True. The inefficient queries yesterday were returning 50,000 rows on what should have been simple queries, now they are not.
On Jul 15, 2011, at 4:57 PM, Pascal Robert wrote: > > Le 2011-07-15 à 16:52, Andrew Kinnie a écrit : > >> Greetings, >> >> I have 2 issues, the first of which I had before WOWODC, and the second >> burst into the open yesterday. >> >> I have an app, an ERRest app, acting as a push notification server, although >> thus far now push notifications have gone out to the public, it works for >> testing. The app is connected to by nearly 50,000 iOS devices which have >> our iOS app installed. The iOS app connects to Apple to get a token, then >> calls home (my server) and tries to register itself with my server so that >> when we do send out notifications, we know which device tokens want which >> types of notifications. The iOS apps on individual iOS devices connect when >> running, and send information to my server about the device, the id of the >> device, other data including the token and whether they want the 4 >> notification types to be on or off. All this works. (there are >> approximately 50,000 registered application_device rows currently, though >> we've not sent any publicly visible notifications). When the app runs, the >> devices also try and get the last notification of any type that the device >> has been set to ask for, and display that in a drop down (data pulled from >> my server) >> >> The server uses ERRest to receive this information and to create a device >> (actually an "ApplicationDevice" object) based on passed in json, checks the >> database (MySQL) to see if there is an existing row in my application_device >> entity/table for the app, and the device id (udid) and if so, updates it >> with the newly created device's data before discarding the new device and >> saving the changes in the fetched one. Otherwise, it saves the newly >> created device. >> >> However, >> >> 1. Occasionally, two devices try to be created at the same time with the >> same primary key (which is provided by the EO_PK_Table in the MySQL >> database). The second device fails with an exception. I gather this is >> because two separate devices are connecting to the server at approximately >> the same time, both asking the EO_PK_Table for the next primary key for the >> entity, then do some things, then try and save. First one succeeds, and the >> second fails because an object with the PK already exists. I was under the >> impression that EOF should handle this, but it doesn't seem to. >> >> 2. We tried to send out some test notifications yesterday, which exposed >> some inefficiencies in my model, which seem to have been resolved, but there >> are still significant issues. Apache and MySQL are fine, neither is >> pressed, but the CPU approaches 100% usage in spikes on the server (a CentOS >> VM with 16 GB RAM running Java 1.6 64 bit). Nothing else is running on it. >> The individual queries in the registration process do not return more than 5 >> rows, though there are 18 separate SQL calls caused by the registration >> method, all but one returns 1 row, and the other returns 5. In any event, >> MySQL has no slow queries. >> >> So it's the WO Java processes that seem to be taking up 100% of the CPU. >> Constantly getting worker-thread broken pipe errors when it is pressed like >> that. > > For the record, yesterday we weren't even able to use jstack to see what was > going on. But we did see working threads going up to 128. No stuck queries in > MySQL. > > _______________________________________________ Do not post admin requests to the list. They will be ignored. Webobjects-dev mailing list ([email protected]) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/webobjects-dev/archive%40mail-archive.com This email sent to [email protected]
