Hi! This is an attempt to collect feedback and provide notification in the context of KUDU-3583 [1].
KUDU-3583 is about changing the default setting for the tablet_history_max_age_sec flag [2] to 15 minutes from its current default setting of 7 days. The motivation for this anticipated change is preventing issues like KUDU-3406 [3] and avoiding waste of CPU cycles, disk IO bandwidth, and memory capacity on processing of irrelevant data. Also, since KUDU-3429 [4] hasn't been addressed yet, there might be issues due to OOM conditions in extreme cases. The default setting for tablet_history_max_age_sec was changed from 15 minutes to 7 days in Kudu 1.10.0 to support backup-and-restore functionality out-of-the-box (see KUDU-2677 [5]), but not every Kudu cluster is backed up incrementally. If implemented, Kudu users who use incremental backup in their clusters will need to provide an override for the flag, setting --tablet_history_max_age_sec to be at least as long as the time interval between two consecutive incremental backup sessions. Also, users who use read-at-snapshot read operations for snapshots beyond 15 minutes from current time will need to customize the setting as needed. I discussed this with Abhishek Chennaka who is running release management for the upcoming Kudu 1.18.0. Abhishek is willing to include this update into the 1.18.0 release. Yes, we are planning to add a corresponding item in the release notes as well. Does this look like a drastic and maybe a breaking change to anybody? If yes, what alternatives should we consider instead? Your feedback and ideas are very welcome. Thank you! Kind regards, Alexey [1] https://issues.apache.org/jira/browse/KUDU-3583 [2] https://kudu.apache.org/docs/configuration_reference.html#kudu-tserver_tablet_history_max_age_sec [3] https://issues.apache.org/jira/browse/KUDU-3406 [4] https://issues.apache.org/jira/browse/KUDU-3429 [5] https://issues.apache.org/jira/browse/KUDU-2677
