[BUG] GatewayService options set for deprecation for many years now

[phvalguima]

Describe the bug

Hi, I was looking into GatewayService code and noted that its options seems to be set for deprecation. However, that has been set for many years now. It is also using the wrong naming convention here. It has been marked for deprecation for the last 3 years at least.

What is the plan for these options?

Personally, having a "holder" option that let's me say: please, do not do start the cluster until N cluster-manager and/or data nodes are present is very valuable. Can we keep these two options (i.e. cluster manager and data nodes) + remove the deprecation notices + update its naming convention?

(If yes, I am more than happy to do a PR :) )

Use Case

The main use-case for us is a cluster that lost most of its eligible managers. In this case, each unit will block saying it cannot form quorum any longer. However, we are currently working on some automation to detect these scenarios and, given certain conditions, reset the quorum to a smaller count. Ideally, we'd use gateway.expected_master_nodes and some of the other options to resolve this problem.

Related component

Cluster Manager

To Reproduce

Check the code references above.

Expected behavior

These options are very valuable and should just be updated to the current standards + removing deprecation notices.

Additional Details

No response

[andrross]

[Catch All Triage - 1, 2, 3]

Thanks for filing. @phvalguima Feel free to open a PR to move the discussion forward.

[rajiv-kv]

Thanks @phvalguima for filing the issue. On your usecase of recovering the cluster with a surviving node, have you explored seeding the masters with ips using the setting initial_master_nodes.