For at least last week or so, bugzilla has been inaccessible at random times, sometimes just for a few minutes, once or twice even for a couple of hours. This is causing lots of duplicate bugs when people are trying to resubmit their bug reports, and also wastes bug wranglers' and developers' time. Could infra please investigate what's wrong here? Needs some a solution as soon as possible. Thanks! Reproducible: Sometimes Steps to Reproduce:
jakub, the db behind bugs/forums has been getting slammed beause of the increasing number of people using it. we're not the only project using that one database box, kerneltrap is one example of a party who is sharing our db box. i believe our host, the OSL, is working on a replacement to off load some of it. but for right now, there's not much we can do :/ -jeffrey
Are we even sure its db1 thats the primary cause of the problem? Forums is a huge db1 user and I haven't heard any complaints about its speed. We have been dealing with DoS issues lately on that box and that could be part of the problem. I've noticed that its been happening more doing the middle of night for me. Is that when you've noticed the problems, or is it any time during the day? Please let me know a more specific timeframe of these problems. Thanks-
I personally have experienced it during the day (3-6PM EST). Its normally unresponsive to queries for a couple of minutes and then comes back.
havent heard any reports of it being slow lately, so closing it for now.
Closing this bug seems to have disastrous effect on bugzilla; http is dead, https really slow. Same for packages.g.o. :( Reopen.
After doing some looking, I'm thinking the problem is more drawn towards packages.g.o which is hosted on the same box. Been having problems of RSS readers hammering the box and/or spiders doing the same. We have been looking into ways to limit that. We may just need to adjust the robots.txt also for that site.
Marduk: Do you have a better robots.txt we could put on packages.g.o to limit search engine's affect on the site?
lance, you mentioned about people nailing packages with RSS. does RSS run over the http protocol? i'm just thinking, can we put limits on the number of spawned apache processes for just that vhost (packages.g.o) so that it doesnt overwhelm nuthatch? is that not the problem?
think this was fixed with lance limiting the memory bugzilla can eat. closing for now.
