10s and > > timeout to 50s. The numbers are written casually. > > > > > > > > As for the issue in FLINK-23216, I think it should be fixed and may not > be > > a main concern for this case. > > > > > > > > On Wed, Jul 21, 2021 at 6:26 PM Till Rohrmann <trohrm...@apache.org> > > wrote: > > > > Thanks for sharing these insights. > > > > > > > > I think it is no longer true that the ResourceManager notifies the > > JobMaster about lost TaskExecutors. See FLINK-23216 [1] for more details. > > > > > > > > Given the GC pauses, would you then be ok with decreasing the heartbeat > > timeout to 20 seconds? This should give enough time to do the GC and then > > still send/receive a heartbeat request. > > > > > > > > I also wanted to add that we are about to get rid of one big cause of > > blocking I/O operations from the main thread. With FLINK-22483 [2] we > will > > get rid of Filesystem accesses to retrieve completed checkpoints. This > > leaves us with one additional file system access from the main thread > which > > is the one completing a pending checkpoint. I think it should be possible > > to get rid of this access because as Stephan said it only writes > > information to disk that is already written before. Maybe solving these > two > > issues could ease concerns about long pauses of unresponsiveness of > Flink. > > > > > > > > [1] https://issues.apache.org/jira/browse/FLINK-23216 > > > > [2] https://issues.apache.org/jira/browse/FLINK-22483 > > > > > > > > Cheers, > > > > Till > > > > > > > > On Wed, Jul 21, 2021 at 4:58 AM Yang Wang <danrtsey...@gmail.com> wrote: > > > > Thanks @Till Rohrmann <trohrm...@apache.org> for starting this > discussion > > > > > > > > Firstly, I try to understand the benefit of shorter heartbeat timeout. > > IIUC, it will make the JobManager aware of > > > > TaskManager faster. However, it seems that only the standalone cluster > > could benefit from this. For Yarn and > > > > native Kubernetes deployment, the Flink ResourceManager should get the > > TaskManager lost event in a very short time. > > > > > > > > * About 8 seconds, 3s for Yarn NM -> Yarn RM, 5s for Yarn RM -> Flink RM > > > > * Less than 1 second, Flink RM has a watch for all the TaskManager pods > > > > > > > > Secondly, I am not very confident to decrease the timeout to 15s. I have > > quickly checked the TaskManager GC logs > > > > in the past week of our internal Flink workloads and find more than 100 > > 10-seconds Full GC logs, but no one is bigger than 15s. > > > > We are using CMS GC for old generation. > > > > > > > > > > > > Best, > > > > Yang > > > > > > > > Till Rohrmann <trohrm...@apache.org> 于2021年7月17日周六 上午1:05写道： > > > > Hi everyone, > > > > Since Flink 1.5 we have the same heartbeat timeout and interval default > > values that are defined as heartbeat.timeout: 50s and heartbeat.interval: > > 10s. These values were mainly chosen to compensate for lengthy GC pauses > > and blocking operations that were executed in the main threads of Flink's > > components. Since then, there were quite some advancements wrt the JVM's > > GCs and we also got rid of a lot of blocking calls that were executed in > > the main thread. Moreover, a long heartbeat.timeout causes long recovery > > times in case of a TaskManager loss because the system can only properly > > recover after the dead TaskManager has been removed from the scheduler. > > Hence, I wanted to propose to change the timeout and interval to: > > > > heartbeat.timeout: 15s > > heartbeat.interval: 3s > > > > Since there is no perfect solution that fits all use cases, I would > really > > like to hear from you what you think about it and how you configure these > > heartbeat options. Based on your experience we might actually come up > with > > better default values that allow us to be resilient but also to detect > > failed components fast. FLIP-185 can be found here [1]. > > > > [1] https://cwiki.apache.org/confluence/x/GAoBCw > > > > Cheers, > > Till > > > > > > ------------------------------ > > > > L'intégrité de ce message n'étant pas assurée sur internet, la société > > expéditrice ne peut être tenue responsable de son contenu ni de ses > pièces > > jointes. Toute utilisation ou diffusion non autorisée est interdite. Si > > vous n'êtes pas destinataire de ce message, merci de le détruire et > > d'avertir l'expéditeur. > > > > The integrity of this message cannot be guaranteed on the Internet. The > > company that sent this message cannot therefore be held liable for its > > content nor attachments. Any unauthorized use or dissemination is > > prohibited. If you are not the intended recipient of this message, then > > please delete it and notify the sender. > > >
