[
https://issues.apache.org/jira/browse/QPID-8496?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17265908#comment-17265908
]
Robbie Gemmell edited comment on QPID-8496 at 1/15/21, 10:47 AM:
-----------------------------------------------------------------
Your confirmation that it didnt reproduce with that output tallies with what I
thought on first looking at it. That still suggests an empty 'more=false' frame
at the end of a message could still be the issue (which the broker should
handle IMO, but typically shouldn't happen), but it just didn't happen here.
The question would be how it gets in that situation. The trace snippets dont
appear to have enough information to reason about any particular issue, not
showing enough time period to reason about the overall interaction.
Some general observations:
- There looks to be 1 receiver, roughly trading aggregate volume of messages
with 3 senders overall, possibly favouring some senders more than others.
- The sent messages look around 3x+ larger than those received.
- The last large message transfer seen indicates more=true, i.e it isnt
complete. But nothing else is seemingly sent. Unclear why from the snippet.
That enabling the logging seemingly influenced things, might suggest it is a
timing issue. Could be protocol interaction, or threading interaction for
example. The proton client is inherently singled threaded, usable directly via
the container thread, and can only be used from other threads in certain
thread-safe manners. If you are unexpectedly multi threading it, that could
perhaps present as abnormal situations like this. Though I'd expect more
obvious signs of problem typically. It could be something else entirely. Hard
to say from the limited info.
It might be worth trying turning on the brokers protocol trace logging, and
leaving the clients disabled, or enabling both, and see whether it still
reproduces then, to at least confirm the specific protocol trace seen when it
happens. Giving the full output could be useful.
You dont say what specific client version you are using.
was (Author: gemmellr):
Your confirmation that it didnt reproduce with that output tallies with what I
thought on first looking at it. That still suggests an empty 'more=false' frame
at the end of a message could still be the issue (which the broker should
handle IMO, but typically shouldn't happen), but it just didn't happen here.
The question would be how it gets in that situation. The trace snippets dont
appear to have enough information to reason about any particular issue, not
showing enough time period to reason about the overall interaction.
Some general observations:
- There looks to be 1 receiver, roughly trading aggregate volume of messages
with 3 senders overall, possibly favouring some senders more than others.
- The sent messages look around 3x+ larger than those received.
- The last large message transfer seen indicates more=true, i.e it isnt
complete. But nothing else is seemingly sent. Unclear why from the snippet.
That enabling the logging seemingly influenced things, might suggest it is a
timing issue. Could be protocol interaction, or threading interaction for
example. The proton client is inherently singled threaded, usable directly via
the container thread, and can only be used from other threads in certain
thread-safe manners. If you are unexpectedly multi threading it, that could
perhaps present as abnormal situations like this. Though I'd expect more
obvious signs of problem typically. It could be something else entirely. Hard
to say from the limited info.
It might be worth trying turning on the brokers protocol trace logging, and
leaving the clients disabled, and see whether it still reproduces then, to at
least confirm the specific protocol trace seen when it happens. Giving the full
output could be useful.
You dont say what specific client version you are using.
> Randomly hit "unknown QBB implementation" error
> -----------------------------------------------
>
> Key: QPID-8496
> URL: https://issues.apache.org/jira/browse/QPID-8496
> Project: Qpid
> Issue Type: Bug
> Components: Broker-J
> Affects Versions: qpid-java-broker-8.0.2
> Reporter: Lisa Hong
> Priority: Major
> Attachments: UnknownQBB_callstack, proton.log
>
>
> Hi team,
> I'm running qpid java broker with modelVersion8.0. And the client is
> qpid-proton with python. Unfortunately, my producer will randomly hit this
> "unknown QBB implementation" error thus failing with on_connection_error.
> Could you please help to advice what could be the root cause? The error call
> stack could be found from attachment.
> I'm suspending some null buffer is provided for this
> QpidByteBufferFactory.concatenate method. But I have no idea why it could
> happen. Would you please help to figure out? Thank you so much!
> Regards,
> Lisa
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@qpid.apache.org
For additional commands, e-mail: dev-h...@qpid.apache.org
