Add example of reopening channels that have been closed #364
Conversation
bluesliverx
force-pushed
the
master
branch
2 times, most recently
from
4c80718
to
38893c7
Compare
|
Thanks @Sinkler for catching that. I found that I actually needed some more infrastructure around this to work properly since channels were attempting to reinitialize queues and exchanges on reopen. I've changed my commit to make sure things look good. |
| async def reopen(self) -> None: | ||
| # Clear out exchanges and queues when reopened | ||
| self._exchanges = defaultdict(set) | ||
| self._queues = defaultdict(set) |
There was a problem hiding this comment.
By the way, we had some memory leaks with set so we replaced it by WeakSet in the __init__ method and now it's ok 🙃
There was a problem hiding this comment.
Thanks for the info! I will make those same changes...
|
@mosquito the py35/py36-uvloop tox tests are failing, but I'm not sure what to do about those. They appear to be unrelated to my PR. |
|
@mosquito I have updated the PR if you could take a look again. Thanks! |
|
Ran across this again, any chance of getting it merged @mosquito? |
|
@bluesliverx ouch, I should be write this comment earlier, but doesn't sorry. I completely do not understand why I will want to apply this example in real project when to be honest. When you want do not reconnecting when the underlay-connection faills you shouldn't use |
|
@mosquito, I do want the robust connection since I want it to re-established. However, I do not want the queues and exchanges to be declared again since some of them are exclusive and cannot be recreated by a new connection. I believe rabbit threw errors in that case. As for why I need the |
This bit me recently when channels were closed unexpectedly due to long running message retrieval on an API where client connections were closed suddenly. This apparently put the channels in a bad state (
ChannelInvalidStateError("writer is None")to be exact), and therefore I had to check channels when retrieving them from the pool.