[jira] [Commented] (KAFKA-3806) Adjust default values of log.retention.hours and offsets.retention.minutes
[ https://issues.apache.org/jira/browse/KAFKA-3806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16000185#comment-16000185 ] Di Shang commented on KAFKA-3806: - This is really problematic for mirrormaker since it manages offset manually (no auto offset commit). We have mirrormaker setup with 'auto.offset.reset=earliest' because we want to replicate the entire topic from beginning, what we see is that if a partition received no new messages for 1 day, mirrormaker will re-replicate everything from that partition up to 6 days ago causing a lot of unnecessary duplicate msg. I don't see the point having default offsets.retention < log.retention. In general if a consumer go offline for 1 day, then when it comes back the default configuration can make it (depending on auto.offset.reset) either reprocess all messages up to 6 days ago or skip messages during the offline period. It's a trap likely to cause issues in many scenarios that will catch people until they realize that they have to configure offsets.retention >= log.retention explicitly, so why not make it the default anyway. > Adjust default values of log.retention.hours and offsets.retention.minutes > -- > > Key: KAFKA-3806 > URL: https://issues.apache.org/jira/browse/KAFKA-3806 > Project: Kafka > Issue Type: Improvement > Components: config >Affects Versions: 0.9.0.1, 0.10.0.0 >Reporter: Michal Turek >Priority: Minor > > Combination of default values of log.retention.hours (168 hours = 7 days) and > offsets.retention.minutes (1440 minutes = 1 day) may be dangerous in special > cases. Offset retention should be always greater than log retention. > We have observed the following scenario and issue: > - Producing of data to a topic was disabled two days ago by producer update, > topic wasn't deleted. > - Consumer consumed all data and properly committed offsets to Kafka. > - Consumer made no more offset commits for that topic because there was no > more incoming data and there was nothing to confirm. (We have auto-commit > disabled, I'm not sure how behaves enabled auto-commit.) > - After one day: Kafka cleared too old offsets according to > offsets.retention.minutes. > - After two days: Long-term running consumer was restarted after update, it > didn't find any committed offsets for that topic since they were deleted by > offsets.retention.minutes so it started consuming from the beginning. > - The messages were still in Kafka due to larger log.retention.hours, about 5 > days of messages were read again. > Known workaround to solve this issue: > - Explicitly configure log.retention.hours and offsets.retention.minutes, > don't use defaults. > Proposals: > - Prolong default value of offsets.retention.minutes to be at least twice > larger than log.retention.hours. > - Check these values during Kafka startup and log a warning if > offsets.retention.minutes is smaller than log.retention.hours. > - Add a note to migration guide about differences between storing of offsets > in ZooKeeper and Kafka (http://kafka.apache.org/documentation.html#upgrade). -- This message was sent by Atlassian JIRA (v6.3.15#6346)
[jira] [Commented] (KAFKA-3806) Adjust default values of log.retention.hours and offsets.retention.minutes
[ https://issues.apache.org/jira/browse/KAFKA-3806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15906349#comment-15906349 ] Eric Bolinger commented on KAFKA-3806: -- People learn of the interaction between log.retention.hours and offsets.retention.minutes through experience rather than documentation. Then they search the internet for clues until they come upon this ticket. _If the Kafka cluster sets both properties to the same duration_ (accounting for hours vs minutes), then you have a stable solution. Any client that always starts with `auto.offset.reset=earliest` will behave consistently. When the offsets retention period is less than log retention, you have a window where the client behavior is unstable. The absolute values are subjective and vary for different use cases. The scenario where multiple people are running intermittent console consumers points to a need for the client to override the server offsets retention period. If the console consumer specifies a new consumer group every time, then it should also specify a short offset.retention.minutes. The solution where clients periodically commit offsets is just poor design. The offsets.retention period is managed by the cluster while the clients (plural) must individually manage their own update intervals. What happens when the values change? What are the trade-offs? Why is the client doing work to manage memory in the cluster nodes? Count the leaking abstractions here... > Adjust default values of log.retention.hours and offsets.retention.minutes > -- > > Key: KAFKA-3806 > URL: https://issues.apache.org/jira/browse/KAFKA-3806 > Project: Kafka > Issue Type: Improvement > Components: config >Affects Versions: 0.9.0.1, 0.10.0.0 >Reporter: Michal Turek >Priority: Minor > > Combination of default values of log.retention.hours (168 hours = 7 days) and > offsets.retention.minutes (1440 minutes = 1 day) may be dangerous in special > cases. Offset retention should be always greater than log retention. > We have observed the following scenario and issue: > - Producing of data to a topic was disabled two days ago by producer update, > topic wasn't deleted. > - Consumer consumed all data and properly committed offsets to Kafka. > - Consumer made no more offset commits for that topic because there was no > more incoming data and there was nothing to confirm. (We have auto-commit > disabled, I'm not sure how behaves enabled auto-commit.) > - After one day: Kafka cleared too old offsets according to > offsets.retention.minutes. > - After two days: Long-term running consumer was restarted after update, it > didn't find any committed offsets for that topic since they were deleted by > offsets.retention.minutes so it started consuming from the beginning. > - The messages were still in Kafka due to larger log.retention.hours, about 5 > days of messages were read again. > Known workaround to solve this issue: > - Explicitly configure log.retention.hours and offsets.retention.minutes, > don't use defaults. > Proposals: > - Prolong default value of offsets.retention.minutes to be at least twice > larger than log.retention.hours. > - Check these values during Kafka startup and log a warning if > offsets.retention.minutes is smaller than log.retention.hours. > - Add a note to migration guide about differences between storing of offsets > in ZooKeeper and Kafka (http://kafka.apache.org/documentation.html#upgrade). -- This message was sent by Atlassian JIRA (v6.3.15#6346)
[jira] [Commented] (KAFKA-3806) Adjust default values of log.retention.hours and offsets.retention.minutes
[ https://issues.apache.org/jira/browse/KAFKA-3806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15838918#comment-15838918 ] Jeff Widman commented on KAFKA-3806: Too often people not super familiar with Kafka don't realize the default is so low... So I would rather see this default to 4 days. If all processes in a consumer group randomly blow out Friday late afternoon and it's a low-priority consumer group, so ops/dev team decides to wait til Monday to fix the problem, thinking that the consumer will just catch back up when they get it fixed, then they'll be surprised to learn their offsets were hosed. For those companies big enough to have a performance hit from maintaining offsets that long, they will generally have the in-house resources to realize they should reduce this value. > Adjust default values of log.retention.hours and offsets.retention.minutes > -- > > Key: KAFKA-3806 > URL: https://issues.apache.org/jira/browse/KAFKA-3806 > Project: Kafka > Issue Type: Improvement > Components: config >Affects Versions: 0.9.0.1, 0.10.0.0 >Reporter: Michal Turek >Priority: Minor > > Combination of default values of log.retention.hours (168 hours = 7 days) and > offsets.retention.minutes (1440 minutes = 1 day) may be dangerous in special > cases. Offset retention should be always greater than log retention. > We have observed the following scenario and issue: > - Producing of data to a topic was disabled two days ago by producer update, > topic wasn't deleted. > - Consumer consumed all data and properly committed offsets to Kafka. > - Consumer made no more offset commits for that topic because there was no > more incoming data and there was nothing to confirm. (We have auto-commit > disabled, I'm not sure how behaves enabled auto-commit.) > - After one day: Kafka cleared too old offsets according to > offsets.retention.minutes. > - After two days: Long-term running consumer was restarted after update, it > didn't find any committed offsets for that topic since they were deleted by > offsets.retention.minutes so it started consuming from the beginning. > - The messages were still in Kafka due to larger log.retention.hours, about 5 > days of messages were read again. > Known workaround to solve this issue: > - Explicitly configure log.retention.hours and offsets.retention.minutes, > don't use defaults. > Proposals: > - Prolong default value of offsets.retention.minutes to be at least twice > larger than log.retention.hours. > - Check these values during Kafka startup and log a warning if > offsets.retention.minutes is smaller than log.retention.hours. > - Add a note to migration guide about differences between storing of offsets > in ZooKeeper and Kafka (http://kafka.apache.org/documentation.html#upgrade). -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (KAFKA-3806) Adjust default values of log.retention.hours and offsets.retention.minutes
[ https://issues.apache.org/jira/browse/KAFKA-3806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15324020#comment-15324020 ] Michal Turek commented on KAFKA-3806: - Hi, thanks for the advises. We will workaround the issue by prolonging offsets.retention.minutes, it seems safe in our environment. > Adjust default values of log.retention.hours and offsets.retention.minutes > -- > > Key: KAFKA-3806 > URL: https://issues.apache.org/jira/browse/KAFKA-3806 > Project: Kafka > Issue Type: Improvement > Components: config >Affects Versions: 0.9.0.1, 0.10.0.0 >Reporter: Michal Turek >Priority: Minor > > Combination of default values of log.retention.hours (168 hours = 7 days) and > offsets.retention.minutes (1440 minutes = 1 day) may be dangerous in special > cases. Offset retention should be always greater than log retention. > We have observed the following scenario and issue: > - Producing of data to a topic was disabled two days ago by producer update, > topic wasn't deleted. > - Consumer consumed all data and properly committed offsets to Kafka. > - Consumer made no more offset commits for that topic because there was no > more incoming data and there was nothing to confirm. (We have auto-commit > disabled, I'm not sure how behaves enabled auto-commit.) > - After one day: Kafka cleared too old offsets according to > offsets.retention.minutes. > - After two days: Long-term running consumer was restarted after update, it > didn't find any committed offsets for that topic since they were deleted by > offsets.retention.minutes so it started consuming from the beginning. > - The messages were still in Kafka due to larger log.retention.hours, about 5 > days of messages were read again. > Known workaround to solve this issue: > - Explicitly configure log.retention.hours and offsets.retention.minutes, > don't use defaults. > Proposals: > - Prolong default value of offsets.retention.minutes to be at least twice > larger than log.retention.hours. > - Check these values during Kafka startup and log a warning if > offsets.retention.minutes is smaller than log.retention.hours. > - Add a note to migration guide about differences between storing of offsets > in ZooKeeper and Kafka (http://kafka.apache.org/documentation.html#upgrade). -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (KAFKA-3806) Adjust default values of log.retention.hours and offsets.retention.minutes
[ https://issues.apache.org/jira/browse/KAFKA-3806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15323527#comment-15323527 ] James Cheng commented on KAFKA-3806: One impact, which I don't know if it's a large issue, is that the most-recent-offset for a given consumer group is stored in memory in the broker. There is a process in the broker that continually reads the __consumer_offsets topic, and stores them into an in-memory structure, so that when a consumer group starts up, the broker can efficiently tell the consumer what its most recent offset was. If you set the offsets.retention.minutes to a very large value, you will end up with fewer garbage collections, which means the in-memory aggregate size will be larger. > Adjust default values of log.retention.hours and offsets.retention.minutes > -- > > Key: KAFKA-3806 > URL: https://issues.apache.org/jira/browse/KAFKA-3806 > Project: Kafka > Issue Type: Improvement > Components: config >Affects Versions: 0.9.0.1, 0.10.0.0 >Reporter: Michal Turek >Priority: Minor > > Combination of default values of log.retention.hours (168 hours = 7 days) and > offsets.retention.minutes (1440 minutes = 1 day) may be dangerous in special > cases. Offset retention should be always greater than log retention. > We have observed the following scenario and issue: > - Producing of data to a topic was disabled two days ago by producer update, > topic wasn't deleted. > - Consumer consumed all data and properly committed offsets to Kafka. > - Consumer made no more offset commits for that topic because there was no > more incoming data and there was nothing to confirm. (We have auto-commit > disabled, I'm not sure how behaves enabled auto-commit.) > - After one day: Kafka cleared too old offsets according to > offsets.retention.minutes. > - After two days: Long-term running consumer was restarted after update, it > didn't find any committed offsets for that topic since they were deleted by > offsets.retention.minutes so it started consuming from the beginning. > - The messages were still in Kafka due to larger log.retention.hours, about 5 > days of messages were read again. > Known workaround to solve this issue: > - Explicitly configure log.retention.hours and offsets.retention.minutes, > don't use defaults. > Proposals: > - Prolong default value of offsets.retention.minutes to be at least twice > larger than log.retention.hours. > - Check these values during Kafka startup and log a warning if > offsets.retention.minutes is smaller than log.retention.hours. > - Add a note to migration guide about differences between storing of offsets > in ZooKeeper and Kafka (http://kafka.apache.org/documentation.html#upgrade). -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (KAFKA-3806) Adjust default values of log.retention.hours and offsets.retention.minutes
[ https://issues.apache.org/jira/browse/KAFKA-3806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15323521#comment-15323521 ] Jun Rao commented on KAFKA-3806: [~dwatzke], I think the original intention is that consumers in Kafka are typically real time. If a consumer is down for more than a day, chances are that it will never come back again and therefore it's offsets won't be needed. [~tu...@avast.com], it's true that for a given consumer-group,topic,partition, after compaction, there will be only one message stored in the offset topic eventually. What we wanted to protect against is lots of short lived consumer groups. For example, currently, each time when one runs tools like console-consumer, a new consumer group is created. In an environment where many people are running those tools for a short period of time, it would be useful to garbage collect the unused offsets. If you know this is not the case, you can increase offsets.retention.minutes. Yes, ideally, perhaps we should only start the expiration clock once the consumer becomes inactive, instead when an offset is last committed. For example, one simple improvement is if there is no new update to the offset but the consumer is still alive, we can automatically write the last offset with a new expiration time. This will address [~wushujames]'s issue better since the consumer no longer has to recommit the same offset. > Adjust default values of log.retention.hours and offsets.retention.minutes > -- > > Key: KAFKA-3806 > URL: https://issues.apache.org/jira/browse/KAFKA-3806 > Project: Kafka > Issue Type: Improvement > Components: config >Affects Versions: 0.9.0.1, 0.10.0.0 >Reporter: Michal Turek >Priority: Minor > > Combination of default values of log.retention.hours (168 hours = 7 days) and > offsets.retention.minutes (1440 minutes = 1 day) may be dangerous in special > cases. Offset retention should be always greater than log retention. > We have observed the following scenario and issue: > - Producing of data to a topic was disabled two days ago by producer update, > topic wasn't deleted. > - Consumer consumed all data and properly committed offsets to Kafka. > - Consumer made no more offset commits for that topic because there was no > more incoming data and there was nothing to confirm. (We have auto-commit > disabled, I'm not sure how behaves enabled auto-commit.) > - After one day: Kafka cleared too old offsets according to > offsets.retention.minutes. > - After two days: Long-term running consumer was restarted after update, it > didn't find any committed offsets for that topic since they were deleted by > offsets.retention.minutes so it started consuming from the beginning. > - The messages were still in Kafka due to larger log.retention.hours, about 5 > days of messages were read again. > Known workaround to solve this issue: > - Explicitly configure log.retention.hours and offsets.retention.minutes, > don't use defaults. > Proposals: > - Prolong default value of offsets.retention.minutes to be at least twice > larger than log.retention.hours. > - Check these values during Kafka startup and log a warning if > offsets.retention.minutes is smaller than log.retention.hours. > - Add a note to migration guide about differences between storing of offsets > in ZooKeeper and Kafka (http://kafka.apache.org/documentation.html#upgrade). -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (KAFKA-3806) Adjust default values of log.retention.hours and offsets.retention.minutes
[ https://issues.apache.org/jira/browse/KAFKA-3806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15322097#comment-15322097 ] Michal Turek commented on KAFKA-3806: - Hi Jun and James, I probably don't see all the assumptions and consequences but I feel there is something wrong. I consider the committed offsets to be only tiny metadata for huge log data. Each committed offset is in its nature only one single number + identification of topic-partition + identification of consumer group. It was exactly this when it was stored in ZooKeeper. The new approach - storing of the offsets by writing to a special Kafka topic is "only" an implementation detail :-). The topic may store a lot of subsequent commits, but each offset commit invalidate and fully overwrite all previous ones. I thought the Kafka's log compaction feature (https://cwiki.apache.org/confluence/display/KAFKA/Log+Compaction) apply here and only the last commit will survive the compaction, so there is nearly no storage overhead long term. Am I wrong? Please correct me if so. Default log.retention.hours = 7 days is pretty fine, but I would expect default for offsets.retention.minutes to be half a year or so. Remember it is basically only reasonably small group of tiny numbers. Can you explain me, what is the reason to have offsets.retention.minutes so small, only 1 day? What will be the consequences if we configure it to one month or one year? Will something wrong happen? I feel it's obvious for you but I don't see anything. I fully agree there should be some TTL expiration for very old "dead" values to be able to GC them and free resources. Even the tiny metadata may grow in time. But "one day" doesn't belong to the "very old" category for me at all. If prolonging offsets.retention.minutes was dangerous, would it be possible to prevent deletion of the committed offsets in case that the topic still exists and the consumer group is active or was active during offsets.retention.minutes timeout? I don't know Kafka code, but I would expect behavior like this to reliably prevent any (meta)data loss. Thanks for explanation! Michal > Adjust default values of log.retention.hours and offsets.retention.minutes > -- > > Key: KAFKA-3806 > URL: https://issues.apache.org/jira/browse/KAFKA-3806 > Project: Kafka > Issue Type: Improvement > Components: config >Affects Versions: 0.9.0.1, 0.10.0.0 >Reporter: Michal Turek >Priority: Minor > > Combination of default values of log.retention.hours (168 hours = 7 days) and > offsets.retention.minutes (1440 minutes = 1 day) may be dangerous in special > cases. Offset retention should be always greater than log retention. > We have observed the following scenario and issue: > - Producing of data to a topic was disabled two days ago by producer update, > topic wasn't deleted. > - Consumer consumed all data and properly committed offsets to Kafka. > - Consumer made no more offset commits for that topic because there was no > more incoming data and there was nothing to confirm. (We have auto-commit > disabled, I'm not sure how behaves enabled auto-commit.) > - After one day: Kafka cleared too old offsets according to > offsets.retention.minutes. > - After two days: Long-term running consumer was restarted after update, it > didn't find any committed offsets for that topic since they were deleted by > offsets.retention.minutes so it started consuming from the beginning. > - The messages were still in Kafka due to larger log.retention.hours, about 5 > days of messages were read again. > Known workaround to solve this issue: > - Explicitly configure log.retention.hours and offsets.retention.minutes, > don't use defaults. > Proposals: > - Prolong default value of offsets.retention.minutes to be at least twice > larger than log.retention.hours. > - Check these values during Kafka startup and log a warning if > offsets.retention.minutes is smaller than log.retention.hours. > - Add a note to migration guide about differences between storing of offsets > in ZooKeeper and Kafka (http://kafka.apache.org/documentation.html#upgrade). -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (KAFKA-3806) Adjust default values of log.retention.hours and offsets.retention.minutes
[ https://issues.apache.org/jira/browse/KAFKA-3806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15322076#comment-15322076 ] David Watzke commented on KAFKA-3806: - [~junrao] I understand the need for the automatic consumer group offsets cleanup but why would you insist on keeping the default so low? It seems like a very aggresive default that would complicate the code for everyone who's using kafka (by having to implement noop offset commits just to reset timers)... It's just I don't see a reason why would anybody prefer this over a saner default that would be closer to the default retention (time) value. Thanks > Adjust default values of log.retention.hours and offsets.retention.minutes > -- > > Key: KAFKA-3806 > URL: https://issues.apache.org/jira/browse/KAFKA-3806 > Project: Kafka > Issue Type: Improvement > Components: config >Affects Versions: 0.9.0.1, 0.10.0.0 >Reporter: Michal Turek >Priority: Minor > > Combination of default values of log.retention.hours (168 hours = 7 days) and > offsets.retention.minutes (1440 minutes = 1 day) may be dangerous in special > cases. Offset retention should be always greater than log retention. > We have observed the following scenario and issue: > - Producing of data to a topic was disabled two days ago by producer update, > topic wasn't deleted. > - Consumer consumed all data and properly committed offsets to Kafka. > - Consumer made no more offset commits for that topic because there was no > more incoming data and there was nothing to confirm. (We have auto-commit > disabled, I'm not sure how behaves enabled auto-commit.) > - After one day: Kafka cleared too old offsets according to > offsets.retention.minutes. > - After two days: Long-term running consumer was restarted after update, it > didn't find any committed offsets for that topic since they were deleted by > offsets.retention.minutes so it started consuming from the beginning. > - The messages were still in Kafka due to larger log.retention.hours, about 5 > days of messages were read again. > Known workaround to solve this issue: > - Explicitly configure log.retention.hours and offsets.retention.minutes, > don't use defaults. > Proposals: > - Prolong default value of offsets.retention.minutes to be at least twice > larger than log.retention.hours. > - Check these values during Kafka startup and log a warning if > offsets.retention.minutes is smaller than log.retention.hours. > - Add a note to migration guide about differences between storing of offsets > in ZooKeeper and Kafka (http://kafka.apache.org/documentation.html#upgrade). -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (KAFKA-3806) Adjust default values of log.retention.hours and offsets.retention.minutes
[ https://issues.apache.org/jira/browse/KAFKA-3806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15321771#comment-15321771 ] James Cheng commented on KAFKA-3806: Thanks Jun. If it's per topic/partition, then that means that the garbage collection of offsets is not only dependent on the consumer, it's also dependent on the partitioning strategy of the producer. Within a given topic, a hot partition is less likely to be incorrectly garbage collected, and a low-traffic partition is more likely to be incorrectly garbage collected. For now, it looks like the only way to mitigate this is on the consumer, by committing offsets even though no data is flowing. It might be more usable if garbage collection was took into account not just time-since-last-committed-offset but also time-since-last-produced-message. Although, there are some corner cases where that might not work. For example, what's the right thing to do if you have an active consumer to a topic, but where the producer has been turned off for a week? Not sure. > Adjust default values of log.retention.hours and offsets.retention.minutes > -- > > Key: KAFKA-3806 > URL: https://issues.apache.org/jira/browse/KAFKA-3806 > Project: Kafka > Issue Type: Improvement > Components: config >Affects Versions: 0.9.0.1, 0.10.0.0 >Reporter: Michal Turek >Priority: Minor > > Combination of default values of log.retention.hours (168 hours = 7 days) and > offsets.retention.minutes (1440 minutes = 1 day) may be dangerous in special > cases. Offset retention should be always greater than log retention. > We have observed the following scenario and issue: > - Producing of data to a topic was disabled two days ago by producer update, > topic wasn't deleted. > - Consumer consumed all data and properly committed offsets to Kafka. > - Consumer made no more offset commits for that topic because there was no > more incoming data and there was nothing to confirm. (We have auto-commit > disabled, I'm not sure how behaves enabled auto-commit.) > - After one day: Kafka cleared too old offsets according to > offsets.retention.minutes. > - After two days: Long-term running consumer was restarted after update, it > didn't find any committed offsets for that topic since they were deleted by > offsets.retention.minutes so it started consuming from the beginning. > - The messages were still in Kafka due to larger log.retention.hours, about 5 > days of messages were read again. > Known workaround to solve this issue: > - Explicitly configure log.retention.hours and offsets.retention.minutes, > don't use defaults. > Proposals: > - Prolong default value of offsets.retention.minutes to be at least twice > larger than log.retention.hours. > - Check these values during Kafka startup and log a warning if > offsets.retention.minutes is smaller than log.retention.hours. > - Add a note to migration guide about differences between storing of offsets > in ZooKeeper and Kafka (http://kafka.apache.org/documentation.html#upgrade). -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (KAFKA-3806) Adjust default values of log.retention.hours and offsets.retention.minutes
[ https://issues.apache.org/jira/browse/KAFKA-3806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15321710#comment-15321710 ] Jun Rao commented on KAFKA-3806: [~wushujames], committing offsets on every poll is likely too frequent and will add unnecessary load on the broker. So, committing offsets in some configured time interval will be better. The retention is done per topic/partition. If the last committed offset for a given topic/partition hasn't been updated for longer than offsets.retention.minutes, the offset will be garbage collected. > Adjust default values of log.retention.hours and offsets.retention.minutes > -- > > Key: KAFKA-3806 > URL: https://issues.apache.org/jira/browse/KAFKA-3806 > Project: Kafka > Issue Type: Improvement > Components: config >Affects Versions: 0.9.0.1, 0.10.0.0 >Reporter: Michal Turek >Priority: Minor > > Combination of default values of log.retention.hours (168 hours = 7 days) and > offsets.retention.minutes (1440 minutes = 1 day) may be dangerous in special > cases. Offset retention should be always greater than log retention. > We have observed the following scenario and issue: > - Producing of data to a topic was disabled two days ago by producer update, > topic wasn't deleted. > - Consumer consumed all data and properly committed offsets to Kafka. > - Consumer made no more offset commits for that topic because there was no > more incoming data and there was nothing to confirm. (We have auto-commit > disabled, I'm not sure how behaves enabled auto-commit.) > - After one day: Kafka cleared too old offsets according to > offsets.retention.minutes. > - After two days: Long-term running consumer was restarted after update, it > didn't find any committed offsets for that topic since they were deleted by > offsets.retention.minutes so it started consuming from the beginning. > - The messages were still in Kafka due to larger log.retention.hours, about 5 > days of messages were read again. > Known workaround to solve this issue: > - Explicitly configure log.retention.hours and offsets.retention.minutes, > don't use defaults. > Proposals: > - Prolong default value of offsets.retention.minutes to be at least twice > larger than log.retention.hours. > - Check these values during Kafka startup and log a warning if > offsets.retention.minutes is smaller than log.retention.hours. > - Add a note to migration guide about differences between storing of offsets > in ZooKeeper and Kafka (http://kafka.apache.org/documentation.html#upgrade). -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (KAFKA-3806) Adjust default values of log.retention.hours and offsets.retention.minutes
[ https://issues.apache.org/jira/browse/KAFKA-3806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15321422#comment-15321422 ] James Cheng commented on KAFKA-3806: Does the offsets.retention.minutes timer apply to a per topic basis? Per partition basis? To any topic in for the consumer group? That is, in the following scenario: * a consumer listening to a 2 partition topic * offsets.retention.minutes = 1 day * partition 1 has no new messages in over a day * partition 2 has a new message every couple minutes. Will the offset for partition 1 get deleted, but the offset for partition 2 remain present? Similarly, in the following scenario: * a consumer listening to a 2 topics * offsets.retention.minutes = 1 day * topic 1 has no new messages in over a day * topic 2 has a new message every couple minutes. Will the offset for topic 1 get deleted, but the offset for topic 2 remain present? -James > Adjust default values of log.retention.hours and offsets.retention.minutes > -- > > Key: KAFKA-3806 > URL: https://issues.apache.org/jira/browse/KAFKA-3806 > Project: Kafka > Issue Type: Improvement > Components: config >Affects Versions: 0.9.0.1, 0.10.0.0 >Reporter: Michal Turek >Priority: Minor > > Combination of default values of log.retention.hours (168 hours = 7 days) and > offsets.retention.minutes (1440 minutes = 1 day) may be dangerous in special > cases. Offset retention should be always greater than log retention. > We have observed the following scenario and issue: > - Producing of data to a topic was disabled two days ago by producer update, > topic wasn't deleted. > - Consumer consumed all data and properly committed offsets to Kafka. > - Consumer made no more offset commits for that topic because there was no > more incoming data and there was nothing to confirm. (We have auto-commit > disabled, I'm not sure how behaves enabled auto-commit.) > - After one day: Kafka cleared too old offsets according to > offsets.retention.minutes. > - After two days: Long-term running consumer was restarted after update, it > didn't find any committed offsets for that topic since they were deleted by > offsets.retention.minutes so it started consuming from the beginning. > - The messages were still in Kafka due to larger log.retention.hours, about 5 > days of messages were read again. > Known workaround to solve this issue: > - Explicitly configure log.retention.hours and offsets.retention.minutes, > don't use defaults. > Proposals: > - Prolong default value of offsets.retention.minutes to be at least twice > larger than log.retention.hours. > - Check these values during Kafka startup and log a warning if > offsets.retention.minutes is smaller than log.retention.hours. > - Add a note to migration guide about differences between storing of offsets > in ZooKeeper and Kafka (http://kafka.apache.org/documentation.html#upgrade). -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (KAFKA-3806) Adjust default values of log.retention.hours and offsets.retention.minutes
[
https://issues.apache.org/jira/browse/KAFKA-3806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15321377#comment-15321377
]
James Cheng commented on KAFKA-3806:
Jun,
Any recommendations on how to structure an application such that it commits
often enough, but not so much that it spams Kafka with committed offsets?
We had an application here recently that would do something like
{code}
messages = poll(100); // 100ms
process(messages)
commitOffsets();
{code}
If the topic had no new messages, we would end up committing offsets 10 times a
second. We had 38 instances of this application, each having multiple threads,
processing a 12-partition topic. We ended up with 12,000 messages per second
being written to the __consumer_offsets topic. The aggregate size was only
2MB/sec tho.
On the one hand, Kafka was able to absorb this load fine, and as you say, we
don't need to worry about offsets going away. On the other hand, it seems
wasteful.
Any thoughts, or is that just the necessary tradeoff? If we want to spam the
logs less often, we could change our applications to:
{code}
messages = poll(100); // 100ms
process(messages)
if (messages.length > 0 || enoughTimeHasPassedSinceLastCommit) {
commitOffsets();
}
{code}
That code isn't that difficult, but it is also logic that we would have to
repeat in all of our consumers.
> Adjust default values of log.retention.hours and offsets.retention.minutes
> --
>
> Key: KAFKA-3806
> URL: https://issues.apache.org/jira/browse/KAFKA-3806
> Project: Kafka
> Issue Type: Improvement
> Components: config
>Affects Versions: 0.9.0.1, 0.10.0.0
>Reporter: Michal Turek
>Priority: Minor
>
> Combination of default values of log.retention.hours (168 hours = 7 days) and
> offsets.retention.minutes (1440 minutes = 1 day) may be dangerous in special
> cases. Offset retention should be always greater than log retention.
> We have observed the following scenario and issue:
> - Producing of data to a topic was disabled two days ago by producer update,
> topic wasn't deleted.
> - Consumer consumed all data and properly committed offsets to Kafka.
> - Consumer made no more offset commits for that topic because there was no
> more incoming data and there was nothing to confirm. (We have auto-commit
> disabled, I'm not sure how behaves enabled auto-commit.)
> - After one day: Kafka cleared too old offsets according to
> offsets.retention.minutes.
> - After two days: Long-term running consumer was restarted after update, it
> didn't find any committed offsets for that topic since they were deleted by
> offsets.retention.minutes so it started consuming from the beginning.
> - The messages were still in Kafka due to larger log.retention.hours, about 5
> days of messages were read again.
> Known workaround to solve this issue:
> - Explicitly configure log.retention.hours and offsets.retention.minutes,
> don't use defaults.
> Proposals:
> - Prolong default value of offsets.retention.minutes to be at least twice
> larger than log.retention.hours.
> - Check these values during Kafka startup and log a warning if
> offsets.retention.minutes is smaller than log.retention.hours.
> - Add a note to migration guide about differences between storing of offsets
> in ZooKeeper and Kafka (http://kafka.apache.org/documentation.html#upgrade).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (KAFKA-3806) Adjust default values of log.retention.hours and offsets.retention.minutes
[ https://issues.apache.org/jira/browse/KAFKA-3806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15320931#comment-15320931 ] Jun Rao commented on KAFKA-3806: offsets.retention.minutes is designed to handle the case that a consumer group goes away forever. In that case, we don't want to store the offsets for that group forever. That retention time potentially can be less than the log retention time. In your case, if the consumer is still alive, you would want to keep committing the offsets even though they are not changing. Once the consumer group coordinator sees the new offset commit, it will extend the time to preserve the offsets. This is what auto commit will do. > Adjust default values of log.retention.hours and offsets.retention.minutes > -- > > Key: KAFKA-3806 > URL: https://issues.apache.org/jira/browse/KAFKA-3806 > Project: Kafka > Issue Type: Improvement > Components: config >Affects Versions: 0.9.0.1, 0.10.0.0 >Reporter: Michal Turek >Priority: Minor > > Combination of default values of log.retention.hours (168 hours = 7 days) and > offsets.retention.minutes (1440 minutes = 1 day) may be dangerous in special > cases. Offset retention should be always greater than log retention. > We have observed the following scenario and issue: > - Producing of data to a topic was disabled two days ago by producer update, > topic wasn't deleted. > - Consumer consumed all data and properly committed offsets to Kafka. > - Consumer made no more offset commits for that topic because there was no > more incoming data and there was nothing to confirm. (We have auto-commit > disabled, I'm not sure how behaves enabled auto-commit.) > - After one day: Kafka cleared too old offsets according to > offsets.retention.minutes. > - After two days: Long-term running consumer was restarted after update, it > didn't find any committed offsets for that topic since they were deleted by > offsets.retention.minutes so it started consuming from the beginning. > - The messages were still in Kafka due to larger log.retention.hours, about 5 > days of messages were read again. > Known workaround to solve this issue: > - Explicitly configure log.retention.hours and offsets.retention.minutes, > don't use defaults. > Proposals: > - Prolong default value of offsets.retention.minutes to be at least twice > larger than log.retention.hours. > - Check these values during Kafka startup and log a warning if > offsets.retention.minutes is smaller than log.retention.hours. > - Add a note to migration guide about differences between storing of offsets > in ZooKeeper and Kafka (http://kafka.apache.org/documentation.html#upgrade). -- This message was sent by Atlassian JIRA (v6.3.4#6332)
