"Not enough replicas" caused keyspace group split to fail randomly

[binshi-bing]

Enhancement Task

What did you do?

QA team ran keyspace group split with pd-ctl

What did you expect to see?

keyspace group split should successfully complete

What did you see instead?

Randomly failed because of "Not enough replicas"

What version of PD are you using (pd-server -V)?

tidbcloud/pd-cse release-6.6-keyspace 9e1e2de

[rleungx]

This is because the watcher is started before the bootstrap process. The alloc node loop exits since there is no group in etcd (introduced by #6390). When the bootstrap process is trying to init the default group, it cannot be assigned the members, so it will persist the null to etcd which triggers the above problem.