Chat log from the meeting on 2014-09-02

From OpenSimulator

Jump to: navigation, search
[11:03]  Nebadon Izumi: nice
[11:03]  Nebadon Izumi:*/
[11:03]  BlueWall.Slade And it seemed that after a process crashed and went zombie, the rest were there, hanging
[11:03]  Justin Clark-Casey: Urgh, I spent 1.5 days looking for a bug only to just discover it was probably coming from debug code I inserted in OpenMetaverse.dll and accidentally introduced into master opensim
[11:03]  Justin Clark-Casey: :(
[11:04]  Nebadon Izumi: that sucks
[11:04]  BlueWall.Slade ahhhh
[11:04]  Richardus.Raymaker hi justin
[11:04]  Justin Clark-Casey: it sucks big balls
[11:05]  BlueWall.Slade
[11:05]  Richardus.Raymaker yiks ?
[11:05]  vegaslon.plutonian the camera on vehicles is messed up now since commit r/25154
[11:05]  BlueWall.Slade what was the consensus about that - is it still useful?
[11:06]  Justin Clark-Casey: interesting, I never knew that was a major reason
[11:06]  Justin Clark-Casey: My opinion is that it causes more difficulty than benefit now
[11:07]  BlueWall.Slade do you want me to start removing it?
[11:07]  Justin Clark-Casey: yes, though I would prefer not to start removing it and then stop unless there's a very good reason
[11:08]  BlueWall.Slade should we take it to the mailing lists?
[11:08]  Justin Clark-Casey: tbh, I would just remove it and see if anyone complains
[11:08]  Justin Clark-Casey: it can always be reinstated
[11:08]  BlueWall.Slade I will study it before starting to remove it.
[11:08]  BlueWall.Slade I think is is probably not that much
[11:08]  Justin Clark-Casey: it's just a shell which forces execution in 32-bit mode, nothing more afair
[11:08]  BlueWall.Slade If I start it, I'll finish
[11:08]  Justin Clark-Casey: and doesn't appea rto work properly if you actually are 32-bit...
[11:08]  Justin Clark-Casey nods
[11:08]  Justin Clark-Casey: bluewall: great
[11:09]  BlueWall.Slade ok, on my TODO list
[11:09]  BlueWall.Slade [11:11] vegaslon.plutonian the camera on vehicles is messed up now since commit r/25154
[11:10]  Justin Clark-Casey: yeah, which commit is r/25154?
[11:10]  Justin Clark-Casey: I expect that was one of the friday changes
[11:10]  vegaslon.plutonian ya the very last one
[11:10]  Justin Clark-Casey: for sitting avatars reutrning velocity?
[11:10]  Justin Clark-Casey: 473c55?
[11:10]  Nebadon Izumi: oh ya i meant to test that
[11:10]  Nebadon Izumi: never did
[11:10]  Mata Hari: that's the very last commit...odd that it would affect camera
[11:11]  Justin Clark-Casey: I expect cameria is assuming something about vel settings that i didn't realize
[11:11]  Justin Clark-Casey: vegaslon: could you confirm that's the one?
[11:11]  vegaslon.plutonian it effects camera becouse the veiwer is getting two versions of velocity now and thinks the avatar is moving
[11:11]  Nebadon Izumi: kind of makes sense
[11:11]  Nebadon Izumi: you were trying to fix that problem where if you were still walking when you sat
[11:11]  Justin Clark-Casey: vegaslon: ok thanks, I will either revert or fix soon
[11:11]  Nebadon Izumi: your avatar would keep its velocity?
[11:12]  Justin Clark-Casey: nebadon: that change was because script functions didn't return velocity properly when avatar was on vehicle
[11:12]  Justin Clark-Casey: was always 0
[11:12]  Justin Clark-Casey: but clearly something is relying on it being zero, ho hum
[11:12]  Nebadon Izumi: ah ok
[11:12]  Justin Clark-Casey: vegaslon: thanks for the info
[11:14]  Justin Clark-Casey: any other opensim topics today?
[11:15]  Dahlia.Trimble hi
[11:15]  Mata Hari: it's been interesting using HG standalone for the last couple of weeks....reminded me of a number of bugs I reported 12-15 months ago that are still broken
[11:15]  Justin Clark-Casey: hi dahlia
[11:15]  Justin Clark-Casey: there aren't enough developers
[11:15]  Richardus.Raymaker hi Mata, more ductape
[11:15]  Mata Hari: things like online/offline HG friends notifications that appear in console but never get transmitted to the viewer
[11:16]  Mata Hari: or standing in the same region as an HG friend and your IMs to one another are all prefixed with the (Saved <date time) even though they're being transmitted instantly
[11:17]  Dahlia.Trimble "patches are welcome" :)
[11:17]  Billy.Bradshaw I have noticed that during the HG friend process, 2 calling cards are created.
[11:17]  Richardus.Raymaker hi dahlia. yes mata now i see. i have soemtimes problems that IM word one way fine but sending it back gives offline message
[11:17]  Mata Hari: you're probably prefer to receive them from someone with a passing familiarity with C#
[11:18]  Justin Clark-Casey: tbh, if it compiles and helps the problem then it can always be refined
[11:18]  BlueWall.Slade Mata - do you have a Hypergrid?
[11:19]  Mata Hari: also on login to my own regions my viewer is stuck showing "Fetching" even though there's 0 http traffic
[11:19]  Justin Clark-Casey: the hardest part can be finding exactly where the issue is.
[11:19]  Mata Hari: but if I empty trash it instantly clears
[11:19]  Mata Hari: and will clear even if my trash is empty
[11:19]  Justin Clark-Casey: Yes, there are bugs
[11:19]  Justin Clark-Casey: look in mantis and you'll see thousands of tehm
[11:20]  Mata Hari: that one could conceivably be related to the other main inventory fetch bug since OSG handles the trash folder differently than some other grids
[11:20]  Nebadon Izumi: that seems unlikely
[11:20]  Billy.Bradshaw Can anyone explain how asset duplication occurs, if it is convoluted I'll pass.
[11:21]  Justin Clark-Casey: Billy: how do you mean?
[11:21]  Billy.Bradshaw Keep hearing about the de-duping?
[11:21]  BlueWall.Slade 100% same content with different uuid
[11:22]  Nebadon Izumi: deduping works by doing a hash of the asset
[11:22]  Nebadon Izumi: usually SHA256 hash
[11:22]  Nebadon Izumi: and comparing all new assets with hash
[11:22]  BlueWall.Slade importing objects
[11:22]  Justin Clark-Casey: billy: there are ways in which it occurs, thuogh I couldn't detail all of them 100%
[11:22]  Billy.Bradshaw Bluewall. that is the answer
[11:22]  Nebadon Izumi: anything that matches an existing asset gets shed and only a new reference is created to the already existing asset
[11:22]  Andrew.Hellershanks Billy, upload the same image twice (for example). You have two asset records in the database with the same data blob.
[11:22]  Nebadon Izumi: the chance of 2 assets having same hash that are not identical is like a trillion to one odds
[11:22]  BlueWall.Slade if several users import simliar object, they may create collisions
[11:22]  Billy.Bradshaw I wondered how 2 identical UUID could exist
[11:22]  Justin Clark-Casey: that's a good example
[11:23]  Justin Clark-Casey: OAR import may actually use the same asset in many cases
[11:23]  Billy.Bradshaw kk
[11:23]  Justin Clark-Casey: so there's a small level of avoidance of dupes, but it's not great
[11:23]  Andrew.Hellershanks justin, I've often wondered how OAR import works re: loading assets. I was hoping it would check for existing asset before creating a new one.
[11:23]  Dahlia.Trimble I dont think the default asset server hashes
[11:24]  Justin Clark-Casey: andrew: assets have their original UUID in OARs
[11:24]  Andrew.Hellershanks It doesn't, AFAIK.
[11:24]  Andrew.Hellershanks That last comment of mine was for Dahlia.
[11:24]  Justin Clark-Casey: so if it's imported and an asset exists with the same UUID, then that is used instead. Which is not a great approach but at the time the only alternative was to insert yet another copy of the asset
[11:24]  Richardus.Raymaker always where thinking oar checks with asset server if asset alread exists , means same uuid = no upload
[11:24]  Justin Clark-Casey: which would have been probably worst
[11:24]  Justin Clark-Casey: worse
[11:25]  Andrew.Hellershanks Richardus, I was hoping it does that.
[11:25]  Justin Clark-Casey: in theory, you could use the hash of the asset also as its UUID directly
[11:25]  Andrew.Hellershanks I've been trying to decide if I switch a grid to SRAS or whether to wait for some built-in deduping of assets.
[11:26]  Justin Clark-Casey: the block to switching to built in, for me, is working out the complete migration strategy
[11:26]  Justin Clark-Casey: it actually will already auto-migrate as you go along - the question is whether that's sufficient or if there needs to be a final command to complete migration for rarely accessed assets
[11:26]  Dahlia.Trimble theres a hashing routine in the materials module that would probably work for other assets
[11:26]  Andrew.Hellershanks I have a migration method for my version of a debuping system.
[11:27]  Andrew.Hellershanks s/debuping/deduping/
[11:28]  Andrew.Hellershanks The time consuming part is the exporting of the asset blobs to disk.
[11:30]  Andrew.Hellershanks makes a note to update his asset blobs to disk patch to make sure it can be applied to 0.8
[11:30]  Justin Clark-Casey: well, I did get my local HG working again, after puzzling through some insanity in the code
[11:31]  BlueWall.Slade lokl
[11:31]  Justin Clark-Casey: it turns out that things don't work very well if you have your simulator configured for HG but not your robust...
[11:31]  BlueWall.Slade what did you find in there JCC?
[11:31]  Justin Clark-Casey: but because of the insane code it's difficult to actually notify the user of that fact in a sane way
[11:31]  BlueWall.Slade Ahh
[11:31]  Richardus.Raymaker yeah, configure iot right or you have fun
[11:31]  BlueWall.Slade yes, last week I had the errors popping up because of a mis-config
[11:32]  Justin Clark-Casey: amazingly, some code assumes that a lack of certain return data signals success and not failure. So if the HG tries to contact a robust endpoint that simply isn't there, it apepars to assume a form of success...
[11:32]  Justin Clark-Casey sighs
[11:32]  Andrew.Hellershanks Justin, couldn't you have the instance make a query to Robust during start up to check that both are HG or not HG?
[11:32]  Justin Clark-Casey: just how many testers are you loading on today neb?
[11:32]  BlueWall.Slade is there a quick way to drop attempts to reach a misconfigured url?
[11:32]  Justin Clark-Casey: I think it depends on the misconfiguration
[11:32]  Nebadon Izumi: i was hoping for 80
[11:32]  Andrew.Hellershanks Justin, Could it be done during the HG version check ??
[11:33]  BlueWall.Slade like mine last week
[11:33]  Richardus.Raymaker i have still the problem that i used in the past a wrong url for objects. and i dont see a way to fix it
[11:33]  Justin Clark-Casey: andrew: maybe, though I'm always cautious about introducing more service calls
[11:33]  Dahlia.Trimble is Online
[11:33]  Richardus.Raymaker wrong \ i think
[11:33]  Andrew.Hellershanks Justin: would it need a new call or just extra piece of info returned from an existing call/check?
[11:33]  BlueWall.Slade maybe blacklist a url after 3-5 attempts ?
[11:33]  Justin Clark-Casey: andrew: I don't know, it's not something I have had the time to think about
[11:34]  Andrew.Hellershanks Justin, np. Just throwing out some ideas.
[11:34]  Justin Clark-Casey: I just spent an age chasing down ghosts in UDP acking :(
[11:34]  Justin Clark-Casey: though hopefully I have found why the keynotes had enormous outbound resend counts
[11:35]  Justin Clark-Casey: I was just super annoyed that opensim didn't report more obviously what was going wrong, and then spent a few hours invesitgating just to find that changing that would involve checking a whole bunch of other stuff :(
[11:39]  Justin Clark-Casey: anyway, anything else to discuss today?
[11:39]  Nebadon Izumi: ok
[11:39]  Nebadon Izumi: all 80 bots are in
[11:39]  Nebadon Izumi: i was able to get all 80 on 1 machine
[11:39]  Mata Hari: one thing I've noticed with HG tping is that there seems to be a ton of data that has to get sent in a very short period of time or the tp will fail
[11:40]  Justin Clark-Casey: nebadon: trying for a new record?
[11:40]  Mata Hari: but if you then immediately re-try the tp it works instantly
[11:40]  Justin Clark-Casey: mata: attachment related?
[11:40]  Mata Hari: prossibly
[11:40]  Andrew.Hellershanks Mata, Might be a timer in the viewer end that is too short.
[11:40]  Nebadon Izumi: well previously Justin
[11:41]  Nebadon Izumi: when i was using Ragegast to sit my avatars
[11:41]  Mata Hari: this was also happing with inbound people to my region who had very little in the way of attachments though
[11:41]  Justin Clark-Casey: there is some sub-optimal stuff going on with HG as well when parsing XML, though not quite as bad as the previous attachments insanity
[11:41]  Nebadon Izumi: i would get to around 50 or 60 and it became impossible to sit them anymore
[11:41]  Dahlia.Trimble ya Ive noticed that too
[11:41]  Nebadon Izumi: now that I dont have to sit them with the client
[11:41]  Nebadon Izumi: i can have 80 logged in and sat in 20-25 minutes at most
[11:41]  Dahlia.Trimble lots of stuff, looks like assets or something
[11:41]  Mata Hari: it seems like it wants *all* appearance data inside that (15-second?) window or it will just sit there for 2 minutes and then fail
[11:42]  Justin Clark-Casey: I have a nasty suspicion I have to audit all the HG comms, just to get an idea of what the difficulties may be for the conference
[11:42]  Dahlia.Trimble its as if it wants all the assets before it teleports you
[11:42]  Nebadon Izumi: if i dont have to run the region scanner to find a chair radegast uses a lot less memory
[11:42]  Mata Hari: but the data still gets there somehow
[11:42]  Nebadon Izumi: i have 80 clients running
[11:42]  Nebadon Izumi: and still have about 5gb ram free
[11:42]  BlueWall.Slade we should put soem #ifdef in the code and set some counters for various url handlers
[11:42]  Nebadon Izumi: out of 16gb total
[11:42]  Justin Clark-Casey: possibly it's cached on first failure
[11:42]  Mata Hari: then on next attempt the data is all cached so it works instantly
[11:42]  Justin Clark-Casey: so doesn't need to be re-sent on the next attempt
[11:42]  Mata Hari: right
[11:42]  Justin Clark-Casey: yeah, that's what I would assume
[11:43]  Justin Clark-Casey: how do you know it's sending a lot of data. You're looking at a network monitor or similar?
[11:43]  Mata Hari: just reams and reams of messages in the console about getting textures, etc
[11:43]  Mata Hari: an inbound tp from a friends gerates 100-200 lines in the console about that stuff
[11:43]  Justin Clark-Casey: have you chased down what those textures are? Whether they related to attachments on the avatar
[11:43]  Mata Hari: (using default loggning)
[11:44]  Mata Hari: no
[11:44]  Justin Clark-Casey: hg stuff is very debug log heavy. The non-HG has been slowly worn away
[11:44]  Dahlia.Trimble there are a lot of console messages
[11:44]  Dahlia.Trimble wear an avatar with a lot of attachments and watch the console as you hg to a new place
[11:44]  Mata Hari: yeah
[11:44]  Justin Clark-Casey: heh, that's easy with neb's aliens
[11:44]  Nebadon Izumi: logging in general when you get heavy console spew can actually degrade overall server performance
[11:44]  BlueWall.Slade that is probably the bulk of the messages?
[11:45]  Nebadon Izumi: there is a point where logging becomes counter productive
[11:45]  BlueWall.Slade attachments.
[11:45]  Mata Hari: I guess the question is more whether it wouldn't be possible to tranfer the agent and let the attachments finish as fast as they can
[11:45]  Justin Clark-Casey: tbh, it's an avenue that I need to check for the conference so may get some attention soon, or at least I will write up/understand exactly what is happening
[11:45]  Mata Hari: vs the current method seems to be an all or nothing thing
[11:45]  BlueWall.Slade probably inventory folder requests too
[11:45]  BlueWall.Slade even though they are protected
[11:45]  Nebadon Izumi: ya i notice here today
[11:45]  Mata Hari: yes, tons of folder requests as well
[11:45]  Nebadon Izumi: when some people log in i get a ton of yellow warnings
[11:45]  Nebadon Izumi: about cant read folder
[11:46]  Justin Clark-Casey: our old inventory friend, eh?
[11:46]  Dahlia.Trimble theres a glitch in the protocol. It shouldnt impede the teleport waiting for asset transfers
[11:46]  BlueWall.Slade we should nix those
[11:46]  Nebadon Izumi: i always assumed that was HG invnetory hitting protected folders
[11:46]  Mata Hari: but all of that has to complete before it will let the tp happen
[11:46]  Mata Hari: and the time window is something like 15 seconds
[11:46]  Dahlia.Trimble maybe its a http queue issue
[11:46]  BlueWall.Slade for HG, we should only request the My Suitcase
[11:46]  Justin Clark-Casey: depends if the warnings are important or unavoidable problems
[11:46]  Nebadon Izumi: that wouldnt be so for OSgrid
[11:46]  Nebadon Izumi: would need to be careful how that is implemented
[11:46]  Billy.Bradshaw Bluewall++
[11:46]  Nebadon Izumi: not everyone uses suitcase
[11:46]  Mata Hari: I think most HG users would be content to get there, then let their stuff catch up to them
[11:46]  Justin Clark-Casey: even if unavoidable, I would prefer to see them recorded somewhere accessible via a console command, just to avoid the frustration of having bugs connected with unrecorded errors
[11:46]  BlueWall.Slade I think a lot of it is R&D releated
[11:47]  Dahlia.Trimble like if it creates a bunch of http requests and then tries to send a tp request via rest, the request might sit for a while
[11:47]  BlueWall.Slade maybe a separate log file?
[11:47]  Billy.Bradshaw Has to be suitcase, all other inventory folders are marked unavailable
[11:47]  Andrew.Hellershanks Mata, sounds reasonable. We often have to wait for our appearance to rez after logging in to a grid or after TPs.
[11:47]  Mata Hari: exactly
[11:47]  BlueWall.Slade it tries to get all folders
[11:47]  BlueWall.Slade the remote server refuses
[11:47]  Nebadon Izumi: you know
[11:47]  Mata Hari: I can live iwth being a cloud for a minute or two if I can actually get there reliably
[11:48]  Nebadon Izumi: i know what this might be
[11:48]  Nebadon Izumi: 1 sec
[11:48]  BlueWall.Slade and tha ttakes cycles and bandwidth
[11:48]  Nebadon Izumi: [InterestManagement]
 UpdatePrioritizationScheme = FrontBack
[11:48]  Nebadon Izumi: this
[11:48]  Nebadon Izumi: if you are running anything other than this
[11:49]  Nebadon Izumi: Avatars will jam up the simulator on login
[11:49]  Dahlia.Trimble well I'm amazed hg works as good as it does but it still has a few rough edges
[11:49]  Nebadon Izumi: giving avatars priority on login is bad news
[11:49]  Justin Clark-Casey: i'm amazed hg works at all
[11:49]  Nebadon Izumi: some grids like it because of RPG stuff
[11:49]  Nebadon Izumi: i suspect if you switch to FrontBack
[11:49]  Nebadon Izumi: you will see major improvements on logins
[11:49]  Nebadon Izumi: otherwise if its BestAvatarAppearance
[11:49]  Nebadon Izumi: avatasr pull priority
[11:50]  Nebadon Izumi: and it can be super nasty
[11:50]  Nebadon Izumi: its not as noticeable with single file loigins
[11:50]  Billy.Bradshaw need to switch computers, brb
[11:50]  Nebadon Izumi: but 2 or 3 simutlaneous logins
[11:50]  Nebadon Izumi: its not good
[11:51]  Dahlia.Trimble I didnt think the interest stuff worked
[11:51]  Nebadon Izumi: oh it does
[11:51]  Nebadon Izumi: the only one that works good though is FrontBack
[11:51]  Nebadon Izumi: in my opinion
[11:51]  Andrew.Hellershanks Is that only for HG grids?
[11:51]  Nebadon Izumi: no
[11:51]  Nebadon Izumi: its for everything
[11:51]  Nebadon Izumi: but using BestAvatarApperance
[11:51]  Nebadon Izumi: if you get someone who gets super long delays
[11:51]  Nebadon Izumi: that could jam up threads
[11:52]  Mata Hari: maybe FrontBack ought to be made the default then?
[11:52]  Nebadon Izumi: because it pushes all priority to loading avatar apperance over everything else
[11:52]  BlueWall.Slade "Hey! You like my threads?"
[11:52]  Mata Hari: I've just now changed to that....will be interesting to see if it makes any different to inbound tps
[11:52]  Nebadon Izumi: ya definitely
[11:53]  Nebadon Izumi: let us know for sure
[11:53]  Mata Hari: (I had it set at the default)
[11:53]  Andrew.Hellershanks I'll have to try that FrontBack setting and see what effect it has.
[11:53]  Nebadon Izumi: I beleive the default is still BestAvatarAppearance
[11:53]  Nebadon Izumi: let me check
[11:53]  Mata Hari: yes, it is
[11:53]  Mata Hari: I'm looking at it right now...that's the default
[11:53]  Nebadon Izumi: it is
[11:54]  Nebadon Izumi: i suspect you will see some improvement there
[11:54]  Nebadon Izumi: it probably wont fix all
[11:54]  Nebadon Izumi: but you should see gains
[11:54]  Dahlia.Trimble I have a scene octree in a private module. I wouldnt mind donating it but it probably should be integrated directly into scene rather than being a region module and I'm not sure of what anyone's plans are for scene
[11:54]  Nebadon Izumi: that is how all the OSGrid plazas and this entire grid is configured
[11:54]  Dahlia.Trimble it would be awesome for interest management
[11:55]  Nebadon Izumi: ya to be honest
[11:55]  Nebadon Izumi: the interest management stuff never worked well
[11:55]  Nebadon Izumi: i saw major problems after it was instituted
[11:55]  Nebadon Izumi: performance tanked bad
[11:55]  Nebadon Izumi: it was never the same after that
[11:56]  Dahlia.Trimble ya you need to use an octree or its horribly slow
[11:56]  Nebadon Izumi: other things have improved and its gotten considerable better
[11:56]  Andrew.Hellershanks still hears some thunder rumbles
[11:56]  Nebadon Izumi: but prior to it existing things ran much better
[11:56]  Nebadon Izumi: though it probably didnt scale well
[11:56]  Nebadon Izumi: thats when all the avatar apparance problems started
[11:56]  Nebadon Izumi: lots of clouds etc..
[11:57]  BlueWall.Slade also LL started baking all the textures on the avatar and sending them vrom the viewr
[11:57]  BlueWall.Slade so, that makes a lot of sense
[11:57]  Dahlia.Trimble well clouds could be any one of hundreds of potential failure points
[11:58]  Nebadon Izumi: ya I just remember arguing about it a bunch with jhurliman and jradford
[11:58]  Nebadon Izumi: who did most of that work
[11:58]  BlueWall.Slade lol
[11:58]  Nebadon Izumi: then melanie made some adjustments later
[11:58]  Justin Clark-Casey: man, how many years ago was this?
[11:58]  Nebadon Izumi: ages
[11:59]  Dahlia.Trimble wonders where the hurlicane is blowing now....
[11:59]  Nebadon Izumi: heh
[11:59]  Nebadon Izumi: he does move around a lot
[11:59]  Justin Clark-Casey: he was clever :)
[11:59]  Dahlia.Trimble ya
[11:59]  Justin Clark-Casey: and knew when to up sticks
[11:59]  BlueWall.Slade he was, at that
[11:59]  Dahlia.Trimble that music site was cool
[12:00]  Dahlia.Trimble cant remember
[12:00]  Nebadon Izumi: i occaisionally get updates from linkedin saying he changed jobs
[12:00]  BlueWall.Slade that was it
[12:00]  Nebadon Izumi: ya i still get requests in SL for people wanting that TV i made for them
[12:00]  Nebadon Izumi: lol
[12:01]  Billy.Bradshaw Does Opensim only support V3 viewerss now?
[12:01]  Nebadon Izumi: no
[12:01]  Justin Clark-Casey: it's 12 btw
[12:01]  BlueWall.Slade I gotta run - I'll jump home then back to keynote 1 to park while I'm AFK
[12:01]  Billy.Bradshaw are V1 and V2 gone?
Personal tools
About This Wiki