Once we switch the release and esr channels to balrog we should be pretty close to being able to kill it. There will be a small amount of work to do still, at least the following: * Redirect all aus/aus2/aus3 requests to aus4. * Switch in-app update server to aus4
Jake gave me some logs over in bug 1133934. From them I grepped away things that were already redirected and silly requests (such as requests for /), which gave me about ~8,000 unredirect requests. The vast majority of those were on the "default" channel. After removing this, the only thing left is (balrog)➜ logs cat * | grep "/update/./" | grep -v '/beta/' | grep -v '/beta-localtest/' | grep -v '/beta-cdntest/' | grep -v '/release' | grep -v '/release-localtest/' | grep -v '/release-cdntest' | grep -v '/esr' | grep -v '/esr-localtest/' | grep -v '/esr-cdntest' | grep -v default | grep -v "/beta-cck" | grep -v "/release-cck" | grep -v "/nightly/" | grep -v "/aurora" 188.8.131.52 aus3.mozilla.org - [27/Feb/2015:10:28:57 -0800] "GET /update/3/Firefox/14.0.1/20120713134347/WINNT_x86-msvc/en-US/false-cck-mozilla14/Windows_NT%184.108.40.206%20(x64)/mozilla14/1.0/update.xml HTTP/1.1" 200 496 "-" "Mozilla/5.0 (Windows NT 6.1; WOW64; rv:14.0) Gecko/20100101 Firefox/14.0.1" "-" "DNT:-" 220.127.116.11 aus3.mozilla.org - [27/Feb/2015:11:07:27 -0800] "GET /update/3/Firefox/35.0.1/20150122214805/WINNT_x86-msvc/x86 HTTP/1.0" 200 497 "-" "curl/7.30.0" "-" "DNT:-" So, false-cck-mozilla14 is an invalid channel. And the second request looks like an Avast one that Zeus doesn't log properly. Given this, I think we're fine to redirect all AUS traffic to aus4 anytime now. It would be ideal to wait until we support /update/1 (bug 1026827), but not strictly necessary.
bug 998721 is a nice to have update path. We'll get to it, but it doesn't block killing aus3.
My understanding is that until aus3 is decommissioned, it will still be using configurations from cvs. So, blocks bug 985022
(In reply to Hal Wine [:hwine] (use needinfo) from comment #3) > My understanding is that until aus3 is decommissioned, it will still be > using configurations from cvs. So, blocks bug 985022 We discussed this in IRC, but for posterity: AFAIK, aus3 does not need CVS around to operate, it only needs it around when we change the code or config. We are NOT doing either of those things anymore, therefore I don't consider this to block the CVS death. This _should_ be done this week, but it's highly dependent on WebOps. I'll leave the dependency, but I don't think this actually blocks CVS death.
I don't think bug 998721 blocks this. Those requests were already redirected to aus4...
All traffic has been redirected Balrog. aus3 is dead as far as RelEng is concerned. IT said they want to keep the machines around a bit longer, but there is no chance we will be using them - we don't have the ability to generate update snippets anymore.