You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Is your feature request related to a problem? Please describe
To restore snapshot from remote store enabled cluster to non remote store enabled cluster or to another remote store enabled cluster, currently user need to explicitly override the remote store related settings (remote_store.repository, remote_translog.repository, remote_store.enabled) by providing them as part of request fields in restore api.
Since remote store is more of a cluster level feature and replication strategy will always be SEGMENT with remote store feature enabled. user should not have to provide override for these index settings. during restore, these settings can be pulled and applied from node attributes.
Describe the solution you'd like
As part of this issue, proposing the following change in behavior:
Override the remote store related settings with the values present in node attributes during restore since remote store related settings are more of cluster level settings than index level settings.
For replication strategy related settings, if remote store feature is enabled in the cluster. override the setting to SEGMENT. otherwise, if cluster.force.index.replication_type settings is present in cluster, override the replication strategy with cluster default. If this setting is not present, let the index come up with different replication_type then the cluster default since user anyway allows indices with different replication strategy in same cluster.
All the restore scenarios will work as well with this solution:
Snapshot Restore Scenario
Works/don't work
Notes
Same Version Remote Store → Same Version Non Remote Store
will work
remote store settings will be ignored during restore, replication strategy will be overriden with cluster default if `cluster.force.index.replication_type is present, else indices will be restored as segrep.
Same Version Non Remote Store → Same Version Remote Store
will work
remote store and replication strategy settings will be applied from the node attributes.
2.12 Version Remote Store → 2.10(<=) Version Non Remote Store
will not work
snapshot restoration to an older version cluster is not allowed.
(<=)2.10 Version → 2.12 Version Non Remote Store
will work
2.12 Version Non Remote Store → 2.10(<=) Version
will not work
snapshot restoration to an older version cluster is not allowed.
(<=)2.10 Version → 2.12 Version Remote Store
will work
remote store and replication strategy settings will be applied from the node attributes.
Related component
Storage:Snapshots
Describe alternatives you've considered
No response
Additional context
[TODO]: add an example after restoring without overriding settings, add an example after restoring from snapshot by overriding settings
The text was updated successfully, but these errors were encountered:
Makes sense to me. The only thing I want to double confirm is what all combinations of Remote Store + Replication type is possible to be deployed by a user. Do we only allow SegRep indices with remote store enabled ? Or we also allow mix of doc rep + seg rep based indices where only later uses the remote store capabilities ? I think we don't allow the mix but wanted to confirm on this.
@sohami AFAIK, for remote store enabled clusters, user can only have seg-rep indices. for non remote store enabled clusters, user can have a mixture of doc-rep and seg-rep indices if they want.
Is your feature request related to a problem? Please describe
To restore snapshot from remote store enabled cluster to non remote store enabled cluster or to another remote store enabled cluster, currently user need to explicitly override the remote store related settings (remote_store.repository, remote_translog.repository, remote_store.enabled) by providing them as part of request fields in restore api.
Since remote store is more of a cluster level feature and replication strategy will always be
SEGMENT
with remote store feature enabled. user should not have to provide override for these index settings. during restore, these settings can be pulled and applied from node attributes.Describe the solution you'd like
As part of this issue, proposing the following change in behavior:
- Override the remote store related settings with the values present in node attributes during restore since remote store related settings are more of cluster level settings than index level settings.
- For replication strategy related settings, if remote store feature is enabled in the cluster. override the setting to
All the restore scenarios will work as well with this solution:SEGMENT
. otherwise, ifcluster.force.index.replication_type
settings is present in cluster, override the replication strategy with cluster default. If this setting is not present, let the index come up with different replication_type then the cluster default since user anyway allows indices with different replication strategy in same cluster.Related component
Storage:Snapshots
Describe alternatives you've considered
No response
Additional context
[TODO]: add an example after restoring without overriding settings, add an example after restoring from snapshot by overriding settings
The text was updated successfully, but these errors were encountered: