1) Moved AUS from Oregon to San Jose. As part of move, put the AUS machine behind a cache box, and added a 5minute timeout. This should improve load when user keeps manually doing check-for-updates rapidly. Vast majority of traffic is once-a-day automatic check-for-updates, which should be unaffected by this 5minute cache 2) As soon as AUS move happened (wed/thu of last week), the usage stats dropped bu approx 30% 3) RE-re-re-checked everything, confirmed all machines running ok. Verified that the loss of traffic is happening at the first point of contact to the network (ie the cache box). Confirmed it is not that we somehow drop the packets in our network. 4) On Sunday night, turned off the 5min timeout, so all traffic is not cached. Confirmed that traffic, including pings to the network cache box returned back to normal levels within hours of removing the timeout setting.
Confirmed drop in traffic was specific to windows. Linux, Mac traffic did not change.
Confirmed drop in traffic was across all versions. Also confirmed drop was across all locations, and across all locales.
11 years ago
There was one problem fixed about how the larger log files in the new location were causing SQL problems when loading. This problem was quickly fixed. Mentioned here for completeness.
from email@example.com we do an XMLHttpRequest for the aus url and set "Cache-Control: no-cache" on the request. See http://lxr.mozilla.org/mozilla1.8/source/toolkit/mozapps/update/src/nsUpdateService.js.in#1892 Could that be related to our problem? from firstname.lastname@example.org If I remember correctly, client headers are ignore since it could be a DoS vector.