Node crashes with Transport endpoint is not connected

Post-merge fix will be provided in 8.0.34 by this PR
https://github.com/percona/galera/pull/270

Hi,

As per the discussion here: https://forums.percona.com/t/occasional-db-crashes-in-pxc-8-0-32-around-remote-endpoint-transport-endpoint-is-not-connected/25158  we believe this is still an issue in PXC 8.0.32.

I’ve been able to replicate the crash with a two node cluster on PXC 8.0.32 using nmap’s tcp connect scan against port 4567 on both nodes.

I’ve got our application connected to <node1> (generating some writeset changes) and if I leave this running from a third host:

…I see a lot of these entries in both node logs as the commands loop:

…and after some time (usually minutes) <node1> falls over e.g:

Hope its helpful!

Thank you, this really helps

Sadly nothing was documented in pxc releases about wsrep version, and variables still shows 26.4.3

I found out I should have checked for GALERA_VERSIONS file in submodules

Minimal testcase assuming default Galera communication port 4567:

1. start single node cluster

2. while true; do nmap -p4567 127.0.0.1; done

It crashes immediately.

The issue was fixed by Galera upstream commit 930c016108d7086b472ad7a8b9d0f6989202b48a and is included in Galera 26.4.12, so:

8.0.27 -> galera 26.4.10 - failure
8.0.28 -> galera 26.4.11 - failure
8.0.29 -> galera 26.4.12 - works fine