vmstorage failure

[Thierry Fournier]

Hi,

I read the docs about Victoria Metrics, but I do not find answer to my questions.

I want to use high availability for data storage. I understand that I must provide 2*N - 1 datastorage node where N is the number of replica. When one datastorage die, can I reconstruct using data of other datastorage ? Backup/restore is the only way to restore datastorage node ?

What about split brain ? If the connexion between my 2 DC is down, there are a quorum mechanism or something like that to avoid data split brain ?

Thanks

Thierry

[Yurii Kravets]

By default every sample is written to a single vmstorage node, which is chosen via consistent hashing. Time series are spread evenly among the configured vmstorage nodes under normal conditions thanks to the consistent hashing.
When the -replicationFactor is greater than 1 (for example, if RF=2), then every sample is written into RF distinct vmstorage nodes.

So in this specific case 3 vmstorage nodes can be configured with  -replicationFactor=2
It will assure that data is written 2 times to every vmstorage. In case 1 node is down, data will be written to other available vmstorage nodes. So there is no need to reconstruct anything as it is already available on other nodes.

More info can be found here:
https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#replication-and-data-safety