(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
[11:00 AM] Andrew Hellershanks: Hello, everyone.
[11:00 AM] Andrew Hellershanks: I'm expecting a visitor so I may need to step out for a short while.
[11:00 AM] Jagga.Meridith @hg.zetaworlds.com: Nearly finished my huge music project, then I can get back to documenting
[11:00 AM] Gavin.Hird @grid.xmir.org:8002: it is not cached. we fixed rebaking on HG teleport without having to use an external server like FS does
[11:00 AM] Andrew Hellershanks: Jagga, what music project?
[11:01 AM] Jagga.Meridith @hg.zetaworlds.com: soundtrack to an audiobook
[11:01 AM] Andrew Hellershanks: Jagga, interesting. I don't think of soundtracks for audio books.
[11:02 AM] Jagga.Meridith @hg.zetaworlds.com: it's a full cast production
[11:02 AM] Andrew Hellershanks: Cool
[11:02 AM] Jagga.Meridith @hg.zetaworlds.com: 8 months work
[11:02 AM] Andrew Hellershanks: Hello, Lyr.
[11:02 AM] Lyr.Lobo @cc.opensimulator.org:8002: Hello Andrew *Grins*
[11:03 AM] Lyr.Lobo @cc.opensimulator.org:8002: I had to install a fresh Firestorm onto a new system. Had no idea that it does not load the grid list. Figured it out, but whew
[11:03 AM] Lyr.Lobo @cc.opensimulator.org:8002: loading them manually with apply was not working
[11:03 AM] Andrew Hellershanks: I recently had to figure out where FS stores its user IDs and passwords.
[11:04 AM] Vincent.Sylvester @hg.zetaworlds.com: No new commits this week, but devs were not lazy, well I wasn't. Squashed a viewer bug today
[11:04 AM] Gavin.Hird @grid.xmir.org:8002: viewer bug? – do we have those?
[11:04 AM] Vincent.Sylvester @hg.zetaworlds.com: Connecting differently sized regions would produce a hard crash on linux in surprisingly reproducible manner
[11:04 AM] Vincent.Sylvester @hg.zetaworlds.com: Heap access after clear
[11:05 AM] Vincent.Sylvester @hg.zetaworlds.com: TJ dug around for a while and was able to fix that
[11:05 AM] Vincent.Sylvester @hg.zetaworlds.com: Now I'm waiting for the next auto build to test cause I fought linux enough for one day trying to get nvidia to not be retarded
[11:06 AM] Vincent.Sylvester @hg.zetaworlds.com: Still wouldn't recommend going for a region config like that
[11:06 AM] Ubit Umarov: ( talking about bugs.. as you noticed there where no new bugs added to opensim during laste week.. isnt that great?)
[11:07 AM] Vincent.Sylvester @hg.zetaworlds.com: We are trying to figure out that (???).(???) thing on hg now, though apparently lucky me the lifeguard has decided to take a spin and call itself Unknown now
[11:07 AM] Vincent.Sylvester @hg.zetaworlds.com: Looks like regions cache the returns for those requests somewhere so when the bug triggers once it requires restart to trigger again
[11:08 AM] Andrew Hellershanks: Unknown? I thought that issue got fixed a while back.
[11:09 AM] Ubit Umarov: the (??) is old thing
[11:09 AM] Vincent.Sylvester @hg.zetaworlds.com: There was something done about hardening the server side against it so it wouldn't get confused what that meant, but this is something with the viewer not processing correctly
[11:09 AM] Ubit Umarov: old bug
[11:09 AM] Vincent.Sylvester @hg.zetaworlds.com: I see this every week here with the lifeguard
[11:09 AM] Ubit Umarov: i keep telling about it, and ignored... no more pacience for that
[11:10 AM] Ubit Umarov: i many cases is kinda self heels.. so whatever..
[11:10 AM] Vincent.Sylvester @hg.zetaworlds.com: It does, usually the radar also updates, though for some reason here it doesn't. Usually they pop in after ten minutes
[11:11 AM] Ubit Umarov: ( the root problem has other manifestations like failes textures etc.. )
[11:11 AM] Vincent.Sylvester @hg.zetaworlds.com: There is something funky going on, but it is something in communication getting lost, cause the data does make it out since it gets cached by the region
[11:11 AM] Vincent.Sylvester @hg.zetaworlds.com: It's probably the old async fun again or something
[11:12 AM] Vincent.Sylvester @hg.zetaworlds.com: Working on getting reliable reproduction then it shouldn't be that hard to fix :)
[11:13 AM] Vincent.Sylvester @hg.zetaworlds.com: The crashing from the bad neighbor region allocation has been around for years now, crashing every now and then on all platforms. Oddly enough I couldn't trigger it on mac or windows
[11:13 AM] Vincent.Sylvester @hg.zetaworlds.com: On linux it blew up almost instantly
[11:13 AM] Andrew Hellershanks: The lifeguard isn't showing as Unknown to me at this time.
[11:13 AM] Vincent.Sylvester @hg.zetaworlds.com: Difference in memory handling I guess
[11:14 AM] Vincent.Sylvester @hg.zetaworlds.com: Normally it comes in as ??? and eventually loads, but yeah it's different, definitely something the viewer hiccups on more than server being wrong though
[11:14 AM] Ubit Umarov: Grrrr i told here and other places and directly to some viewer devs..
[11:15 AM] Vincent.Sylvester @hg.zetaworlds.com: Once it works once when you teleport again it always shows up fine, so that makes me think timing related in some way
[11:15 AM] Ubit Umarov: during teleports, viewers keep asking to older region about things on the new region
[11:15 AM] Ubit Umarov: at sl that is fine, so they do not care
[11:16 AM] Ubit Umarov: at opensim, it fails a lot during HG...
[11:16 AM] Ubit Umarov: it also fails during normal tps, but that may one be visible on dynmaic textures and similar
[11:17 AM] Vincent.Sylvester @hg.zetaworlds.com: TJ telling me that's being looked into and should get fixed soon, since it was fixed upstream, so we'll see
[11:17 AM] Ubit Umarov: it is a timing issue.. and a bug on viewers main control than no one cars to fix, because it is really hard and not a issue at Sl
[11:18 AM] Ubit Umarov: her.. i said it again.. m Bahhh
[11:18 AM] Gavin.Hird @grid.xmir.org:8002: noted, not promising a fix
[11:18 AM] Ubit Umarov: it is hard
[11:19 AM] Vincent.Sylvester @hg.zetaworlds.com: Supposedly even LL finally realized that was a bad way of doing it so time will tell
[11:19 AM] Vincent.Sylvester @hg.zetaworlds.com: But it is being looked at now directly so good news
[11:19 AM] Vincent.Sylvester @hg.zetaworlds.com: Some things just take longer :)
[11:19 AM] Andrew Hellershanks: :)
[11:20 AM] Ubit Umarov: at sl does not matter.. all assets and names affected are global, so region does not matter
[11:20 AM] Ubit Umarov: about viewer crashs... i have no reports
[11:20 AM] Vincent.Sylvester @hg.zetaworlds.com: Still kinda seems like a backwards to leave a region and as parting gift bug it about some more data heh
[11:21 AM] Ubit Umarov: possilble still bad handling of large regions
[11:21 AM] Gavin.Hird @grid.xmir.org:8002: timing of when to start asking the other
[11:21 AM] Vincent.Sylvester @hg.zetaworlds.com: Gotta love async *starts shaking*
[11:23 AM] Gavin.Hird @grid.xmir.org:8002: can't stop asking the old till we are absolutely sure we are properly connected to the new or most likely there will be no reliable fallback if the TP fails
[11:24 AM] Gavin.Hird @grid.xmir.org:8002: on HG TP I mean
[11:24 AM] Vincent.Sylvester @hg.zetaworlds.com: I have only ever seen this behavior on hg or when going to heavy regions that respond more slowly so it tracks this is just down to taking too long. Always fun
[11:24 AM] Ubit Umarov: i did add delays on region side. only start sending things after seeing viewer fetch caps.. etc..
[11:25 AM] Ubit Umarov: more delays and things break bc other reasons
[11:25 AM] Vincent.Sylvester @hg.zetaworlds.com: The only odd thing about that is it seems on the second teleport it works flawlessly even if you restart the viewer. It's like the region has the data at the ready faster
[11:26 AM] Vincent.Sylvester @hg.zetaworlds.com: No doubt there is some caching somewhere maybe even underneath in dotnet or something, cause simulator restart triggers it again
[11:26 AM] Vincent.Sylvester @hg.zetaworlds.com: But we'll look into, hopefully get it fixed at some point in near future
[11:28 AM] Vincent.Sylvester @hg.zetaworlds.com: Rolled out some new binaries yesterday to track that asset cache overflowing a little better. Checking data today and nothing makes sense. The in use asset only counts a few mb in size while the folder has 26gb in it, so either my math is off or the counting routine is not counting the correct things
[11:29 AM] Vincent.Sylvester @hg.zetaworlds.com: The good thing is that it doesn't leak, it eventually goes back down
[11:29 AM] Vincent.Sylvester @hg.zetaworlds.com: Kinda goes up and down over time
[11:30 AM] Ubit Umarov: ( a not to casual reader> Vincent.Sylvester did release binaries to his own grid Zetaworlds, not opensim core )
[11:30 AM] Ubit Umarov: ( a note ... )
[11:31 AM] Vincent.Sylvester @hg.zetaworlds.com: Going through the code I started a rework of the caching system aiming to reduce asset server load on re-fetching when simulators share the data since they can interfere with each other. Since sharing is useful for deduplication
[11:32 AM] Vincent.Sylvester @hg.zetaworlds.com: Combined with better handling of dynamic textures that should largely cut the asset server load in half
[11:32 AM] Vincent.Sylvester @hg.zetaworlds.com: The latter might even finally allow for storing the creation string of the textures somewhere for later editing as it was meant to be
[11:33 AM] Vincent.Sylvester @hg.zetaworlds.com: But that is still distant, makes my brain hurt trying to figure out how to do that without nuking the local disk with db writes
[11:33 AM] Vincent.Sylvester @hg.zetaworlds.com: Tradeoffs
[11:33 AM] Ubit Umarov: by default dyn textures do not get into assets services, they are LOCAL
[11:33 AM] Vincent.Sylvester @hg.zetaworlds.com: Yes it's when the local cache fetch request fails they burden the asset server
[11:34 AM] Vincent.Sylvester @hg.zetaworlds.com: There is a small chance the asset server then might even return data if the uuids collide somehow
[11:34 AM] Vincent.Sylvester @hg.zetaworlds.com: Imagine it trying to render a mesh as texture
[11:35 AM] Vincent.Sylvester @hg.zetaworlds.com: Thankfully it shouldn't based on type
[11:35 AM] Ubit Umarov: if you do that then you get what you deserve :p
[11:36 AM] Vincent.Sylvester @hg.zetaworlds.com: Best option would be to have each face on a prim store a bit about whether the texture is dynamic or not, so a request for its texture can be canceled before it hits assets if cache fails, but that would mean changes to libomv
[11:36 AM] Ubit Umarov: dyn textures have some nasty problems.. lost from cache is not the worse of those :(
[11:37 AM] Vincent.Sylvester @hg.zetaworlds.com: Also another bit on each face does mean more memory and storage
[11:37 AM] Vincent.Sylvester @hg.zetaworlds.com: Have to scratch my head some more on that
[11:37 AM] Vincent.Sylvester @hg.zetaworlds.com: Solve one problem create another if not careful with this
[11:38 AM] Vincent.Sylvester @hg.zetaworlds.com: (The bigger worry is a colliding uuid returning a texture of something you may not want to show publicly)
[11:39 AM] Ubit Umarov: you let users type uuids..
[11:39 AM] Gavin.Hird @grid.xmir.org:8002: :-)
[11:39 AM] Vincent.Sylvester @hg.zetaworlds.com: No, but uuids are not 100% free of collisions
[11:40 AM] Ubit Umarov: this days ... almost free
[11:40 AM] Vincent.Sylvester @hg.zetaworlds.com: You know how cursed I am, if it can happen it will
[11:40 AM] Ubit Umarov: if you let the tools create them
[11:40 AM] Ubit Umarov: and keep using same uuid type
[11:40 AM] Ubit Umarov: ( think we use type 4 )
[11:42 AM] Ubit Umarov: As noted by Wikipedia, “the number of random version-4 UUIDs which need to be generated in order to have a 50% probability of at least one collision is 2.71 quintillion.”
[11:42 AM] Ubit Umarov: ..
[11:42 AM] Ubit Umarov: You would need to create 1 billion IDs every second for 86 years to hit a 50% chance of a single duplicate.
[11:42 AM] Ubit Umarov: ;;;
[11:42 AM] Ubit Umarov: so says google ai
[11:44 AM] Ubit Umarov: ( note that basic random numbers gens are bad... tools use more complex ones now )
[11:44 AM] Vincent.Sylvester @hg.zetaworlds.com: Chance just waits until you least expect and it then drops a piano on you
[11:45 AM] Ubit Umarov: ask dino about his grandgrand..grandparents
[11:46 AM] Vincent.Sylvester @hg.zetaworlds.com: Don't jinx it now xD
[11:49 AM] Andrew Hellershanks: hehe
[11:50 AM] Andrew Hellershanks: We are at 10 minutes before the top of the hour. Any other last minute items for today?
[11:50 AM] Gavin.Hird @grid.xmir.org:8002: Have done quite a bit of progress on the Mac viewer this week – all published. I will list
[11:50 AM] Ubit Umarov: :)
[11:50 AM] Lyr.Lobo @cc.opensimulator.org:8002: Great, Gavin
[11:50 AM] Ubit Umarov: fired claude?
[11:50 AM] Vincent.Sylvester @hg.zetaworlds.com: lol
[11:51 AM] Andrew Hellershanks: Sounds good, Gavin.
[11:51 AM] Gavin.Hird @grid.xmir.org:8002: On HG teleports we now detect an avatar rebake is needed without reaching for an external server like FS does
[11:51 AM] Lyr.Lobo @cc.opensimulator.org:8002: that's good
[11:52 AM] Gavin.Hird @grid.xmir.org:8002: the viewer now supports sound input from all system supported audio formats, converts them to the special wav format that the ogg decoder needs on upload
[11:52 AM] Gavin.Hird @grid.xmir.org:8002: so you can pretty much take any audio file and upload
[11:53 AM] Gavin.Hird @grid.xmir.org:8002: you can chose between streaming audio with the FMOD or VLC plugins and the VLC stream will do track announcements pretty much like the FMOD does
[11:53 AM] Gavin.Hird @grid.xmir.org:8002: audio decoding is not happening on as many cores as the system will allow you to use rather than on one thread
[11:54 AM] Vincent.Sylvester @hg.zetaworlds.com: That's just audio, video still multi core?
[11:54 AM] Gavin.Hird @grid.xmir.org:8002: and we found that Claude can write, test/compile and debug lsl scripts
[11:55 AM] Gavin.Hird @grid.xmir.org:8002: video is handled by the VLC plugin How multi-treaded that is I don't know
[11:55 AM] Vincent.Sylvester @hg.zetaworlds.com: Ok
[11:56 AM] Gavin.Hird @grid.xmir.org:8002: is now it should be and not is not
[11:57 AM] Ubit Umarov: no idea why we type not instead of now.. i keep doing it all the time pff
[11:57 AM] Vincent.Sylvester @hg.zetaworlds.com: The rtmp side of things has been a bit glitchy, but it is the fastest protocol, hls is supported, but caches an entire block which can mean delays of a minute while rtmp can be as low as 15 seconds
[11:57 AM] Vincent.Sylvester @hg.zetaworlds.com: That's mostly on vlc though
[11:58 AM] Gavin.Hird @grid.xmir.org:8002: we also found that the current viewers could block audio decoding for up to 5 minutes if a audio file requested never replied, so we fixed that too
[11:59 AM] Vincent.Sylvester @hg.zetaworlds.com: Reminds me. Testing that neighbor bug on linux today I could not get FS to start at first cause the machine had no audio (nuked wireplumber etc. cause system instability) and it just got hard stuck loading fmod and pulseaudio, had to --noaudio to get it started
[12:01 PM] Andrew Hellershanks: Interesting. Didn't know about that option. Never had the program hang up due to audio issue.
[12:01 PM] Vincent.Sylvester @hg.zetaworlds.com: After the damn nvidia driver nonsense I wasn't in the mood to debug that
[12:02 PM] Andrew Hellershanks: I totally understand. I've been battling issues with an embedded system for the past couple of weeks. Dealing with lots of headaches.
[12:03 PM] Andrew Hellershanks: Last call for any other topics or comments for today. If there are none then I will wrap up this meeting.
[12:03 PM] Andrew Hellershanks: Seeing nothing more, that does it for this week. Thank you all for coming. See you again next week.
