Skip to main content
Version: 5.11.0

Routine maintenance

Common cluster operations: bringing a chunkserver in or out of service, and upgrading LeilFS. Adding and removing individual disks on an existing chunkserver is covered separately in Storage device setup/removal; this page covers whole chunkserver nodes and cluster upgrades.

Adding a chunkserver​

  1. Install leil-chunkserver on the new node.

  2. Configure leil-chunkserver.cfg with the master's address (MASTER_HOST, MASTER_PORT).

  3. Create /etc/saunafs/leil-hdd.cfg listing the storage directories on the new node.

  4. Start the service:

    sudo systemctl start saunafs-chunkserver

  5. Confirm in the CGI monitor that the new chunkserver appears and that chunks begin replicating onto it.

Removing a chunkserver​

Data must be evacuated before the node is taken out, or copies below goal could be lost.

  1. Mark every disk of the node in leil-hdd.cfg with the * prefix to trigger evacuation (see hdd-config).

  2. Watch replication drain the node — use leil fileinfo on affected paths and/or the CGI monitor — until no chunks remain on it.

  3. Stop the chunkserver service:

    sudo systemctl stop saunafs-chunkserver

  4. Remove the node from any leil-topology.cfg entries.

caution

Never physically remove a chunkserver until you have confirmed every chunk it held has been replicated elsewhere. Pulling it early can drop copies below the configured goal and cause data loss.

Upgrading LeilFS​

LeilFS follows semantic versioning: minor and patch upgrades within a major version keep backward compatibility and can be done with rolling restarts. Major upgrades may need a migration step.

  1. Read the release notes and the Upgrade/Downgrade notes for the target version.
  2. Upgrade in this order: master → shadow masters → metaloggers → chunkservers → clients.
  3. For major upgrades, run the provided migration helper scripts (saunafs-migrations).
  4. Verify health after the upgrade with leil-admin info and by checking the logs.
note

The strict upgrade order keeps components that speak newer protocol versions behind those that must remain compatible. Do not upgrade chunkservers or clients ahead of the master.