Following feedback from 0003440, i've made some changes to the new texture pipeline to optimise
performance. The changes are:
- Fixed a math issue where a small percentage of images with a certain size (on the packet boundary) would not have their final data delivered. This issue has been present since pre- 0003440
- It was suggested that a discardlevel of -1 and a prioriy of 0 meant to abandon the transfer, this is incorrect and caused some textures to clog.
- The texture throttle blocking queue is now only filled in relation to the actual throttle amount.. i.e, on a connection throttled to 300k, only twenty packets will be placed in the queue at a time, on a larger connection it will be much more. This is to balance responsiveness to requests and speed, and to minimise wasted packets.
- The engine now keeps track of the number of pending textures, and the stack will not be walked if there's no textures pending, saving CPU. Textures are only considered "pending" when they've already been decoded.
- As part of the above, some textures may receive twice as much data per cycle if the number of pending textures is below the cycle threshold, this should prevent loading from slowing down when there are fewer textures in the queue.
* WebStatsModule doesn't crash on restart. GodsModule doesn't crash when there is no Dialog Module. LLUDPServer doesn't crash when the Operation was Aborted.
* ODEPlugin does 'Almost NaN' sanity checks.
* ODEPlugin sacrifices NaN avatars to the NaN black hole to appease it and keep it from sucking the rest of the world in.
* Revamps the server side texture pipeline
* Textures should load faster, get clogged less, and be less blurry
* Minor tweak to ensure the outgoing texture throttle stays private.
* Fixes mantis 3440
Addresses Mantis #3381
The current implementation works as expected if the object has no rotation or
only rotation around the Z axis; you can spin the object left or right (around
the world Z axis).
It works a little unexpectedly if the object has a non-Z-axis rotation; in this
case the body is spun about its local Z axis, not the world Z-axis. (But SL
also behaves oddly with a spin on an arbitrarily rotated object.)
* Now, the 144 unit tests takes roughly as long time to run (16s on my laptop) that the 10 long running takes. The database tests takes forever.
* Feel free to run the unit tests as you code, and the rest before commit.
* Notes: this requires heavy testing, it may cause new issues where LL have recycled agent block data for non-security purposes. It can be disabled on Line 4421 of LLClientView.cs by changing m_checkPackets to false.
* This represents approx 1/8th of the packets being checked.
* Important: HttpServer.dll was changed to HttpServer_OpenSim.dll so that the HttpServer references do not conflict if you've copied the OpenMetaverse.Http.dll and requirements to the OpenSimulator bin folder.
This means that if you reference HttpServer.dll in any projects, you will need to change the reference to HttpServer_OpenSim.dll. It still uses the Same HttpServer namespace though.
Also truncate messages that may exceed the limit set by the packet size. The limit in OpenMetaverse is 1100 bytes including a zero byte terminator.
Fixes Mantis #3244
- handle GetUser request for nonexistent user gracefully
- include throttle levels in ClientInfo
- code to save/restore throttles in client stack
- only update/send updates to active clients
- make animation classes serializable
* Added log4net dependency to physxplugin in prebuild.xml.
* Added missing m_log fields to classes.
* Replaced Console.WriteLine with appropriate m_log.Xxxx
* Tested that nant test target runs succesfully.
* Tested that local opensim sandbox starts up without errors.
* Added estate ban table to migration scripts of all supported databases.
* Added nhibernate mapping for EstateBans property of EstateSettings
* Refactored property accessors for EstateBan object.
* Added comments for EstateBan properties.
* Ensured that NHibernate tests pass with NUnitGUI.
* Ensured that nant test target passes.
This fixes mantis #3210. Thank you, tlaukkan!
and dropped nonths ago, because it is required to get smooth region
crossings with AO running. Without it, in some corner cases, anims will
continue to run in an unstoppable state.
NOTHING has been deleted or moved off to forge at this point. what
has happened is that OpenSim.Region.Environment.Modules has been split
in two:
- OpenSim.Region.CoreModules: all those modules that are either
directly or indirectly referenced from other OpenSim packages, or
that provide functionality that the OpenSim developer community
considers core functionality:
CoreModules/Agent/AssetTransaction
CoreModules/Agent/Capabilities
CoreModules/Agent/TextureDownload
CoreModules/Agent/TextureSender
CoreModules/Agent/TextureSender/Tests
CoreModules/Agent/Xfer
CoreModules/Avatar/AvatarFactory
CoreModules/Avatar/Chat/ChatModule
CoreModules/Avatar/Combat
CoreModules/Avatar/Currency/SampleMoney
CoreModules/Avatar/Dialog
CoreModules/Avatar/Friends
CoreModules/Avatar/Gestures
CoreModules/Avatar/Groups
CoreModules/Avatar/InstantMessage
CoreModules/Avatar/Inventory
CoreModules/Avatar/Inventory/Archiver
CoreModules/Avatar/Inventory/Transfer
CoreModules/Avatar/Lure
CoreModules/Avatar/ObjectCaps
CoreModules/Avatar/Profiles
CoreModules/Communications/Local
CoreModules/Communications/REST
CoreModules/Framework/EventQueue
CoreModules/Framework/InterfaceCommander
CoreModules/Hypergrid
CoreModules/InterGrid
CoreModules/Scripting/DynamicTexture
CoreModules/Scripting/EMailModules
CoreModules/Scripting/HttpRequest
CoreModules/Scripting/LoadImageURL
CoreModules/Scripting/VectorRender
CoreModules/Scripting/WorldComm
CoreModules/Scripting/XMLRPC
CoreModules/World/Archiver
CoreModules/World/Archiver/Tests
CoreModules/World/Estate
CoreModules/World/Land
CoreModules/World/Permissions
CoreModules/World/Serialiser
CoreModules/World/Sound
CoreModules/World/Sun
CoreModules/World/Terrain
CoreModules/World/Terrain/DefaultEffects
CoreModules/World/Terrain/DefaultEffects/bin
CoreModules/World/Terrain/DefaultEffects/bin/Debug
CoreModules/World/Terrain/Effects
CoreModules/World/Terrain/FileLoaders
CoreModules/World/Terrain/FloodBrushes
CoreModules/World/Terrain/PaintBrushes
CoreModules/World/Terrain/Tests
CoreModules/World/Vegetation
CoreModules/World/Wind
CoreModules/World/WorldMap
- OpenSim.Region.OptionalModules: all those modules that are not core
modules:
OptionalModules/Avatar/Chat/IRC-stuff
OptionalModules/Avatar/Concierge
OptionalModules/Avatar/Voice/AsterixVoice
OptionalModules/Avatar/Voice/SIPVoice
OptionalModules/ContentManagementSystem
OptionalModules/Grid/Interregion
OptionalModules/Python
OptionalModules/SvnSerialiser
OptionalModules/World/NPC
OptionalModules/World/TreePopulator
These changes replace all direct references to the AssetCache with
IAssetCache. There is no change to functionality. Everything works as
before.
This is laying the groundwork for making it possible to register
alternative asset caching mechanisms without disrupting other parts of
OpenSim or their dependencies upon AssetCache functionality.
OpenSim.Region.Environment into a "framework" part and a modules only
part. This first changeset refactors OpenSim.Region.Environment.Scenes,
OpenSim.Region.Environment.Interfaces, and OpenSim.Region.Interfaces
into OpenSim.Region.Framework.{Interfaces,Scenes} leaving only region
modules in OpenSim.Region.Environment.
The next step will be to move region modules up from
OpenSim.Region.Environment.Modules to OpenSim.Region.CoreModules and
then sort out which modules are really core modules and which should
move out to forge.
I've been very careful to NOT BREAK anything. i hope i've
succeeded. as this is the work of a whole week i hope i managed to
keep track with the applied patches of the last week --- could any of
you that did check in stuff have a look at whether it survived? thx!
* Increases the number of ImageDataPackets we send per PriorityQueue pop and tweak it so that the number of packets is ( (2 * decode level) + 1 ) * 2, and (((2 * (5-decode level)) + 1) * 2). The first one sends more data for low quality textures, the second one sends more data for high quality textures.
sceneB
* However, I'm not convinced that the actual process in the test completely reflects reality, and a lot of stuff had to be rigged up (which should get resolved over time)
* Adds a prototype web stats module which is disabled by default. It's functional with one report right now, however, the database structure may change, so I don't recommend enabling this to keep actual stats right now. I'll let you know when it's safe.
* Adds Prototype for ajaxy web content
* removed a warning or two.
* This facility allows you to save changes to an object that you've rezzed into a region back into their original inventory item without having to take a copy of the rezzed
object.
* However, this is not currently activated since it's not absolutely fully tested and there's a bug lurking in there to do with the sending of the BulkInventoryUpdate packets
Adds a new method to IClientAPI to allow adding message handlers
for GenericMessages (of which "autopilot" is one). Part 2 adds a
specific autopilot handler in ScenePresence.cs.
2) Removing unused variables and functions.
3) Simplifying the navigation logic in ScenePresence.cs. The
original patch was somewhat complex because it included orientation
logic for a future enhancement of orienting the avatar to point
towards the direction being walked. Currently this isn't working,
though, so I removed the orientation code, which leaves just the
smaller and hopefully simpler-to-understand movement code.
* SceneObjectPartInventory.cs isn't a particularly good name but it's probably not got a long life
* A proper inventory interface to follow
* Parallel changes for other inventory partial classes to follow at a later date
release a direct descendant of the stable 7364, with all the features and
none of the issues.
This omits the following patch chain: r7383 r7382 r7381 r7377 r7375 r7373
r7372 r7370 r7369 r7368 r7367 r7366
the IM module and makes it into a module of it's own, which can be used by
all other modules. Removes some ugly hacks. Refer to the IM module to see
how it's used. Also fixes the persistence issue (Mantis #2598)
Contains a migration. May contain nuts.
Please back up your inventory data store. This revision changes the interface
version!! No older regions can connect to these new UGAIM, and the new regions
can't connect to the old UGAIM. Fixes a long-standing issue of permissions loss
Currently persisted on MySQL only.
* This may alleviate a little the freezing experienced by existing avatars when a new client logs in
* Race condition risks look minimal since one wouldn't expect another thread to start fiddling with that presence
* Code Cleanliness Fixes in LLClientView
* Using field instead of local variable for handlerUpdatePrimGroupRotation (if you notice any new oddities with prim group rotation after this patch, please mantis)
* Implements the SendInitiateDownload method in IClientAPI
* Uses the ITerrainModule Interface to write a terrain file to disk then uses a FileStream to read the binary file from the disk and put it in a byte array. and save to the xFer list.
* It then tells the client to download the file and the client initiates an Xfer request.
* Items will now be locally cached for only 24 hours from last access. (Rather than until restart)
* Caveat: Implementing the new caching mechanism means statistics gathering on AssetCache is no longer functional. (Justin - you might want to take a look and see if you can somehow get that back and running if you still need it)
* This means the recent memory fix should now be working correctly - so the current largest memory leak should be fixed. AssetCache still needs to be addressed however.
* This patch is highly experimental and may cause clients to not be able to connect, if this is the case, it will be rolled back in approximately 5 minutes.
* Introducing IClientCore - this will be the key replacement for IClientAPI in the long run, it has a very minimal set of methods designed to allow you to access specialist API's.
* See https://lists.berlios.de/pipermail/opensim-dev/2008-September/003049.html for the early discussion on this.
* This should (hopefully) allow TestClient and stuff built on top of it to work again
* Will probably come back later and change variable names to stop this happening again
* Really this should be 1, but I think that this would be too slow compared to a Second Life server until we improve our ability to send textures of variable quality
* This may improve one aspect of sim performance where there are many avatars. However, there are still other performance problems that are unrelated to this change
* Value may be further tuned
* Removed temporary decals since the multipler setting will stick around now
* This should probably be 1, but currently by default it is 8, to reflect what was being eon3 in OpenSim before this revision. So if the client requested a maximum throttle
of 1500 kilobits per second, we would actually send out 1500 kilobytes per second
* Adjusting this multiplier down towards 1 may improve your OpenSim experience, though in other situations it may degrade (e.g. if you're using a standalone over high bandwidth
links)
* This is currently a user setting because adjusting it down may currently reveal other OpenSim bugs.
on-/offline updates, calling cards for friends.
This adds methods in the DB layer and changes the MessagingServer, so a full
update (incl. UGAIM) is necessary to get it working. Older regions shouldn't
break, nor should older UGAIM break newer regions, but friends/presence will
only work with all concerned parts (UGAIM, source region and destination
region) at this revision (or later).
I added the DB code for MSSQL, too, but couldn't test that.
BEWARE: May contain bugs.
* This may help us detect if mysterious UDP disconnects are happening because of this.
* Shouldn't be any functional change but I would appreciate a buddy check from Teravus if he has time (as for all client stack changes)
* Whilst this does automatically get the client to rebake, on crossing a region border the 'local' assets are left behind
* There may be a cunning solution (such as squirting the assets on region crossing, or having them fetched from the original region) but
instead I'm going to opt for the easy solution of keeping them in the asset database, for now
* And hopefully rebaking all the time should no longer be necessary now
* It turns out that when the client baked the texture, the uploaded asset had the Temporary flag to true (Temporary is actually deprecated).
* It also had the StoreLocal flag set to true, which signifies that the asset should be stored locally. If it disappears we should reply to the asset request with
ImageNotInDatabasePacket
* However, last time this was enabled some clients started crashing. This may well no longer be the case and needs to be tested, but in the mean time we will store
the asset instead.
* This needs to be resolved in a better way, possibly by starting to send the ImageNotInDatabase packet again instead
* This moves authentication from the client thread (where failure was difficult to detect) to the particular thread handling that packet
* I've kept the authentication outside of the crucial clientCircuits lock (though any delay here is probably swamped by the other delays associated with login)
* Also added more to the unit test to ensure this doesn't regress
* This checks that a client circuit is established when the udp server is given a use client circuit code packet
* And checks that other circuit codes do not exist
* Not sure why things still worked in the presence of this bug - possibly the problem is compensated for later on. If you are having udp session problems this bug fix may help
(though no guarantees).
* Guys, there's an endless loop there *ON PURPOSE*. Please don't try to *fix* it. We must continue to process the UDP stream buffer on clients that disconnected nastily until it ends or the UDP server accept thread will die a horrible death.
Add rezzing time to objects. Add Object return and traffic fields to land
database. Add plumbing for auto return. Implement auto return.
Contains a migration. May contain nuts.
* Regarding an earlier change, I think it would be possible to eliminate the creation of new IPEndPoints on every end receive if we did the client circuit lookup before starting
the next receive. However, this would be a performance trade off and hence not worth trying without performance testing
* This widened what I think is an existing race condition where asynchronous recieves could potentially stomp on each other's end points (though this must occur very rarely, if at
all, in reality)
* This allows multiple user profile providers to be specified in OpenSim.ini separated by commas
* If multiple providers are specified then a request for a user profile will query each in turn until the profile is either found or all have been queried
* Unfortunately I don't believe this order can currently be specified, which if true is something that will need to be fixed.
* Thanks to smeans for the original patch.
* I believe this is reasonable since code outside the Linden client stack shouldn't be aware of the packet format being used
* I would love to have made the method protected, but the LoadBalancerPlugin is still calling it and resolving that would require more work
* This should fix a long standing issue where you often wouldn't see other people simply turn around without moving at all
* Arguably lastPhysRot (to mirror lastPhysPos) is not a good name, may change variable names later
* This may help http://opensimulator.org/mantis/view.php?id=2377 where large linksets do not always correctly delete - since a lost kill packet to the client could result in
the symptoms described
* Experimenting with the PacketPool mechanism.
* It's still disabled in the code, however there's now a flag to enable it.
* Converted to use Generic Collections vs Hashtables, also now uses a list of 'OK to pool' packets, starting with the high volume PacketAck packet.
The average-value of modify.ModifyBlock.Height in LLClientView.cs:4170
seem to be incorrect or it isn't the average? Mhhh...
So the terrain build -> Flaten Sphere is unuseable.
I have put in a patch that contains a workaround while
the main problem is not solved.
* I believe this was the cause of the remaining packet_out_of_order messages in the Linden client logs
* There were race conditions where multiple clientstacks would overwrite each other's sequence numbers
* Changing network bandwidth in the preferences will now have a much more noticeable effect - a user may want to increase this if data is being slow to download from opensim
Implementation of llModifyLand() and There is a bug on
permission-check of land-terraforming: x an y-coordinates
are interchanged on function-call ExternalChecksCanTerraformLand.
Correct: x is west, and y is north. 2) Missing check of
"Other allow to terraform-flag" (Parcel.ParcelFlags.AllowTerraform)
Introduce a resend counter on the ack queue. The header "Resent" field is
now obsolete. Implement 3 resends on reliable packets, variable.
Increase default resend timeout to 3000ms and default silence threshold
to 350ms.
old way: each region module interested in chat from client had to
- subscribe to scene.EventManager.OnNewClient
- then in its OnNewClient delegate it would subscribe to
client.OnChatFromViewer to capture chat messages coming
new way: ChatModule is the only region module that uses the "old
way" approach but is now forwarding all client chat via
scene.EventManager.OnChatFromClient
- each region module interested in chat from client now only
subscribes to scene.EventManager.OnChatFromClient
this not only simplifies code, but also allows us to substitute
ChatModule with derived classes (ConciergeModule is going to be one
example).
Also, this changeset changes ChatFromViewer to ChatFromClient as it
doesn't necessarily have to be a viewer that is a chat source.
i've taken great care to only comment out those OnNewClient delegates
that were only used for getting at the client chat --- hope it's not
breaking anything.
* Switched it on by default
* Updated OpenSim.ini.example to reflect this
* Caught a UDP Server issue that occurs when the network pipe is saturated
* Still experimental :D
* This actually probably doesn't make a huge difference. We still end up resending a massive number of packets in quite a few situations, but it's not easy to consistently
reproduce this.
* Might really need some throttling of packet send to the client on the OpenSim end.
Estate/ Manager Owner Uses the Region/Estate Menu Region TAB,
and uses "Teleport Home All Users..." (Action Button), the
action will complete but no one will be teleported and all
users still function in the region ok.
* returns FAKEEVENT instead of the connection returning a 502. It doesn't like our 502's for some reason.. so, in leau of this.. send it a fake event.
* Once again, this is still 'really early' code, so please don't blame us if you have no more threads left.
* It seems kind of silly to be building a 256x256 array just to use two 16 float blocks.. but for now the layerdata routine requires it so we'll go along with that.
* We only fill a 32x16 area of the 256x256 float array with data.
* We use patches 0,0 and 0,1 for the first and second patch to determine the direction and magnitude of the wind.
Check the client dialog box (from top menu) WORLD / REGION ESTATE
/ REGION tab. The client dialog box seems to have a hard limit of
about 32 characters per line available for displaying the region
version number. Our regions are sending a string which is greater
than the limit, causing the client to wrap the text and look ugly.
* A little wind wouldn't hurt anyone, right? This is the 'slightly breezy' setting.. hopefully you won't notice 'much' of a difference.
* It turns out the terrain patch routine is similar enough to the wind version that it can be used to hack together a breeze generator with a few mods.
* Not much configuration.. yet. You only get breeze updates in the general vicinity of your camera now to keep bandwidth usage down.. and we're not talking about 'much' movement at the moment.
* initial version... could use improvement I'm sure.
and makes it use a common set of types in both engine. Fixes the issues with
running both engines and HTTP requests / listens / timers etc..
Also fixes a couple of minor Scene issues and a CTB by nullref.
for inventory REST calls for the time being, as firefox, curl, and
also python's urllib2 cannot authenticate using digest auth.
fix permission checking for prim inventory to be the same as for
normal edit ops.
* if a packet was really null it would be caught by the general exception handler at the top of the client thread, which would also provide more information and attempt a clean
shutdown
* There was a small window where region logins were allowed before modules were loaded - avatars logins that hit this window could have caused bad things to happen.
* A similar change will follow for grid mode sometime soon
Charles Krinke