Jira (PDB-5548) Partition GC's query bulldozer not guaranteed to make progress

[Rob Browning (Jira)]

Title: Message Title

Rob Browning updated an issue   PuppetDB / PDB-5548 Partition GC's query bulldozer not guaranteed to make progress Change By: Rob Browning Release Notes Summary: PuppetDB should drop old partitions (e.g. reports) more effectively. Because dropping the partitions currently requires an exclusive lock that can block all all other queries, PuppetDB tries to eliminate any queries that are blocking a drop attempt, but the method previously used to stop queries was insufficient.  In addition, as constructed right now, competing queries can deadlock with the drop attempt.  If a deadlock doesn't resolve within the configured `deadlock_timeout`, PostgreSQL's deadlock detector will break the deadlock, and to do so in some cases it has chosen to kill the partition drop attempt every time. To fix this problem, PuppetDB now terminates any queries competing with a drop attempt via `pg_terminate_backend` instead of `pg_cancel_backend`.  This should be very reliable (cancel was more advisory), and PuppetDB also makes sure that it has the connection it needs for the terminations before it makes making a drop attempt.  Finally, it will try to issue those terminations termination requests more frequently than it did before (every fifth of a second instead of every second). Add Comment   This message was sent by Atlassian Jira (v8.20.11#820011-sha1:0629dd8) -- You received this message because you are subscribed to the Google Groups "Puppet Bugs" group. To unsubscribe from this group and stop receiving emails from it, send an email to puppet-bugs+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/puppet-bugs/JIRA.470479.1664997895000.56202.1667839320166%40Atlassian.JIRA.