Hi Anirban, > But this happened only once and now it is not getting reproduced at all.
This does make it sound a lot like https://issues.apache.org/jira/browse/FLINK-31632. > 1. What is the default watermarking strategy used in Flink. Can I quickly check the default parameters being used by calling some function or so ? Are you using the DataStream API and using the KafkaSource (and not the older FlinkKafkaConsumer connector)? If that's the case, I believe you always have to set a watermark strategy when adding the KafkaSource to the job topology, so there isn't a default. If you're using SQL, the strategy would be noWatermarks if not specified, which shouldn't let you bump into the bug. > 2. Are there any other conditions in which a source can be marked as Idle apart from the watermarking issue mentioned below ? That bug is the only known cause of incorrect permanent idleness that I'm aware of. > 3. If a Flink source is marked as Idle, is there any way to make it active without having to re-submit the Flink Job ? With the currently supported watermark strategies, the only way for a source subtask to resume activeness is when it reports a new watermark (e.g. when new data is produced). Best, Gordon On Fri, Jun 16, 2023 at 7:10 AM Anirban Dutta Gupta < anir...@indicussoftware.com> wrote: > Hello All, > > Sorry to be replying to an existing thread for my question. Actually we > are also facing the issue of the Flink Kafka source stopping consuming > messages completely. > It only started consuming messages after we re-submitted the Job. But this > happened only once and now it is not getting reproduced at all. > We are not using any watermarking strategy in specific. > > I have a few questions: > 1. What is the default watermarking strategy used in Flink. Can I quickly > check the default parameters being used by calling some function or so ? > 2. Are there any other conditions in which a source can be marked as Idle > apart from the watermarking issue mentioned below ? > 3. If a Flink source is marked as Idle, is there any way to make it active > without having to re-submit the Flink Job ? > Or Is it that the source automatically becomes active after a certain > duration ? > > Many thanks in advance, > Anirban > > On 16-06-2023 02:27, Ken Krugler wrote: > > I think you’re hitting this issue: > > https://issues.apache.org/jira/browse/FLINK-31632 > > Fixed in 1.16.2, 1.171. > > — Ken > > > On Jun 15, 2023, at 1:39 PM, Piotr Domagalski <pi...@domagalski.com> > wrote: > > Hi all! > > We've been experimenting with watermark alignment in Flink 1.15 and > observed an odd behaviour that I couldn't find any mention of in the > documentation. > > With the following strategy: > > WatermarkStrategy.<Event>forBoundedOutOfOrderness(Duration.ofSeconds(60)) > .withTimestampAssigner((e, t) -> e.timestamp) .withIdleness(Duration. > ofSeconds(3600)) .withWatermarkAlignment("group-1", Duration.ofSeconds(15 > )); > > Kafka sources stop consuming completely after 3600s (even when the data is > flowing into all the partitions). Is this an expected behaviour? Where > could I find more information on this? > > -- > Piotr Domagalski > > > -------------------------- > Ken Krugler > http://www.scaleunlimited.com > Custom big data solutions > Flink, Pinot, Solr, Elasticsearch > > > > >
