Testing of new double region crossing fixes by Pepper Linden
Log In
AlettaMondragon Resident
Mercury Linden Pepper Linden Maestro Linden Is there any news about this lately? Does the Blake Sea test look good enough so that a large-scale implementation can be rolled out? Even if not the entire grid yet, but entire Heterocera where many of the worst region corner crossings on roads are and many railroad region corner crossings, too? It would be nice to see what happens there with these fixes.
Maestro Linden
AlettaMondragon Resident: Funny you should ask! We have actually incorporated the Blake Sea region crossing changes into the next release, which went out to the 'Second Life RC Ferrari' channel this morning:
If all goes well, the changes will go to many more regions over the next 2 weeks.
Giante Waffle
We did another long test today.
We set up a boat with an auto-steering script to help it navigate a course through 22 sims in Blake Sea, then let it run for the past 10 hours with 2–4 passengers as people came and went. The speed was low at 7 kts, but it averaged about one sim crossing per minute for the entire test. Nobody got disconnected, and no boats were lost. The only funky thing that happened was that one passenger’s head vanished at some point.
I feel like this type of testing is more in line with how we actually use vehicles in Second Life. Nobody really heads out with 10+ people on a boat at full throttle, because that’s never worked-- the fact that such scenarios are even survivable now is awesome. For most of us, we're out boating or driving with four people or less, moving along at a low to moderate speed as we're chatting and playing a fishing game or doing GTFO/GFN deliveries. This patch is a major improvement for those scenarios.
edit: I ended up running this for 17 hours without issue before calling it a day.
animats Resident
Re previous mention of region crossing problem that only occurred with Sharpview. See: https://feedback.secondlife.com/server-bugs/p/intermittent-timing-race-with-event-poller
This should not block deployment of the new region crossing fixes.
It's something to be addressed the next time someone has to work on the event poller, either viewer or server side.
Jelly Doll
Took part in Eris Ravenwood's insane bus crossing adventure and I have a few thoughts ...
Bumping off a region border as avatars are migrated / crash is not a pleasant experience. The old system was better when it worked, interpolating the client while the server caught up was only ever bad when the servers dropped the ball.
The new code seems to treat every passenger as a separate entity causing some back and forth to happen as some avatars get transferred and then yanked back during a stall.
A vech and all it's passengers needs to be treated as a single entity, either everyone makes it all in one go, or no one does.
By the end, my avatar was utterly wrecked, missing many attachments that did not come back following a relog.
As an end user - It should feel seamless and smooth, even if I'm doing donuts on the region border. I shouldn't have to care about the map grid more than I care about the experience.
AlettaMondragon Resident
Jelly Doll "when it worked" Most of the time it did not.
Other than that, yes it should be smooth and reliable, but if Eris had up to 20 people on that flying bus, it is a miracle that it even worked the way it did. I think the most people on Animats' test train on the beta grid were 7, but in two cars, so their crossings were distributed. I don't think anyone had even 10 people sitting on the same objects to test this. But it's nice that this group of you did this test and I agree that crossings and the entire experience should be nice even with so many people in a vehicle. Definitely something LL should aim for.
I had two weird, severe attachment ghosting issues on the beta grid after hundreds of corner crossings on the test train, but then it didn't happen again on other days, but it looks like that issue wasn't eliminated yet if it happened to you now.
Eris Ravenwood
Tested in the Blake Sea with about 15-20 people on a flying bus.
We started at 16:00 SLT in the Blake Sea - Turnbuckle region, and continued northeast -> east to Blake Sea - Bering, then turned around, and made it back to Blake Sea - Crow's Nest by 16:45 SLT.
Overall, results were good. We did have several cases where we were bashing up against a sim crossing for up to a couple mins before crossing successfully.
As a test, we decided to also cross into the Full Private Regions, Blake Diego and Blake Roatan, which don't have the fix. Results there were disastrous. While we were "able" to cross, half of us ended up underwater, off the bus, and bugged out.
I don't know how much of the "invisible wall bumping" was due to simply crossing into homestead regions with a bus full of 20 people (and homesteads have a max of 25, 20 for non-premium residents), or the new crossing code. Either way, it does seem like the crossings were improved. I hope y'all have logs lol.
I asked riders to also share feedback, since we did have a few people get kicked off the bus or crash, but I wasn't tracking everyone.
Photo Viewer
View photos in a modal
animats Resident
Eris Ravenwood > We did have several cases where we were bashing up against a sim crossing for up to a couple mins before crossing successfully.
Right. An implication of Pepper's fix is that you can't start a second crossing until all the avatars have completed the previous crossing. There's a point in a region crossing where the sim server is waiting for the viewer to send CompleteAgentMovement for each avatar. The slowest viewer/avatar connection stalls the second region crossing. Previously, in that situation you'd lose an avatar at the crossing, so this is an improvement.
DiomedesMarteens Resident
Did a test with several other avatars, the only issue i have notice is a bit of rubberbanding when going with fast vehicles ([sau]Kazu jetski), which sometimes turns into real invisible wall blocks when there are several avatars and one of them has not completed the transition yet (although i suppose this is intended behaviour). I hope this helps
Nelson Jenkins
Testing on Agni with an experimental plane with a ton of scripts and a bunch of scripted attachments, no disconnection issues but control does stall out on crossings as often as it used to, which causes stuck controls that send the plane... out of control. This is probably not as noticeable in boats and ground vehicles since there's no need for constant pitch corrections. There were also a lot of scripts moving around, so this is probably not as common of a problem as I intentionally made it.
There is also the issue that bumping head-on into an "invisible wall" causes the vehicle to basically stop dead-on, which for planes, kills lift and causes it to immediately stall and start spiraling out of control (at least in my model). Obviously not a concern for other vehicles, and is tolerable if approaching at a sharply oblique angle (though that's also an atypical flight path across borders in general). I don't think there is an obvious architectural solution for this, but I can't think of a way to reliably mitigate it in a script off the top of my head, either.
AlettaMondragon Resident
Nelson Jenkins Maybe the stuck controls could be avoided if the sim simply dropped the control input on the crossing no matter what, but then what direction should the vehicle keep going? Straight ahead based on the last vector when the sim dropped the control input? Normally, most of us just don't press any controls right before each crossing and that works, even with aircraft that might need "constant pitch corrections".
Donovan Michalski
I'm now on a boat with three other avatars, so four of us total - the boat is The Mesh Shop's "Blue Marlin".
Everyone is surviving the crossings just fine, though I'll admit the 'invisible wall' part is a bit jarring. With four of us on the boat, all crossings are taking roughly 5 to 7 seconds to complete.
Donovan Michalski
Just to follow up on this - looks like it was a viewer issue. One of the avatars was logged in through Radegast during the initial test. When all four were logged in using Firestorm, the crossing performance was much improved and the invisible wall issue went away. Crossing times dropped significantly, 1 to 3 seconds on average.
Neon Zombie
I've spent the last hour out here in the blake sea trying everything I could to crash or get disconnected and so far its been really solid. I even setup a stress test forcing my character quickly between region crossings and its still going 20 mins later. Great work so far. from what I've seen its only when you have a large group of people on the same vehical crossing at the same time that it hangs hardest but no disconnects.
Artemis Elara
Yeah, been driving around testing it out and it's great. 50 X crossings later and I was still alive! we might have a future where we don't have to keep our eyes pinned on the mini map!
Load More
→