Thread: are analyze statistics synced with replication?
We have a master/slave setup with replication. Today we failed over to the slave and saw disk I/O go through the roof.
Are the pg_statistic statistics synced along with streaming replication? Are you expected to have to do a vacuum analyze after failing over? That's what we're trying now to see if it makes a difference. Our next step will be to fall back to the first host and see where this one went wrong (society? lax discipline at home? the wrong sort of friends?)
If you don't do read queries on the slave than it will not have hot data/pages/rows/tables/indexes in ram like the primary ? (it smoked weed and was happy doing nothing so it was happy, but when responsibility came (being promoted to master) it failed hard)
On Thu, May 15, 2014 at 6:46 AM, Kevin Goess <kgoess@bepress.com> wrote:
We have a master/slave setup with replication. Today we failed over to the slave and saw disk I/O go through the roof.Are the pg_statistic statistics synced along with streaming replication? Are you expected to have to do a vacuum analyze after failing over? That's what we're trying now to see if it makes a difference. Our next step will be to fall back to the first host and see where this one went wrong (society? lax discipline at home? the wrong sort of friends?)
On Thu, May 15, 2014 at 6:39 AM, Dorian Hoxha <dorian.hoxha@gmail.com> wrote:
If you don't do read queries on the slave than it will not have hot data/pages/rows/tables/indexes in ram like the primary ?
Yeah, that was the first thing we noticed, the cacti graph shows it took two hours for the page cache to fill up our 64GB of RAM, but I/O didn't stop sucking after that.