I'm having trouble tracking down the cause of an error message that is appearing in my logs every 15 minutes or so.

The cause of the error happened whilst performance testing a 3 node Galera Cluster (running on MariaDB 10.1.13). Two of the read-only nodes fell too far behind the write master causing them to request an SST sync and pulling the cluster off-line whilst the write master fulfilled the sync.

Since that point though the master has regularly been posting the following message into the logs:

2016-08-26 18:03:16 139975496911616 [ERROR] WSREP: failed to initialize io-cache
2016-08-26 18:03:16 139975496911616 [Warning] WSREP: binlog trx cache not empty (0 bytes) @ connection close 6510
2016-08-26 18:03:16 139975496911616 [Warning] WSREP: binlog stmt cache not empty (0 bytes) @ connection close 6510

Google is sparse on any details and the only good hit I've had is from the source code, which appears to suggest that it's unable to initialise a cache for reading or writing (hmmm, much like the error message suggests).

   32   if (reinit_io_cache(cache, READ_CACHE, 0, 0, 0))
   33   {
   34     WSREP_ERROR("failed to initialize io-cache");
   35     return ER_ERROR_ON_WRITE;
   36   }

This message continues despite a reboot and full SST sync from the other nodes, so I'm a little lost as to the cause. Does anyone know what this means and how to fix it?

  • Just in case someone comes across this thread and has this sort of moment; we never got to the bottom of this issue. Further testing of Galera showed that the synchronous replication overhead made it unsuitable for our use case (the application was too write heavy under load), so we converted to a standard async replication in order to meet our performance targets.
    – Dan
    Commented Aug 29, 2016 at 15:18


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.