Moving elk data path to dedicated volume

[agiorgi (Alessandro Giorgi)]

Hello everyone,

I've been asked to move Elasticsearch's data path from default path, to a new dedicated volume.
I still do not have any detail (except that cluster is running on RHEL VMs), but i'm afraid that the volume is going to be provided through an NFS share.
I know that data on NFS kills indexing performance, but besides performance, do i need to care about something else? e.g. Indexes getting corrupted or similar etc...
There are some serious consequences on having data on an NFS share?
Beside that, which are going to be the steps to actually do this migration?
Assuming that i already have the new volume mounted and available on my nodes, these are the steps that i would do:
Cluster level:

• Disable shard allocation

For every Elasticsearch node:

• Shut down ES
• Copy/mv data to the new volume (maybe using rsync)
• Amend ES config file with correct data.path
• Start ES

Once all nodes are up and back running
Cluster level:

• Re-enable shard allocation

What do you think about that?
is is enough?
Do i have to completely take off the cluster?

Many thanks!
Regards,
Alessandro

[DavidTurner]

@agiorgi

There are some serious consequences on having data on an NFS share?

It depends how the NFS share is configured. It needs to behave like a local filesystem for Elasticsearch to work correctly, which I think means it needs to be at least NFS 4.1. I don't have any more details on how to configure it, however.

@agiorgi

Assuming that i already have the new volume mounted and available on my nodes, these are the steps that i would do:

Yes, that seems reasonable. You must move the data (or copy it and then delete the original) to avoid starting up two copies of the same node by accident.