You can start old master, wait for crash recovery to complete, stop it cleanly and then use pg_rewind. It works.
Test 2 :
- On a successfully running streaming replication with one master and one slave, i did a clean shutdown of master - promoted slave - performed some operations (data changes) on newly promoted slave and did a clean shutdown - Executed pg_rewind on the old master to sync with the latest changes on new master. I got the below message
The servers diverged at WAL position 0/A2000098 on timeline 1. No rewind required.
I am not getting this too.
In this case the master WAL visibly did not diverge from the slave WAL line. A rewind is done if the master touches new relation pages after the standby has been promoted, and before the master is shutdown. -- Michael