When a small number of hosts down, on cluster ddl will not return timeout exception

[marising]

Use case
#21574
#23062

When a small number of hosts down, whether it is a permanent or temporary, ON CLUSTER DDL will not return timeout exception.

Describe the solution you'd like

1. When the host starts, run a background thread to periodically check for 60s, and register the current host as the ephemeral node of zooKeeper, the path is is_active.
   Activate current host as ephemeral node in zookeeper #26269

2. Distributed_ddl_output_mode is set to QUORUM, when DDLQueryStatusInputStream is output, all active hosts are executed successfully, there are only inactive nodes in unfinished_hosts, and inactive hosts are a minority
   2.1. unfinished_hosts are all inactive hosts
   2.2. Check each shard : the count of inactive host replicas / the total count of shard replicas <= 1/2
   Add quorum mode to the distributed_ddl_output_mode when executing DDL… #27004

Satisfy the above 2.1 and 2.2, will not wait for more than distributed_ddl_task_timeout, directly return the information of the number of successful hosts and the number of failed hosts, otherwise it will wait for timeout.

I plan to implement this feature, can you give me some suggestions? @tavplubix

[marising]

@filimonov @alexey-milovidov Do you have any comments?

[alexey-milovidov]

The idea is to simply add a mode - wait for all active servers.
The server is active if it holds an ephemeral node in Keeper.

[azat]

It seems that the alternative can be to use cluster discovery right now - #31442