cleanup connections about django-sse HOT 10 CLOSED

Hi!

I'm glad you're interested in the project.

In respect for your question. Actually on a pubsub class is removed by a garbage collector, it disconnects the connection and releases a connection to the pool. Theoretically, there is no problem.

Are you find some unusual behavior?

from django-sse.

Hi,
You are right, connection will get closed when GC kicks because there are not extra references in your code that can lead to a connection leak.
My concern is about something else, correct me if am wrong but as far as i know when a client closes its connection the web server will not notice it until a write is performed because redis pubsub does not (for obvious reasons) have a timeout.
This way the connection pool can get full at some point or the connection count really high.

from django-sse.

I think you're right in what you say. I'll do some research and see if the behavior need to look for a solution or behavior is appropriate.

Thank you very much! I mention the conclusions, in a couple of days at most.

from django-sse.

I really liked your approach with class based views :)
I made sort of copycat project for longpolling and implemented a timeout approach to avoid this problem: https://github.com/tbarbugli/django_longpolling

from django-sse.

Sorry it took me to respond, I have a lot of work ...
I've been doing some research, and sure enough, the connection is not returned to the connection pool.

Your approach is good, but I think it only works well for long polling.
More research is needed, now do not know the best way to handle this ...

Once again thanks for making me see this problem.

from django-sse.

Any news here? I have also runned into this proble.

Too many open files errors, and 500+ of "addr=127.0.0.1:54999 fd=226 name= age=127922 idle=127922 flags=N db=0 sub=1 psub=0 multi=-1 qbuf=0 qbuf-free=0 obl=0 oll=0 omem=0 events=r cmd=subscribe" listed with "redis-cli client list"..
Any more information I can provide to make it easier to solve this problem?

from django-sse.

you have few options (last two need should be combined)
fix the library to closed unused connections or to only use 1 redis
connection per webserver process
increase kernel's file descriptor (and redis max connections setting) for
redis server
restart your webserver every x hours (depending on how quickly you run out
of connections)

2013/12/7 xeor [email protected]

Any news here? I have also runned into this proble.

Too many open files errors, and 500+ of "addr=127.0.0.1:54999 fd=226
name= age=127922 idle=127922 flags=N db=0 sub=1 psub=0 multi=-1 qbuf=0
qbuf-free=0 obl=0 oll=0 omem=0 events=r cmd=subscribe" listed with
"redis-cli client list"..
Any more information I can provide to make it easier to solve this problem?

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/9#issuecomment-30052980
.

from django-sse.

Thanks for the quick reply..

About using 1 redis connection, I see https://github.com/niwibe/django-sse/blob/master/django_sse/redisqueue.py#L55 , but that information is static for each connection.. Where can I change the logic, so I can example enforce max 10 connections per ip? I am using uniqe channels per user, so having max 10 connections is my goal.

Inceasing the file descriptor count is something I will do, but not in a development face where I want to detect leaks like this :)

I guess I can restart the web-server holding the sse connections once in a while.. But that really feels dirty..

Is there a way to kill off connections that havent gotten any data in 1 minute or so? My sse server is sending out pings every 40 sec, and the client will try to reconnect if there is no data for 45 sec.

from django-sse.

To whoever else wants to do this. I went for something like

redis = redisqueue._connect()
kill_status = [ redis.client_kill(c['addr']) for c in redis.client_list() if int(c['idle']) > 360 and c['cmd'] == 'subscribe']
print('Killed %s redis pubsub connections' % str(len(kill_status)))

in my already existing rabbitmq schedueler. Which also takes care of pinging all connected sse connections every 40 seconds. Them theirself makes sure with javascript code to reconnect if they dont have any data for 45 seconds.

from django-sse.

This project is unmaintained. Sorry for inconvenience.

from django-sse.