-
Notifications
You must be signed in to change notification settings - Fork 453
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Spurious ZMQ disconnects in case of oversubscribe CPU #636
Comments
@jsmoon please review |
@hhaim looks good. I think this is related to #336 and I'll close it now also. For your information, I've been used to check the event messages from the additional trex-console client. And I have a minor suggestion regarding this change. |
@jsmoon I didn't realize that my patch solve #366. now all the events are in any way in sync beacuse the async is reading from a queue which is reliable. async is just an interrupt to read the queue (per session). In regard to the verbose mode, what happen before this patch? I don't expect so many events with small number of profiles. could you elaborate your suggestion? |
@hhaim I got the following message when I set verbose on in the trex-console.
The rate is about 2 times per second. Is there any periodic event on the SUB/PUB channel? |
@jsmoon understood now, in case of SUB/PUB timeout (500msec) there is a read from the queue too. maybe in case of debug we can disable the call in case of timeout |
Could you check if this helps
|
@hhaim it doesn't work. Each time, the following event message is received periodically.
I found that |
events should have
|
@hhaim no, it's one of the periodic messages generated by the following routine.
|
@jsmoon, I missed removing this, it is not needed in interactive mode. |
@hhaim, ok. |
@hhaim, ok. I'll keep on watching it. |
@jsmoon I see the other threads on backward compatibility |
Signed-off-by: Hanoh Haim <hhaim@cisco.com>
The issue: Python client reports as disconnected while the server is up. re-connect is possible.
The reason: ZMQ SUB/PUB report events to the clients, some of the events are critical for the state of the client so it must be reliable while SUB/PUB can discard messages
How to reproduce: run
iperf3
orstress-ng
on the machine and attach it to master (taskset
) core and connect with 1000 threads from a remoteThe solution: in case of interactive (STL/ASTF) save the events into a queue and add an RPC to pool the the queue events. STF will keep the same interface. The MAJOR version was incremented so only latest clients could work with the new server. (GUI need to be updated)
The text was updated successfully, but these errors were encountered: