I created a test bed on VirtualBox and test HA cluster by random failures in loop. Sometime, in case of longest switching from failure, I get strange behaviour. The walreceiver is stopped by timeout, but don't restarted for 2 hours. May be I agree with stopping by timeout, but why postgresql wait for 2 hour to start it again?
2019-08-12 16:34:31.118 MSK  FATAL: terminating walreceiver due to timeout
2019-08-12 16:34:31.119 MSK  LOG: record with incorrect prev-link DC7A2D84/100 at 0/D078A38
2019-08-12 18:34:50.222 MSK  FATAL: could not connect to the primary server: server closed the connection unexpectedly
This probably means the server terminated abnormally
before or while processing the request.
2019-08-12 18:34:50.234 MSK  LOG: fetching timeline history file for timeline 4 from primary server
2019-08-12 18:34:50.235 MSK  LOG: started streaming WAL from primary at 0/D000000 on timeline 3
2019-08-12 18:34:50.237 MSK  LOG: replication terminated by primary server
2019-08-12 18:34:50.237 MSK  DETAIL: End of WAL reached on timeline 3 at 0/D078A38.
2019-08-12 18:34:50.238 MSK  LOG: new target timeline is 4
2019-08-12 18:34:50.239 MSK  LOG: restarted WAL streaming at 0/D000000 on timeline 4
May be reason is in restart_after_crash=off option (recommended fo HA clusters).
shared_buffers = 256MB
# may be this is a good compromise
synchronous_commit = remote_write
# other DC is the first, our is the last (this is different for each node)
synchronous_standby_names = 'FIRST 1 (tuchanka2a,tuchanka2c,tuchanka2b)'