We’ll be upgrading the server systems making up Second Life on Thursday, November 8th. This update has a number of bug fixes and some new functionality.
The upgrade will not require any planned service outages. No new viewer will be required.
There may be slight delays to log-ins and to online notifications while some central systems are updated, which is planned for the morning (10:00am PST, 18:00 GMT). This will be followed by a rolling restart to update all regions with the new code. The rolling restart is expected to begin around 1:00pm PST (21:00 GMT) and last about 4 hours. Each region will be restarted in a wave that progresses North to South across the map, with a 5-minute warning preceding the restart.
- llGiveMoney() will now be throttled to limit disruptive effects on servers
- Users with 1.18.4 viewers will see friend glyphs in a different color in the minimap
- SVC-94: Build system changes rotations slightly at random (including when “Edit linked parts” is on)
- SVC-247: Failed attempt to rez an object from trash results in content loss
- SVC-583: Five minute wait due to “System Logging You Out” happens frequently with scripted attachments
- SVC-912: Sky Eclipse’s avatar crashes regions EVERYTIME she logs in, regardless of where she logs in
- Fix: Can once again set For Sale on an inventory item via right-click Properties
[Update, 2007-11-06 @ 9:30 am]
Quick answers to some of the questions that came up in the comments
- llGiveMoney() throttling is now documented in the wiki
- “friend glyphs” means the resident indicators that show in the minimap, for friends. Historically, all of the glyphs were green dots (same altitude) or a T/inverted-T (above/below). As of the 1.18.4 viewer they’re dots, up or down chevrons. After this server update, the viewer will receive enough information to show friends in yellow, everyone else in green.
- SVC-94: There are several similar issues with “prim drift”; this fix addresses some but almost certainly not all of them.
- SVC-583: Q Linden has some excellent comments, so I defer to his expertise – see comment #99
- SVC-912: The crash was due to corrupted data in an attachment. Once this was diagnosed it was easy to have Sky remove the attachment, and track down the place in the server code that wasn’t doing proper checks.
[Update, 2007-11-06 @ 2:45 pm]
Well, that was fun.
We started the updates Tuesday @ 10am as originally planned. About an hour into it we were updating the systems that keep track of who’s where in-world (“agent presence”). At around 11am, an issue that didn’t show up in testing appeared which stopped the “login” servers from talking to the “presence” servers. Unfortunately, it also interfered with our monitoring systems so it took several minutes to notice the problem. We investigated, hoping for a quick fix, but since this was not forthcoming we rolled back the changes which fortunately worked. (Rollbacks are not always a slam-dunk.) This also affected the PJIRA logins, but in a different way, and required rollbacks on a different host.
We are still trying to track down the cause of the problem. We’re tentatively scheduling the continuation of the update for tomorrow (central servers in the morning, rolling restart in the afternoon), but of course will be doing so with even more extreme paranoia.
[Update, 2007-11-07 @ 11:15 am]
We believe the source of the problems has been identified. We’ll be testing the fixes throughout the day, and plan to perform the server updates on Thursday. Same time frame – central service updates in the morning, rolling restart in the afternoon.