[Opensim-users] Couple questions about master (Unhandled Packet AgentUpdate, and HTTP Inventory)

[Justin Clark-Casey]

How strange.  I inspected all the packet pool code again visually for possible race conditions or bugs and couldn't find 
anything.  I also tested with 2 viewers, LL 3.3.4 and LL 1.2.3 (on different machines) on both mono 2.10.9 and Windows 
with .net without seeing the issue.  This particularly aspect of packet pooling has been around since early October, afair.

However, clearly you're having some kind of problem.  I added a tiny bit more logging in git master 0b93a68 to tell us 
whether one or both viewers are having their AgentUpdate packets rejected and how these are interleaved.  Please could 
you retest with this and open a Mantis with the OpenSim.log.  Thanks.

On 13/12/12 04:57, Chris wrote:
> Hi Justin,
>
> This started to happen just recently as soon as I started to use commit 905d7c4 (r/21385); although I am not sure
> exactly when this may have started as I made a bit of a jump (895d28f r/20917 to 905d7c4 r/21385)
>
> This appears to happen when there is more than one person logged in.
>
> Did some tests in a few different variations of connectivity configurations (As far as which computer(s) the viewers are
> running on):
>
> Have 2 Computers set up, 1 with OS and all related services running, the other does not have anything OS related
> running. Running a typical home network, 1 network, nothing fancy =)
>
> Results:
> 1 viewer running on the non OS computer: No apparent issues
> 1 viewer running on the OS computer: No apparent issues
>
> 2 (or more) viewers running on the non OS computer: Shows the AgentUpdate packet warnings from time to time
> 2 (or more) viewers running on the OS computer: Shows the AgentUpdate packet warnings although it is noticeably a bit
> spammier in this configuration
>
> 1 viewer running on OS computer and 1 viewer running on non OS computer: Shows the AgentUpdate packet warnings although
> it is noticeably a bit spammier in this configuration. The viewer on the OS computer tends to experience network lag
> quite a bit when walking (i.e. The avatar drifts off into space for a few seconds after every few steps) while the
> viewer on the non OS computer will be able to walk around just fine but the warnings will still pop up in both cases.
>
>
> If I set RecyclePackets = false in [PacketPool] then the issues seem to go away. I repeated the above tests and threw in
> a couple more viewer instances for good measure to check as well.
>
>
>
> For the HTTP Inventory question; I set:
>
> [ClientStack.LindenCaps]
>      Cap_FetchInventoryDescendents2 = ""
>      Cap_FetchInventory2 = ""
>
> And inventory seems to load now (very quickly I might add) for viewers with HTTP Inventory disabled in Debug Settings.
>
>
> I did run into an exception along the way in my testing right after I made the changes in [ClientStack.LindenCaps]. The
> below stack trace came up (One exception per script in the item) when I detached one of my attachments. I am not certain
> if it is related to any of the above but I do recall that you wanted some more information on script state related
> issues so I thought this might be a good thing to add =)
>
> Thank you!
>
>
> Exception:
>
> 2012-12-12 22:35:38,843 ERROR - OpenSim.Region.ScriptEngine.XEngine.XEngine [XEngine]: Failed to save state of script
> AOHUD.AO v4, item UUID 14eba70e-
>
> 0bf8-4b5e-92d2-382f20eb4117, prim UUID 74cf7aca-c5f7-4dd7-8373-9dd8620ef8af in Test Region 2.  Exception
> System.AppDomainUnloadedException: The target application domain has been unloaded.
>
> Server stack trace:
>     at System.Threading.Thread.InternalCrossContextCallback(Context ctx, IntPtr ctxID, Int32 appDomainID,
> InternalCrossContextDelegate ftnToCall, Object
>
> [] args)
>     at System.Runtime.Remoting.Channels.CrossAppDomainSink.DoTransitionDispatch(Byte[] reqStmBuff,
> SmuggledMethodCallMessage smuggledMcm,
>
> SmuggledMethodReturnMessage& smuggledMrm)
>     at System.Runtime.Remoting.Channels.CrossAppDomainSink.SyncProcessMessage(IMessage reqMsg)
>
> Exception rethrown at [0]:
>     at System.Runtime.Remoting.Proxies.RealProxy.HandleReturnMessage(IMessage reqMsg, IMessage retMsg)
>     at System.Runtime.Remoting.Proxies.RealProxy.PrivateInvoke(MessageData& msgData, Int32 type)
>     at OpenSim.Region.ScriptEngine.Shared.ScriptBase.IScript.GetVars()
>     at OpenSim.Region.ScriptEngine.Shared.Instance.ScriptInstance.GetVars() in
> c:\Users\admin\Desktop\opensim-905d7c4\OpenSim\Region\ScriptEngine\Shared
>
> \Instance\ScriptInstance.cs:line 919
>     at OpenSim.Region.ScriptEngine.Shared.Instance.ScriptSerializer.Serialize(ScriptInstance instance) in
> c:\Users\admin\Desktop\opensim-905d7c4\OpenSim
>
> \Region\ScriptEngine\Shared\Instance\ScriptSerializer.cs:line 80
>     at OpenSim.Region.ScriptEngine.Shared.Instance.ScriptInstance.SaveState(String assembly) in
> c:\Users\admin\Desktop\opensim-905d7c4\OpenSim\Region
>
> \ScriptEngine\Shared\Instance\ScriptInstance.cs:line 961
>     at OpenSim.Region.ScriptEngine.XEngine.XEngine.DoBackup(Object o) in
> c:\Users\admin\Desktop\opensim-905d7c4\OpenSim\Region\ScriptEngine\XEngine
>
> \XEngine.cs:line 753
> 2012-12-12 22:35:38,890 ERROR - OpenSim.Region.ScriptEngine.XEngine.XEngine [XEngine]: Failed to save state of script
> AOHUD.Menu v1.0, item UUID
>
> c8c29c17-048e-4abf-b650-8122f75eda26, prim UUID 74cf7aca-c5f7-4dd7-8373-9dd8620ef8af in Test Region 2.  Exception
> System.AppDomainUnloadedException: The target application domain has been unloaded.
>
> On 12/12/2012 5:37 PM, Justin Clark-Casey wrote:
>> On 11/12/12 22:28, Chris wrote:
>>> Hello every one, got a couple of questions about git master 905d7c4 (r/21385).
>>>
>>> I recently downloaded and compiled master and am noticing a bunch of yellow text popping up (warnings?) saying [CLIENT]:
>>> unhandled packet AgentUpdate primarily when walking around but they do pop up on occasion when standing still. Is anyone
>>> else seeing this?
>>
>> AgentUpdates are extremely common messages.  If you do "debug packet 255" on the console you will see a constant
>> stream of them come in since they tell the server about agent movements, head rotation, etc, even when you're standing
>> completely still.
>>
>> So this must only be happening on rare occasions.  In the current code this should only happen if an agent update
>> message has somehow been sent to the client port from another source but with wrong session and agent IDs.
>>
>> This seems very unlikely, so I think a stronger candidate is some packet pooling issue.
>>
>> So in connection with this
>>
>> 1) How recently did this start happening?
>>
>> 2) Does it happen when you are alone on a region with nobody in the same or in any neighbouring regions?
>>
>> 3) What happens if you set
>>
>> [PacketPool]
>> RecyclePackets = false
>>
>> ?
>>
>>>
>>> Also a quick question about HTTP inventory and more recent viewers (i.e. Firestorm, SL v3.3.x, and possibly v1.23 based
>>> viewers that support it (have not tested these though)); It seems that when I have HTTP Inventory disabled my inventory
>>> refuses to load up until HTTP Inventory is enabled. I prefer to keep it off because UDP inventory loads a lot quicker
>>> for me and having it disabled also seems to result in more stable teleports on LL grid. Is there anyway to fix this in
>>> the config files?
>>
>> On the server side, UDP inventory is always available.  You could disable HTTP inventory on the server side with the
>> instructions at [1] just to be sure.  But I think a viewer bug is more likely where fallback is not happening properly
>> when HTTP inventory is disabled.  I don't know if there are any viewer config values which would help in this situation.
>>
>> [1] http://opensimulator.org/mantis/view.php?id=6412
>>
>
>


-- 
Justin Clark-Casey (justincc)
OSVW Consulting
http://justincc.org
http://twitter.com/justincc