What I believe is happening is that on initial terrain send, this is done one packet at a time.
With WaitOne, the outbound loop has enough time to loop and wait again after the first packet before the second, leading to a slower send.
This approach instead does not wait if a packet was just sent but instead loops again, which appears to lead to a quicker send without losing the cpu benefit of not continually looping when there is no outbound data.
cap is something other than "localhost". A new interface for handling
external caps is supported with an example implemented for Simian. The
only linden cap supporting this interface right now is the GetTexture
cap.
Only HG Visitors get this var, to avoid spamming local users.
The config var is now called MapTileURL, to be consistent with the login one, and its being picked up from either [LoginService], [HGWorldMap] or [SimulatorFeatures], just because I have a bad memory.
This reflects the actual use of this stat - it hasn't recorded general exceptions for some time.
Make the sim extra stats collector draw the data from the stats manager rather than maintaing this data itself.
The major departure from flotsam is to send only one message per destination region, as opposed to one message per group member. This reduces messaging considerably in large groups that have clusters of members in certain regions.
- Child and root agents are only closed after 15 sec, maybe
- If the user comes back, they aren't closed, and everything is reused
- On the receiving side, clients and scene presences are reused if they already exist
- Caps are always recreated (this is where I spent most of my time!). It turns out that, because the agents carry the seeds around, the seed gets the same URL, except for the root agent coming back to a far away region, which gets a new seed (because we don't know what was its seed in the departing region, and we can't send it back to the client when the agent returns there).
In this new protocol, and as committed before, the viewer is not sent EnableSimulator/EstablishChildCommunication for the destination. Instead, it is sent TeleportFinish directly. TeleportFinish, in turn, makes the viewer send a UserCircuitCode packet followed by CompleteMovementIntoRegion packet. These 2 packets tend to occur one after the other almost immediately to the point that when CMIR arrives the client is not even connected yet and that packet is ignored (there might have been some race conditions here before); then the viewer sends CMIR again within 5-8 secs. But the delay between them may be higher in busier regions, which may lead to race conditions.
This commit improves the process so there are are no race conditions at the destination. CompleteMovement (triggered by the viewer) waits until Update has been sent from the origin. Update, in turn, waits until there is a *root* scene presence -- so making sure CompleteMovement has run MakeRoot. In other words, there are two threadlets at the destination, one from the viewer and one from the origin region, waiting for each other to do the right thing. That makes it safe to close the agent at the origin upon return of the Update call without having to wait for callback, because we are absolutely sure that the viewer knows it is in th new region.
Note also that in the V1 protocol, the destination was getting UseCircuitCode from the viewer twice -- once on EstablishAgentCommunication and then again on TeleportFinish. The second UCC was being ignored, but it shows how we were not following the expected steps...
If this consistently increases then this is a problem since it means the simulator is receiving more requests than it can distribute to other parts of the code.
If we're not receiving packets with multiple threads (m_asyncPacketHandling) then this is critical since it will limit the number of incoming UDP requests that the region can handle and affects packet loss.
If m_asyncPacketHandling then this is less critical though a long process will increase the scope for threads to race.
This is an experimental stat which may be changed.
AgentUpdate packet. This fixes the problem with vehicles not moving forward
after the first up-arrow.
Code to fix a potential exception when using different IClientAPIs.
- The existing event to scene has been split into 2: OnAgentUpdate and OnAgentCameraUpdate, to better reflect the two types of updates that the viewer sends. We can run one without the other, which is what happens when the avie is still but the user is camming around
- Added thresholds (as opposed to equality) to determine whether the update is significant or not. I thin these thresholds are ok, but we can play with them later
- Ignore updates of HeadRotation, which were problematic and aren't being used up stream
Added here since it was the most convenient place
Number is in the last column, "Sig. AgentUpdates" along with percentage of all AgentUpdates
Percentage largely falls over time, most cpu for processing AgentUpdates may be in UDP processing as turning this off even earlier (with "debug lludp toggle agentupdate" results in a big cpu fall
Also tidies up display.
Enabling this will stop anybody from moving on a sim, though all other updates should be unaffected.
Appears to make some cpu difference on very basic testing with a static standing avatar (though not all that much).
Need to see the results with much higher av numbers.
This appears to improve cpu usage since launching a new thread is more expensive than performing a small amount of inline logic.
However, needs testing at scale.
successfully tested, and I'm merging back those changes, which proved to
be good.
Revert "Revert "Cleared up much confusion in PollServiceRequestManager. Here's the history:""
This reverts commit fa2370b32e.
When Melanie added the web fetch inventory throttle to core, she made the long poll requests (EQs) effectively be handled on an active loop. All those requests, if they existed, were being constantly dequeued, checked for events (which most often they didn't have), and requeued again. This was an active loop thread on a 100ms cycle!
This fixes the issue. Now the inventory requests, if they aren't ready to be served, are placed directly back in the queue, but the long poll requests aren't placed there until there are events ready to be sent or timeout has been reached.
This puts the LongPollServiceWatcherThread back to 1sec cycle, as it was before.
This adds explicit cap poll handler supporting to the Caps classes rather than relying on callers to do the complicated coding.
Other refactoring was required to get logic into the right places to support this.
Justin, if you read this, there's a long story here. Some time ago you placed SendInitialDataToMe at the very beginning of client creation (in LLUDPServer). That is problematic, as we discovered relatively recently: on TPs, as soon as the client starts getting data from child agents, it starts requesting resources back *from the simulator where its root agent is*. We found this to be the problem behind meshes missing on HG TPs (because the viewer was requesting the meshes of the receiving sim from the departing grid). But this affects much more than meshes and HG TPs. It may also explain cloud avatars after a local TP: baked textures are only stored in the simulator, so if a child agent receives a UUID of a baked texture in the destination sim and requests that texture from the departing sim where the root agent is, it will fail to get that texture.
Bottom line: we need to delay sending the new simulator data to the viewer until we are absolutely sure that the viewer knows that its main agent is in a new sim. Hence, moving it to CompleteMovement.
Now I am trying to tune the initial rez delay that we all experience in the CC. I think that when I fixed the issue described above, I may have moved SendInitialDataToMe to much later than it should be, so now I'm moving to earlier in CompleteMovement.
reception thread to sleep for 30ms if the number of available user worker
threads got low. It doesn't look like any of the UDP packet types are
marked async so this check is 1) unnecessary and 2) really crazy since
it stops up the reception thread under heavy load without any indication.
This currently prints caps requests received and handled, so that overload of received compared to handled or deadlock can be detected.
This involves making BaseStreamHandler and BaseOutputStream record the ints, which means inheritors should subclass ProcessRequest() instead of Handle()
However, existing inheriting classes overriding Handle() will still work, albeit without stats recording.
"show caps" becomes "show caps list" to disambiguate between show caps commands
also moves the implementing code into LLUDPServer.cs along with other debug commands from OpenSim.cs
gets all debug lludp commands to only activate for the set scene if not root
hashes for the purpose of accurately responding to AgentTextureCached
packets. There is a change to IClientAPI to report the wearbles hashes
that come in through the SetAppearance packet. Added storage of the
texture hashes in the appearance. While these are added to the
Pack/Unpack (with support for missing values) routines (which means
Simian will store them properly), they are not currently persisted in
Robust.
packet can be pulled out of LLClientView and moved to
AvatarFactory. The first pass at reusing textures (turned off by
default) is included. When reusing textures, if the baked textures
from a previous login are still in the asset service (which generally
means that they are in the simulator's cache) then the avatar will not
need to rebake. This is both a performance improvement (specifically
that an avatars baked textures do not need to be sent to other users
who have the old textures cached) and a resource improvement (don't
have to deal with duplicate bakes in the asset service cache).
The previous wrong behavior caused the debug setting "UseHTTPInventory" to fail
on all viewers when turned off. UDB inventory would not be correctly used in
that case.
This was an undocumented interface which I think was for long defunct region load balancing experiments.
Also adds method doc for some IClientNetworkServer methods.
This tells the viewer to enable the UI for export permissions.
WARNING: If your inventory store contains invalid flags data, this will result
in items becoming exportable! Don't turn this on in production until it's complete!
with our own and add export permissions as well as a new definition for "All" as meaning "all conventional permissions" rather than "all possible permissions"
* Add zero length blocks to the new packet blocks to remain compatible with older viewers and avoid a NullRef when _packets_.cs calls the Length parameter.. which adds up the Length property all of the blocks.
Extend implementors of IStatsCollector to return an OSDMap of stats.
Update UserStatsCollector and AssetStatsCollector to return both
string and OSDMap data (as well as console format).
This is mostly Bluewall's work but I am also bumping the general version number
OpenSimulator 0.7.5 remains in the release candidate stage.
I'm doing this because master is significantly adding things that will not be in 0.7.5
This update should not cause issues with existing external binary DLLs because our DLLs do not have strong names
and so the exact version match requirement is not in force.
On Windows, concurrent multi-threaded processing of inbound UDP somehow allows different data input processing to interfere with each other.
Possibly the endpoint reference is being switched, though I don't yet know the mechanism. Not seen on Mono.
Also resolveable by setting RecyclePackets = false or RecycleBaseUDPPackets = false in [PacketPool]
Or async_packet_handling = false in [ClientStack.LindenUDP]
For now, will simply disable this particular pooling though will revisit this issue.
In response to http://opensimulator.org/mantis/view.php?id=6468
This prevents a slow grid information network call from holding up the main packet handling thread.
There's no obvious race condition reason for not doing this asynchronously.
This is to avoid the entire scene loop being held up when the group service is slow to respond.
There's no obvious reason for these queries to be sync rather than async.
Viewer 3 will discard such a message if the chat message owner does not match the avatar.
We were filling the ownerID with the primID, so this never matched, hence viewer 3 did not see any script error messages.
This commit fills the ownerID in with the prim ownerID so the script owner will receive script error messages.
This does not affect viewer 1 and associated viewers which continue to process script errors as normal.
This introduces a pull stat type in addition to the push stat type.
A pull stat takes a method on construction which knows how to update the stat on request.
In this way, special interfaces for pull stat collection are not necessary.
Even when an avatar is standing still, it's sending in a constant stream of AgentUpdate packets that the client creates new UDPPacketBuffer objects to handle.
This option pools those objects. This reduces memory churn.
Currently off by default. Works but the scope can be expanded.
This is controlled via the "debug lludp start <in|out|all>" and "debug lludp stop <in|out|all>" region console commands.
The command "debug lludp status" will show current status.
I think this is more useful right now since it tells us if the viewer requested a seed caps at all in various scenarios (such as when teleporting to a new region).
These were neither being returned or in many places reused.
Getting packets from a pool rather than deallocating and reallocating reduces memory churn which in turn reduces garbage collection time and frequency.
This allows different categories of stats to be shown, with options to list categories or show all stats.
Currently categories are scene and simulator and only a very few stats are currently registered via this mechanism.
This commit also adds percentage stats for packets and blocks reused from the packet pool.
The viewer would not see the folder move without this, either on accept or decline.
This commit also updates the TaskInventoryOffered message to better conform with the data LL uses
Changes are, agentID is prim owner rather than prim id, agent name is now simply object name rather than name with owner detail,
message is just folder name in single quotes, message is not timestamped.
However, folder is not renamed "still #RLV/~<name>". Long term solution is probably not to do these operations server-side.
Notes will be added to http://opensimulator.org/mantis/view.php?id=6311
This is to resolve previous build break.
This unnecessarily but harmlessly reads and sets the parameter multiple times - scene was doing the same thing.
This is to allow a second attempt to remove an avatar even if "show connections" shows them as already inactive (i.e. close has already been attempted once).
You should only attempt --force if a normal kick fails.
This is partly for diagnostics as we have seen some connections occasionally remain on lbsa plaza even if they are registered as inactive.
This is not a permanent solution and may not work anyway - the ultimate solution is to stop this problem from happening in the first place.
Giving a large folder from one avatar to another was causing a long delay when handled synchronously, since it took some time to retrieve the necessary data from the inventory service.
Handling this asynchronously instead stops this delay from disrupting all avatars in the scene. This has been shown in OSGrid.
I see no reason for not handling all IM messages asynchronously, just as incoming chat is handled asynchronously, so this has been switched for all instant messages.
Thanks to Nebadon for testing this change out.
This may well come back in the future when this subinterface is actually used but it currently isn't and I feel the name was poor.
Everything uses IClientAPI.RemoveEndPoint which also returned the full endpoint rather than just the ip address.
Serialization of attachments requires IsAttachment = false so that correct positions are serialized instead of avatar position.
However, doing this when a hud is still attached allows race conditions with update threads, resulting in hud artifacts on other viewers.
This change sets SOG.IsDeleted before serialization changes take place (IsDeleted itself is not a serialized property).
LLClientView then screens out any deleted SOGs before sending updates to viewers.
If this happens, then the non-owners would see unremovable huds that they did not own until relog, and sometimes even beyond that.
This was due to a race between the entity update and the attachment code when moving an object from within scene to a hud.
IsActive is more appropriate since unack timeout is not due to voluntary logout.
This is in line with operations such as manual kick that do not set the IsLoggingOut flag.
It's also slightly better race-wise since it reduces the chance of this operation clashing with another reason for client deactivation (e.g. manual kick).
teravus