You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@qpid.apache.org by "Marcel Meulemans (JIRA)" <ji...@apache.org> on 2018/05/29 20:32:00 UTC

[jira] [Updated] (DISPATCH-1019) Messaging instability in networks with many clients / addresses.

     [ https://issues.apache.org/jira/browse/DISPATCH-1019?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Marcel Meulemans updated DISPATCH-1019:
---------------------------------------
    Description: 
After DISPATCH-966 has been fixed I am still experiencing problems in a network with many clients / addresses. I am running a three node fully connected mesh of dispatch routers with 10000 attached clients all with two address messaging at around 100 msg/sec.

In the logs I am seeing the following errors:

{{2018-05-29 14:31:05.145732 +0000 ERROR (error) Invalid message: Insufficient Data to Determine Tag}}
 {{2018-05-29 14:31:05.145748 +0000 ERROR (error) Invalid message: Can't convert message field body}}

Which, in turn, lead to python errors like:

{{2018-05-29 14:31:05.145971 +0000 ROUTER_MA (trace) RCVD: MAU(id=None pv=1 area=0 mobile_seq=0)}}
{{ 2018-05-29 14:31:05.146130 +0000 ROUTER (error) Exception in control message processing}}
{{ Traceback (most recent call last):}}
{{File "/usr/lib/python2.7/qpid_dispatch_internal/router/engine.py", line 157, in handleControlMessage}}
{{self.mobile_address_engine.handle_mau(msg, now)}}
{{File "/usr/lib/python2.7/qpid_dispatch_internal/router/mobile.py", line 97, in handle_mau}}
{{node = self.node_tracker.router_node(msg.id)}}
{{File "/usr/lib/python2.7/qpid_dispatch_internal/router/node.py", line 363, in router_node}}
{{return self.nodes[node_id]}}
{{ KeyError: None}}
{{ 2018-05-29 14:31:05.146175 +0000 ROUTER (error) Control message error: opcode=MAU body=None}}

I have tracked down the cause of the "Insufficient Data to Determine Tag" message to the following: During the call to {{qd_parse}} of the {{MAU}} message the {{qd_iterator_t}} reaches the end of the buffer list before it should. Specifically the call to {{qd_iterator_advance}} here [https://github.com/apache/qpid-dispatch/blob/master/src/parse.c#L151,] "fails" to move forward a certain number of bytes (e.g. 31) even though the {{iterator->view_pointer->remaining}} value has plenty of bytes left (e.g. 84802). The "fail" is because we reach the end of the buffer list before we should (here [https://github.com/apache/qpid-dispatch/blob/master/src/iterator.c#L323|https://github.com/apache/qpid-dispatch/blob/master/src/iterator.c#L323).]).

What I have not been able to figure out yet is why this happens because it is not consistent. Many large MAU message are parsed correctly only sometimes not. I am able to reproduce these errors every time I run my tests. There may be a time component involved because the more logging I to the router code, the less often the errors seem to occur.

  was:
After DISPATCH-966 has been fixed I am still experiencing problems in a network with many clients / addresses. I am running a three node fully connected mesh of dispatch routers with 10000 attached clients all with two address messaging at around 100 msg/sec.

In the logs I am seeing the following errors:

{{2018-05-29 14:31:05.145732 +0000 ERROR (error) Invalid message: Insufficient Data to Determine Tag}}
{{2018-05-29 14:31:05.145748 +0000 ERROR (error) Invalid message: Can't convert message field body}}

Which, in turn, lead to python errors like:

{{2018-05-29 14:31:05.145971 +0000 ROUTER_MA (trace) RCVD: MAU(id=None pv=1 area=0 mobile_seq=0)}}
{{2018-05-29 14:31:05.146130 +0000 ROUTER (error) Exception in control message processing}}
{{Traceback (most recent call last):}}
{{ File "/usr/lib/python2.7/qpid_dispatch_internal/router/engine.py", line 157, in handleControlMessage}}
{{ self.mobile_address_engine.handle_mau(msg, now)}}
{{ File "/usr/lib/python2.7/qpid_dispatch_internal/router/mobile.py", line 97, in handle_mau}}
{{ node = self.node_tracker.router_node(msg.id)}}
{{ File "/usr/lib/python2.7/qpid_dispatch_internal/router/node.py", line 363, in router_node}}
{{ return self.nodes[node_id]}}
{{KeyError: None}}
{{2018-05-29 14:31:05.146175 +0000 ROUTER (error) Control message error: opcode=MAU body=None}}

I have tracked down the cause of the "Insufficient Data to Determine Tag" message to the following: During the call to {{qd_parse}} of the {{MAU}} message the {{qd_iterator_t}} reaches the end of the buffer list before it should. Specifically the call to {{qd_iterator_advance}} here [https://github.com/apache/qpid-dispatch/blob/master/src/parse.c#L151,] "fails" to move forward a certain number of bytes (e.g. 31) even though the {{iterator->view_pointer->remaining}} value has plenty of bytes left (e.g. 84802). The "fail" is because we reach the end of the buffer list before we should (here [https://github.com/apache/qpid-dispatch/blob/master/src/iterator.c#L323|https://github.com/apache/qpid-dispatch/blob/master/src/iterator.c#L323).]).

What I have not been able to figure out yet is why this happens because it is not consistent. Many large MAU message are parsed correctly only sometimes not. I am able to reproduce these errors every time I run my tests. There may be a time component involved because the more logging I to the router code, the less often the errors seem to occur.


> Messaging instability in networks with many clients / addresses.
> ----------------------------------------------------------------
>
>                 Key: DISPATCH-1019
>                 URL: https://issues.apache.org/jira/browse/DISPATCH-1019
>             Project: Qpid Dispatch
>          Issue Type: Bug
>          Components: Routing Engine
>    Affects Versions: 1.1.0
>            Reporter: Marcel Meulemans
>            Priority: Major
>
> After DISPATCH-966 has been fixed I am still experiencing problems in a network with many clients / addresses. I am running a three node fully connected mesh of dispatch routers with 10000 attached clients all with two address messaging at around 100 msg/sec.
> In the logs I am seeing the following errors:
> {{2018-05-29 14:31:05.145732 +0000 ERROR (error) Invalid message: Insufficient Data to Determine Tag}}
>  {{2018-05-29 14:31:05.145748 +0000 ERROR (error) Invalid message: Can't convert message field body}}
> Which, in turn, lead to python errors like:
> {{2018-05-29 14:31:05.145971 +0000 ROUTER_MA (trace) RCVD: MAU(id=None pv=1 area=0 mobile_seq=0)}}
> {{ 2018-05-29 14:31:05.146130 +0000 ROUTER (error) Exception in control message processing}}
> {{ Traceback (most recent call last):}}
> {{File "/usr/lib/python2.7/qpid_dispatch_internal/router/engine.py", line 157, in handleControlMessage}}
> {{self.mobile_address_engine.handle_mau(msg, now)}}
> {{File "/usr/lib/python2.7/qpid_dispatch_internal/router/mobile.py", line 97, in handle_mau}}
> {{node = self.node_tracker.router_node(msg.id)}}
> {{File "/usr/lib/python2.7/qpid_dispatch_internal/router/node.py", line 363, in router_node}}
> {{return self.nodes[node_id]}}
> {{ KeyError: None}}
> {{ 2018-05-29 14:31:05.146175 +0000 ROUTER (error) Control message error: opcode=MAU body=None}}
> I have tracked down the cause of the "Insufficient Data to Determine Tag" message to the following: During the call to {{qd_parse}} of the {{MAU}} message the {{qd_iterator_t}} reaches the end of the buffer list before it should. Specifically the call to {{qd_iterator_advance}} here [https://github.com/apache/qpid-dispatch/blob/master/src/parse.c#L151,] "fails" to move forward a certain number of bytes (e.g. 31) even though the {{iterator->view_pointer->remaining}} value has plenty of bytes left (e.g. 84802). The "fail" is because we reach the end of the buffer list before we should (here [https://github.com/apache/qpid-dispatch/blob/master/src/iterator.c#L323|https://github.com/apache/qpid-dispatch/blob/master/src/iterator.c#L323).]).
> What I have not been able to figure out yet is why this happens because it is not consistent. Many large MAU message are parsed correctly only sometimes not. I am able to reproduce these errors every time I run my tests. There may be a time component involved because the more logging I to the router code, the less often the errors seem to occur.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@qpid.apache.org
For additional commands, e-mail: dev-help@qpid.apache.org