[jira] [Commented] (SLING-4533) OakResourceListener does not allow to configure the OAK observation queue length
[ https://issues.apache.org/jira/browse/SLING-4533?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14379612#comment-14379612 ] Marc Pfaff commented on SLING-4533: --- bq. Am I correct that increasing the queue size will only move the threshold at which the problem occurs? I would say so, yes. I don't necessarily see increasing the queue size as the solution to the problem. But it's IMHO still useful to have this possibility for troubleshooting and as an additional option for tuning. Especially as the Oak default size is only 1000. bq. Is there a loud warning from Oak when that threshold is hit? Yes and no. There is the following warning in the log: {code} 24.03.2015 12:22:03.977 *WARN* [qtp699308246-690] org.apache.jackrabbit.oak.jcr.observation.ChangeProcessor Revision queue is full. Further revisions will be compacted. {code} But that is coming from the Oak ChangeProcessor and AFAIK this one can have a different, as well configurable, queue size as per org.apache.jackrabbit.oak.jcr.repository.RepositoryImpl. So I would say yes there is a warning, but IMO that only makes sense if both, the Oak ChangeProcessor and the Sling OakResourceListener use the same queue length value. OakResourceListener does not allow to configure the OAK observation queue length Key: SLING-4533 URL: https://issues.apache.org/jira/browse/SLING-4533 Project: Sling Issue Type: Bug Components: JCR, Oak Affects Versions: JCR Resource 2.5.0 Reporter: Marc Pfaff Currently the OakResourceListener does not allow to configure the OAK observation queue length when instantiating the OAK BackgroundObserver. Thus the default queue size of 1000 is used. If the max queue size is reached, OAK seem's to turn the observation in a 'graceful degratation' mode, which leads to a compacted view of observations. Means, not every single observation makes it to the observer no more. From the point of view of the Sling resource changed events, this looks like some events are missing. I would propose to make the OAK observation queue size configurable in order to have the possibility to avoid 'graceful degratation'. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (SLING-4533) OakResourceListener does not allow to configure the OAK observation queue length
[ https://issues.apache.org/jira/browse/SLING-4533?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14379979#comment-14379979 ] Marc Pfaff commented on SLING-4533: --- bq. If you have ideas about how that can be implemented it would be good to create an Oak ticket. I created OAK-2680 accordingly. OakResourceListener does not allow to configure the OAK observation queue length Key: SLING-4533 URL: https://issues.apache.org/jira/browse/SLING-4533 Project: Sling Issue Type: Bug Components: JCR, Oak Affects Versions: JCR Resource 2.5.0 Reporter: Marc Pfaff Attachments: SLING-4533.patch Currently the OakResourceListener does not allow to configure the OAK observation queue length when instantiating the OAK BackgroundObserver. Thus the default queue size of 1000 is used. If the max queue size is reached, OAK seem's to turn the observation in a 'graceful degratation' mode, which leads to a compacted view of observations. Means, not every single observation makes it to the observer no more. From the point of view of the Sling resource changed events, this looks like some events are missing. I would propose to make the OAK observation queue size configurable in order to have the possibility to avoid 'graceful degratation'. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (SLING-4533) OakResourceListener does not allow to configure the OAK observation queue length
[ https://issues.apache.org/jira/browse/SLING-4533?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14379756#comment-14379756 ] Bertrand Delacretaz commented on SLING-4533: Ok thanks for the info - having a reliable warning if events might be lost is essential IMO as those are very tricky things to troubleshoot. If you have ideas about how that can be implemented it would be good to create an Oak ticket. OakResourceListener does not allow to configure the OAK observation queue length Key: SLING-4533 URL: https://issues.apache.org/jira/browse/SLING-4533 Project: Sling Issue Type: Bug Components: JCR, Oak Affects Versions: JCR Resource 2.5.0 Reporter: Marc Pfaff Currently the OakResourceListener does not allow to configure the OAK observation queue length when instantiating the OAK BackgroundObserver. Thus the default queue size of 1000 is used. If the max queue size is reached, OAK seem's to turn the observation in a 'graceful degratation' mode, which leads to a compacted view of observations. Means, not every single observation makes it to the observer no more. From the point of view of the Sling resource changed events, this looks like some events are missing. I would propose to make the OAK observation queue size configurable in order to have the possibility to avoid 'graceful degratation'. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (SLING-4533) OakResourceListener does not allow to configure the OAK observation queue length
[ https://issues.apache.org/jira/browse/SLING-4533?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14377563#comment-14377563 ] Bertrand Delacretaz commented on SLING-4533: Am I correct that increasing the queue size will only move the threshold at which the problem occurs? Is there a loud warning from Oak when that threshold is hit? Configuration is good but IMO the most important is to expose the issue clearly, I don't know if that's currently the case. OakResourceListener does not allow to configure the OAK observation queue length Key: SLING-4533 URL: https://issues.apache.org/jira/browse/SLING-4533 Project: Sling Issue Type: Bug Components: JCR, Oak Affects Versions: JCR Resource 2.5.0 Reporter: Marc Pfaff Currently the OakResourceListener does not allow to configure the OAK observation queue length when instantiating the OAK BackgroundObserver. Thus the default queue size of 1000 is used. If the max queue size is reached, OAK seem's to turn the observation in a 'graceful degratation' mode, which leads to a compacted view of observations. Means, not every single observation makes it to the observer no more. From the point of view of the Sling resource changed events, this looks like some events are missing. I would propose to make the OAK observation queue size configurable in order to have the possibility to avoid 'graceful degratation'. -- This message was sent by Atlassian JIRA (v6.3.4#6332)