On Nov 26, 2007 12:08 AM, Toshio Kuratomi <a.badger@xxxxxxxxx> wrote: > Yaakov Nemoy wrote: > > On Nov 25, 2007 4:00 PM, Toshio Kuratomi <a.badger@xxxxxxxxx> wrote: > Not if the app doesn't. From a brief test, TG apps do not do this. > > The script is asking supervisor to shutdown the application. supervisor > sends a TERM to the TG app (we can configure it to send something other > than TERM if we want but I don't see any documentation that leads me to > believe it will be different with a HUP or QUIT). At that point it > looks like a TG app will immediately shutdown and rollback any current > transactions. It's got my vote then > smolt is on shaky ground if it's not using transactions correctly... At > the beginning of the month when smolt was getting hit hard we did pretty > much this same thing except manually instead of via a script when we > noticed that smolt was giving timeouts and taking up 1G+ of RAM. I > think the current smolt code is using SQLAlchemy, correct? It's pretty > easy to use transactions so that you don't leave the db in an > inconsistent state with that configuration. Using the session's > implicit transaction flushed just before the return should do the safe > thing. You can look through the code later and find additional places > where you can safely flush the transaction if there's a need. We do use transactions where we can, but since most of the code is not tested at all, let alone stress tested, i can't vouch for it doing The Right Thing. (Winter break is only a few weeks away) -Yaakov _______________________________________________ Fedora-infrastructure-list mailing list Fedora-infrastructure-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/fedora-infrastructure-list
