+1-208-473-2904 (USA - Sales)
0-800-051-8984 (UK - Sales)
0-800-181-0665 (GER - Sales)
Based on Percona Server 5.6.19-67.0 including all the bug fixes in it, Galera Replicator 3.6 and on Codership wsrep API 25.6 is now the current General Availability release. All of Percona‘s software is open-source and free, all the details of the release can be found in the 5.6.19-25.6 milestone at Launchpad.
During joiner joining the group, state message exchange provides us with gcache seqno limits. That info is now used to choose a donor through IST first, if not possible, only then SST is attempted. wsrep_sst_donor is also honored here. This is also segment aware. (#1252461)
Asynchronous replication slave thread is stopped when the node tries to apply next replication event while the node is in non-primary state. But it would then remain stopped after node successfully re-joined the cluster. A new variable wsrep_restart_slave has been implemented, which controls if MySQL slave should be restarted automatically when the node joins back to the cluster. (#1288479)
Handling install message and install state message processing has been improved to make group forming more stable in case many nodes are joining the cluster. (#14)
Percona XtraDB Cluster now supports storing the Primary Component state to disk by setting the pc.recovery variable to true. The Primary Component can then recover automatically when all nodes that were part of the last saved state reestablish communications with each other. This feature can be used for automatic recovery from full cluster crashes, such as in the case of a data center power outage and graceful full cluster restarts without the need for explicitly bootstrapping a new Primary Component. (#10)
Node consistency issues with foreign keys have been fixed. This fix introduces two new variables: wsrep_slave_FK_checks and wsrep_slave_UK_checks. These variables are set to TRUE and FALSE respectively by default. They control whether Foreign Key and Unique Key checking is done for applier threads. (#1260713).
Fixed the race condition in Foreign Key processing that could cause assertion. Bug fixed #1342959.
The restart sequence in scripts/mysql.server would fail to capture and return if the start call failed to start the server, so a restart could occur that failed upon start-up, and the script would still return 0 as if it worked without any issues. Bug fixed #1339894.
Updating a unique key value could cause server hang if slave node has enabled parallel slaves. Bug fixed #1280896.
Percona XtraDB Cluster has implemented threadpool scheduling fixes. Bug fixed #1333348.
garbd was returning incorrect return code, ie. when garbd was already started, return code was 0. Bug fixed #1308103.
wsrep_sst_rsync would silently fail on joiner when rsync server port was already taken. Bug fixed #1099783.
Example wsrep_notify script failed on node shutdown. Bug fixed #1132955.
When gmcast.listen_addr was configured to a certain address, local connection point for outgoing connections was not bound to listen address. This would happen if OS has multiple interfaces with IP addresses in the same subnet, it may happen that OS would pick wrong IP for local connection point and other nodes would see connections originating from IP address which was not listened to. Bug fixed #1240964.
Percona XtraDB Cluster couldn’t be built with Bison 3.0. Bug fixed #1262439.
mysqld wasn’t handling exceeding max writeset size wsrep error correctly. Bug fixed #1270920.
When FLUSH TABLES WITH READ LOCK was used on a node with wsrep_causal_reads set to 1 while there was a DML on other nodes then, subsequent SELECTs/SHOW STATUS didn’t hang earlier providing non-causal output, that has been fixed here. Bug fixed #1271177.
Percona XtraDB Cluster server package no longer conflicts with mysql-libs package from CentOS repository. Bug fixed #1278516.
The mysql-debug UNIV_DEBUG binary was missing from RPM/DEB server packages. Bug fixed #1290087.
When Query Cache was used and a node would go into non-PRIM state, queries which returned results earlier (and cached into query cache) would still return results whereas newer queries (or the ones not cached) would return unknown command. Bug fixed #1296403.
Brute Force abort did not work with INSERTs to table with single unique key. Bug fixed #1299116.
Close referenced table opened in the same function when foreign constraints were checked, otherwise it could lead to server stall when running DROP TABLE. Bug fixed #1309241.
Compiling on FreeBSD 10.0 with CLANG would result in fatal error. Bug fixed #1309507.
Truncating the sorted version of multi-byte character conversion could lead to wsrep certification failures. Bug fixed #1314854.
Cluster node acting as async slave would stop with the wrong position after hitting max_write_set_size. Bug fixed #1309669.
Fixed the events replication inconsistencies. Bug fixed #1312618.
garb couldn’t be started with init script on RHEL 6.5. Bug fixed #1323652.
GTID of TOI operations is now also synced to InnoDB tablespace in order to get consistent backups. Bug fixed #1329055.
mysql-debug is now distributed with binary tar.gz along with RPM and DEB packages. Bug fixed #1332073.
Startup failure with Undetected state gap has been fixed. Bug fixed #1334606.
Galera3 is now installed in /usr/lib/galera3/libgalera_smm.so with a compatibility symlink to /usr/lib/libgalera_smm.so. Bug fixed #1279328.
Galera could not be compiled on PowerPC. Bug fixed #59.
Cluster could stall if leaving node failed to acknowledge all messages it had received due to exception and remaining nodes failed to reach consensus because of that. Bug fixed #37.
When two node instances were set up on the same server with the two different IPs, they couldn’t not work well because they were use wrong IP addresses. Bug fixed #31.
Automated donor selection with segments gave inconsistent results. Bug fixed #29.
Help us improve quality by reporting any bugs you encounter using our bug tracking system. As always, thanks for your continued support of Percona!