In cluster reconfiguration scenarios where one marks log-server as read-only we can be losing write availability of logs whose open segment contains too many read-only nodes has not been reconfigured yet. Usually, the reconfiguration follows the partition processor leader. If at the same time when marking those nodes as read-only, the system is also moving a pp leader, then the candidate might not be able to write its AnnounceLeadership message to the log. Since auto log improvement only works after a PP has become the leader, auto improvement won't help us in this situation either.

A possible solution is to let the appender trigger a reconfiguration if it detects that a log has become write unavailable.