On Tue, Mar 20, 2012 at 14:35, Asher Feldman <afeldman(a)wikimedia.org> wrote:
On Tuesday, March 20, 2012, Roan Kattouw
<roan.kattouw(a)gmail.com> wrote:
So yeah /normally/
you hit DB servers at random and different servers might respond
differently (or be lagged to different degrees), but in this
particular case it was always the same DB server returning the same
lag value. Nothing strange going on here, this is how the maxlag
parameter works.
How do you feel about a switch to change that behavior (maxlag - 1)? It
would be nice to be continue guiding developers towards throttling API
requests around maxlag without complicating schema migrations by requiring
config deployments before and after every db for this reason only.
How about just having a new file in the config that lists the servers
currently migrating and the migration script can update and deploy as
it goes. Then apaches can check if a high maxlag is on the lag's
expected list and ignore if it is. Should be fairly safe as long as
it's a file that no human will edit and the deploy only pushes the one
file. But, I don't actually know if it's possible (with current
scripts) to push just one file for the config.
-Jeremy