HDDS-13594. Use a different endpoint for fetching the OM checkpoint tarball.

[sadanand48]

What changes were proposed in this pull request?

HDDS-12984 introduced a newer implementation for the OM tarball transfer process but replaces the older endpoint with the new impl. Instead introducing a newer endpoint maybe a better choice as it will handle upgrade related scenarios. An older client will always hit the older endpoint and newer client will hit the newer one. This way no layout change is needed.

What is the link to the Apache JIRA

https://issues.apache.org/jira/browse/HDDS-13594

How was this patch tested?

compat acceptance tests covering:

• New client requesting checkpoint from new server (V1,V2) - pass
• New client requesting checkpoint from old server (V1) - pass
• Old client requesting checkpoint from old server (V1) pass
• Old client requesting checkpoint from new server (V1) pass

[errose28]

An older client will always hit the older endpoint and newer client will hit the newer one. This way no layout change is needed.

We currently don't have rolling upgrade support, so the clients in this case would be other OMs or Recon, which would be in the same version. Is the new endpoint still necessary? As long as it still has the option to fetch the whole DB without snapshots for Recon's purposes I think the same endpoint is fine since it is for internal use.

[sadanand48]

We currently don't have rolling upgrade support, so the clients in this case would be other OMs or Recon, which would be in the same version. Is the new endpoint still necessary?

Yeah it's not very relevant at the moment but might be relevant during a rolling upgrade/partial upgrade like scenarios where a follower OM/Recon has a different version than the leader OM.

Also with the newer endpoint the obtained sst files in the tar won't be placed correctly in the snapshot dirs/om.db and will have inodeID based names etc. The client has additional logic to execute for the new endpoint .
Maintaining the old endpoint will help in manual bootstrap scenarios where one could directly hit it without extra handling on the client.

[priyeshkaratha]

Looks good for me.

[jojochuang]

I think this looks fine. @sadanand48 do you want to update the endpoint name to /v2/dbCheckpoint ? I am okay either way.

[smengcl]

Thanks @sadanand48 . lgtm. Just one doc change requested

[sadanand48]

Thanks @errose28 @priyeshkaratha @jojochuang @smengcl for the reviews