On 04:19 am, luper.ro...@gmail.com wrote:
My first attempt was to run the reactor in the main thread, but the
application was "freezing" when e.g. doing a SELECT with SQLAlchemy/
psycopg2 (I guess because psycopg2 just waits for data and blocks
the twisted reactor). So I put reactor.run() in a thread and it
works, but I'm getting strange issues on some platforms (for
example I can't connect at all on *some* OSX computers, getting a
"reactor stopping" log from twisted without any further error), and
I feel I'm doing the whole thing terribly wrong.
These errors sound like you are making Twisted API calls from
threads other than the reactor thread by accident. There's no
obvious fix except for "don't do that" :).
My suggestion would be to move the SQLAlchemy/psycopg2 into threads,
with e.g. deferToThread, rather than putting the reactor into a
thread. Then the reactor won't freeze while the blocking SQL stuff
happens.
So my question is, how to do proper SSH port forwarding with conch
in an application that also uses the tunnels not the asynchronous
way ? If running the reactor in a thread is OK, are there
particular things to watch out for ?
There's no reason that running the reactor in a thread shouldn't
work; it's just generally tricky (especially if you're working with
an existing codebase) to completely avoid calling Twisted APIs from
other threads.
Just wanted to thank you, I isolated the Twisted part of my
application using the 'multiprocessing' module, and all the weird
issues are gone, it seems I was mixing threads somewhere ...
So executing the reactor in a child process was a good solution for my
problem, no need to use deferToThread everywhere or to do tedious
research about what is called from where.
Cheers,
Luper
_______________________________________________
Twisted-Python mailing list
Twisted-Python@twistedmatrix.com
http://twistedmatrix.com/cgi-bin/mailman/listinfo/twisted-python
