Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Mon, Dec 10, 2012 at 09:25:32AM -0500, Nathan Joyes wrote: > >> We are investigating the possibility of using DRBD to perform > >> synchronous replication of our PostgreSQL database server. After > >> reading the Replication modes section of the DRBD features > >> documentation we concluded that we would need to use protocol C, which > >> is describes writes to be considered complete only after both the > >> local and the remote disk writes have been confirmed. However, during > >> our testing we were able to continuously write to the primary device > >> after disabling network traffic on the secondary. > > > >http://www.drbd.org/users-guide/s-node-failure.html > > > >> We also noticed that during high write loads on the primary the > >> number of kilobytes out of sync would grow, but eventually the > >> secondary would catch up. > > > >Only when in disconnected mode. And then after the connection is restored, you resync: > >http://www.drbd.org/users-guide/s-resync.html > > > >> After reading the documentation, we were surprised at the testing results. > > > >Um, sorry, you seem to not have reviewed the documentation in sufficient detail. :) > > > >> Is this expected and/or correct behavior, or have we incorrectly > >> configured something for the DRBD device? > > > >Perfectly expected. > > > >Now, if you're looking for PostgreSQL HA, the alternative to DRBD is > >obviously to use Postgres synchronous replication, available since > >9.1. Also integrated with Pacemaker. Since that option forgoes >DRBD, > >though, a discussion of that option would be off-topic for this list, > >and if you're interested in pursuing that option I'd encourage you to > >take this discussion to the Pacemaker or linux-ha >mailing list. > > Thanks for the links. After re-reading them it sounds like protocol C > is synchronous when in a connected state, otherwise it suspends replication > until the connection is restored, and it then performs a resync. Right. Though we also can behave differently (e.g. block IO). What behaviour would you like to see? Why? -- : Lars Ellenberg : LINBIT | Your Way to High Availability : DRBD/HA support and consulting http://www.linbit.com DRBD® and LINBIT® are registered trademarks of LINBIT, Austria.