The statistics of miss-peer is not accurate sometimes

[rleungx]

Bug Report

What did you do?

Use pd-ctl to set the max-replicas to 2 with a cluster having 15w regions. And after all extra peers are removed, change back the max-replicas to 3.

As you can see from the above screenshot, after changing back the parameters, there should be a lot of missing peers. But actually, there are a few missing peers.

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

master

[rleungx]

There is another issue: After we change the placement rule to add more peers, there are some missed peers which are always exist, but the log shows the region doesn't have missed peers.