Greg Stark wrote:
> This isn't the first time we've run into the problem that we've run
> out of signals. I think we need to multiplex all our event signals
> onto a single signal and use some other mechanism to indicate the type
> of message.
Yeah. A patch to do that was discussed a while ago, as Fujii's
synchronous replication patch bumped into that as well. I don't feel
like changing the signaling so dramatically right now, however.
> Do we really need a "promote to master" message at all? Is pg_standby
> responsible for this or could the master write out the configuration
> changes necessary itself?
The way pg_standby works is that it keeps waiting for new WAL files to
arrive, until it's told to stop and return a non-zero exit code.
Non-zero exit code from restore_command basically means "file not
found", making the startup process to end recovery and start up the
database. There's two ways to tell pg_standby to stop: create a trigger
file with a particular name, or signal it with SIGINT or SIGQUIT.
-- Heikki Linnakangas EnterpriseDB http://www.enterprisedb.com