1 Sep 2010 02:47
what is happening on the subscriber side?
Hi everyone, Just wanted to look for an explanation regarding what happens on the subscriber's end of a replication set. I currently have 4 nodes (1 thru 4), and node 4 also has the "-a <dir>" flag turned on for log shipping (but I think this is irrelevant) Occasionally, I will see in test_slony_state_dbi.pl, that one of the subscribers has really old events or that the provider is lagging behind the provider, so I decided to harvest some data. Wrote up a cronjob that will fetch the average slony lag on node 4 (I could've picked any of them, but just chose this one because load was lowest). Basically, I ran this command from the shell every hour: `psql -tc "select avg(st_lag_time) from _slony_schema.sl_status" mydb postgres` Now, I logged it into a file (http://pgsql.privatepaste.com/e4ce8f8f67) and it shows that the other nodes average > 70 days' worth of lag at times. (see period from May 10 to Jul 09) There are 3 other replication clusters I tracked, and one of them even went up to 153 days before dropping right back down to zero. Could someone explain why this happens, or perhaps more importantly--what causes the lag to drop from high 70s of days down to 0. Is it sl_log_{1,2} rotation? Sorry, I might be able to find the answer by scouring the logs, but I'm hoping to find a quick answer here.(Continue reading)
RSS Feed