I couldn’t reproduce the issue with 8.0.30, and on 8.0.31 both joining nodes were using a different GTIDsets, but this might be caused by the later reverted fix in 8.0.31. Version 8.0.34 worked fine. 8.0.35 introduces the issue it seems. There was some GTID change mentioned in the release notes that sounds suspicious.
Environment
None
Activity
Show:
Aaditya Dubey
September 17, 2024 at 12:09 PM
Hi @Oliver Dala
Yes, I’m exploring the other way around.
Oliver Dala
September 17, 2024 at 12:08 PM
I don’t know how to help you i’m afraid. Maybe use another type of test environment if this doesn’t work for you? I don’t think the bug itself is docker specific.
Aaditya Dubey
September 17, 2024 at 7:31 AM
Hi @Oliver Dala
Please ignore the query, however, I'm still having trouble with the above error.
Oliver Dala
September 17, 2024 at 7:09 AM
I don’t understand what you mean by that?
Aaditya Dubey
September 16, 2024 at 2:14 PM
Hi @Oliver Dala
Thank you for the updates. Do we really need 180+ inactive threads to simulate the issue?
Medium
During some testing i discovered a weird issue when trying to join 2 pxc nodes at once to a cluster.
The second node, that has to wait in line for the donor to be available, will use a different GUID for all new transactions after joining.
I can reproduce it every time in my docker testing setup starting with version 8.0.35 by executing these steps:
Bootstrap node1
Start node2 and node3 at the same time
After both are done: check gtid_executed. It will be the same on all 3 servers.
Execute a write query.
Check gtid_executed again, node1 and node2 will match, node 3 will have a new GTIDset.
Config used on all 3 nodes:
[mysqld] character-set-server=utf8 collation-server=utf8_general_ci log-slave-updates=1 skip-host-cache skip-name-resolve server-id=1 gtid_mode=on enforce_gtid_consistency=on log_bin binlog_format=row slave_parallel_workers=1Nothing special in the docker-compose eiter:
--- version: '3' networks: xdb: driver: bridge services: node1: container_name: xtradb_node1 image: percona/percona-xtradb-cluster:8.0 environment: - MYSQL_ROOT_PASSWORD=root - MYSQL_DATABASE=galera_test_db - MYSQL_USER=galera_test_db - MYSQL_PASSWORD=galera_test_db - CLUSTER_NAME=pxc-cluster - CLUSTER_JOIN= - PERCONA_TELEMETRY_DISABLE=1 volumes: - ./8.0-cluster-node-conf.d:/etc/percona-xtradb-cluster.conf.d - ./certs:/certs networks: xdb: {} node2: container_name: xtradb_node2 image: percona/percona-xtradb-cluster:8.0 environment: - CLUSTER_NAME=pxc-cluster - CLUSTER_JOIN=node1,node2,node3 - PERCONA_TELEMETRY_DISABLE=1 depends_on: - node1 volumes: - ./8.0-cluster-node-conf.d:/etc/percona-xtradb-cluster.conf.d - ./certs:/certs networks: xdb: {} node3: container_name: xtradb_node3 image: percona/percona-xtradb-cluster:8.0 environment: - CLUSTER_NAME=pxc-cluster - CLUSTER_JOIN=node1,node2,node3 - PERCONA_TELEMETRY_DISABLE=1 depends_on: - node1 volumes: - ./8.0-cluster-node-conf.d:/etc/percona-xtradb-cluster.conf.d - ./certs:/certs networks: xdb: {}I couldn’t reproduce the issue with 8.0.30, and on 8.0.31 both joining nodes were using a different GTIDsets, but this might be caused by the later reverted fix in 8.0.31.
Version 8.0.34 worked fine. 8.0.35 introduces the issue it seems. There was some GTID change mentioned in the release notes that sounds suspicious.