Detecting the crashed replica

Question

Hello,
In the replication group, one master with hundreds of replica, how can master and other replicas detect the crashed replica?
Thanks,
Ritchie Li

Answer 1

The way you've phrased your question is very open-ended, so it's hard
to tell just what you're asking. Can you narrow it down for us?
Typically the master and clients are connected via some sort of simple
telecommunications link, often TCP/IP. In that case you can't tell
for sure when a remote site has crashed; all you can do is try to
decide when the communication link has failed, and then guess that
that may have been caused by a crash. But actually the remote site
may still be running, on the other side of a failed link.
Conversely, even after a crash of a remote site, a TCP/IP
communication failure may not be evident, at least for a little while.
Using heartbeat messages is the usual way to avoid this latter
problem.
In an application that uses the replication "Base API" all of this is
entirely under control of the application, so there's not much more we
can say about it.
Replication Manager can use heartbeat messages to allow clients to
detect the failure of the communication link to the master, but not
the other way around: it doesn't detect failures of remote replicas.
Also, even in the case of detecting a master failure, Replication
Manager does not notify the application of the failure; it just starts
an election internally to choose a new master. (However, when a new
master is established, then Replication Manager does notify the
application.)
You could use the DB_ENV->repmgr_site_list() API to query
Replication Manager's opinion of the health of each connection. But
do remember that, because of the caveats mentioned above, this can
only be an approximate guess as to the actual status of the remote
sites themselves.
Alan Bram
Oracle