guilbaults / puppet-lustre Goto Github PK

Require a file in /tmp before puppet will format storage disk. Just in case the auto detection does not work, this will prevent data loss on a real system.

Corosync timeout are too short when importing a pool that was not cleanly exported.

The disks scan of zpool import can make the monitoring checks of corosync time out since zpool status hang until the zpool import command returns.

The timeout is currently hardcoded, it should be a variable. Its currently increased from the default corosync value, but this is not enough on large JBODs.

Should be enabled by default when formating on 2.11, can be enable on older versions
https://jira.whamcloud.com/browse/LU-8342

Currently using a fork with a incompatible stonith

https://github.com/voxpupuli/puppet-corosync#configuring-stonith-resources

A script is required to verify that HSM is enabled and that copytool are connected.

The service_nodes parameter should be by OST instead of being at the OSS level.

This will enable the correct ordering of the ip address used when a client is connecting.

Without this all clients tries to connect to the first ip, while about half the OST will reside on the second OSS. This is related to the prefered_host parameter which set the primary location of each OST, this prefered host should be the first ip address in service_nodes.

The check_targets script need to handle INACTIVE target status.

INACTIVE simply means all the clients were disconnected before the target was mounted again, this is not an error.

https://lustre-discuss.lustre.narkive.com/54eF5qsF/recovery-status-inactive