Redis issues about cyclone HOT 22 CLOSED

notification.peer.0847b249-6af1-42f2-ae41-d2bc972e6576 is a channel I use. Also, downgrading to rc8 fixes this issue and problem does not present itself.

from cyclone.

i'm trying to replicate this issue but i can't. the changes in redis driver are on sharded connections only... also, all test cases pass, and cyclone's redis demo is working fine here: https://github.com/fiorix/cyclone/tree/master/demos/redis

all operations and pub/sub are apparently working as expected. could you please instruct me on how to replicate it?

from cyclone.

Managed to replicate it using demo you linked.

Ran it, subscribed to some channel 'q':
curl -D - http://localhost:8888/queue/q

created test_redis.py with contents:
https://gist.github.com/3251299

Now with rc8 I am getting flawless answers:
http://pastebin.com/L174kckN

With rc10:
some are good, but I am also getting crap + lots of content is cut:
http://pastebin.com/0WZTzGg3

from cyclone.

Also, one more issue - this actually happens with rc8, haven't tried rc10.

I ran demo. Played with it. Waited some time and I got:
http://pastebin.com/zb9zgHZG

This happend after 5 minutes and it is not properly reconnecting. After this I try to subscribe etc - nothing works until I restart it.

from cyclone.

ok, i'll try to replicate myself and figure out how to fix it and what's the relation between last night's changes...
as for the disconnect issue, i have to investigate... first thing i'd recommend is to disable redis' timeout for idle connections

from cyclone.

Ok never mind that. It was just caused due to unsubscribe_all() function missed at the end RedisMixin.channels.clear()

Not a cyclone bug, just a demo bug.

from cyclone.

Of course I was referring to the disconnect issue. Not the main one.

from cyclone.

bug here: https://github.com/fiorix/cyclone/blob/master/cyclone/redis.py#L247

gotta fix

from cyclone.

Glad you figured that out! Hoping for a fix soon as this is rather huge bug for anyone integrating cyclone with redis.

from cyclone.

I think I got it working. Please check and let me know, so I can close this and upload the new version to pypi.

Here's what happened: http://musta.sh/2012-08-08/a-victim-of-recursion.html

from cyclone.

Yup, seems to be working just fine!

One thing though. This is not really a bug and I am not sure if this isn't redis itself. I found that sending multiple zeroes through channel would result in only one zero being received. But I am not talking about ints where it would be obvious.

redis_client.publish('q', '00000000000')
q: 0

This presents itself only if there is nothing left but the zeroes. E.g. q: 00000000000} is sent correctly.
Seems like str.isdigit() -> int(str) is being done somewhere. But no idea if on redis or on cyclone side. Anyway, this is not an issue of any kind unless someone would find usefulness in sending a bunch of zeroes ;)

from cyclone.

That's the automatic conversion the driver makes: https://github.com/fiorix/txredisapi/blob/master/txredisapi.py#L317

It tries to figure out whether it's a float, or integer, or unicode, otherwise it defaults to "as-is" - and probably will become a binary later in py3.

I guess it's ok to close this ticket now. Please let me know if you find anything else. Thanks!

from cyclone.

I think that might automagically affected/fixed websockets. Will check it out later.

from cyclone.

I highly doubt it.
Changes were made to txredisapi.

IlyaSkriblovsky/txredisapi@5f9e76e

from cyclone.

I mean that at some point websockets uses LineReceiver changing to raw mode. Gotta check it tho.

from cyclone.

i don't think web sockets keep switching back and forth from line mode to raw mode, but feel free to check and let me know if you foresee any issues with that

the hand crafted version of the line receiver can be used anywhere, but should be avoided unless we hit the recursion depth thing

from cyclone.

Where is the push to pypi coming? This bug breaks cyclone redis completely so I thought it will be of higher priority to push it.

from cyclone.

It was fix 18 days ago, when I closed the ticket.

The reason you don't see the push here is because we keep a separate repo just for the redis driver, with it's own documentation and unit tests: https://github.com/fiorix/txredisapi

The driver is then synced/copied to cyclone's tree.

from cyclone.

Wait. I didnt mean push to repo but an updated pypi version. It's still rc10. Unless I'm missing something and downloading txredisapi is done during automatically installation through pip/easy_install?

from cyclone.

Or is it even different - cyclone rc10 is a fixed version now. Just no version bump, reinstall for those experiencing the issue?

from cyclone.

Sorry, my bad.
I'll update both github and pypi to rc11

Sent from the future

On 2012-08-26, at 2:24 PM, Robert Kopaczewski [email protected] wrote:

Or is it even different - cyclone rc10 is a fixed version now. Just no version bump, reinstall for those experiencing the issue?

—
Reply to this email directly or view it on GitHub.

from cyclone.

done

from cyclone.