Fix scenario where xlog locations differ but apply_lag has returned 0
In this case we don't know how far we are behind time-wise, so just set to 1 week;
this is where using a lag value as bytes rather than seconds would be better.
glyn [Thu, 6 Aug 2015 12:42:13 +0000 (13:42 +0100)]
Mark experimental smartmach feature as enabled to suppress warnings in perl 5.18+.
This doesn't mitigate the fact that smartmatch is now considered experimental,
but use of smartmatch in this script is limited to simple given/when with string literals.
glyn [Tue, 13 Jan 2015 13:45:58 +0000 (13:45 +0000)]
Add an initial state to cluster hash, this way we can tell the difference between a cluster that has no suitable target nodes and where the previous cluster state has not yet been initialized.
We also want to register that things have changed and there is now no suitable node, that way we can write nothing and effectively shut off the pool.
glyn [Mon, 1 Dec 2014 20:51:03 +0000 (20:51 +0000)]
Changes:
- Ensure we always read Slony configuration from the origin node when possible
- Add max_ro_lag=<n> seconds parameter to allow forced reconfiguration of read-only pools when subscribers start to lag
glyn [Wed, 22 Oct 2014 13:17:15 +0000 (14:17 +0100)]
A couple of minor changes:
- Added extra configuration parameter 'only_follow_origins' to only repoint pgbouncer configuration if the origin of the monitored sets changes, which is what we usually want.
- Only write cluster status information file when cluster status has actually changed.