maria-discuss team mailing list archive

[Lyle Franklin <cf-core-services-eng@xxxxxxxxxx>]

Recently we observed that when network and IO latency increased
significantly (and correspondingly, that bandwidth reduced significantly)
all nodes in a 3-node mariadb galera cluster reported the other nodes as
partitioned.

We understand that by default galera is configured to wait 6 seconds before
declaring other nodes partitioned. Is there any additional documentation on
the performance on mariadb/galera with respect to network/IO latency and
bandwidth? Are there hard limits beyond which a cluster is known to fail,
and soft limits which are known to degrade performance?

Thanks,

Rob and Lyle
Cloud Foundry Core Services