Very poor ARM performance about canto-next HOT 8 CLOSED

I think this is related to your other issue. In #41 the daemon is crashing because the Python standard threading library throws an unexpected exception that it couldn't start a thread. That shouldn't happen unless Linux is out of pids or something crazy like that.

Can you attempt to reproduce this issue on 0.9.7 or git instead of 0.9.1?

If you can reproduce, it could be that either multiprocessing.cpu_count() is returning something insane, or it could be that Python isn't cleaning up threads properly because on x86_64 and daemon 0.9.7 this has been pretty much bulletproof

from canto-next.

I have reproduced it on 0.9.7: here is the log. I can also say with great certainty that multiprocessing.cpu_count() is not to blame here, as it returns the very sensible 2 on the server machine when tested.

from canto-next.

The top line of that log is still daemon 0.9.1...

from canto-next.

I just noticed in the backtrace you just ran bin/canto-daemon from the source. You have to actually install canto-next. If you don't want to install to root, you can use a prefix with setup.py and change your PYTHONPATH. Some details here

from canto-next.

Ah, OK. Sorry about that - I'm not very Python-literate. Will read and try again.

from canto-next.

Sorry for the delay - I was in a different country for the last two weeks. I've modified my version to match, but now I get even weirder behaviour: specifically, canto-curses doesn't even connect to it, and as the log indicates, it seems to just time out. For reference, canto-curses --version outputs the version as 0.9.3 on my machine.

from canto-next.

That daemon log doesn't look bad, and isn't spawning threads everywhere so I'll call that progress. The curses-log is the file that's probably interesting in this case, but you're also six releases behind there.

Try with canto-curses 0.9.9 and see if you can reproduce. If you can, attach both daemon- and curses-log from the same directory.

from canto-next.

Alrighty, having tested with canto-curses 0.9.9, I can confirm that performance is fine. It's a little bit slow, but given the (lack of) horsepower on the device, and the fact that I'm running it over an SSH tunnel, I think it's OK. Gonna close this now - thank you for your patience.

from canto-next.