-
Notifications
You must be signed in to change notification settings - Fork 18
Connections not fully closed #21
Comments
Note: Unlike the other connection closing issue, this was not a symptom of that secio change and/or a lack of connection closing. This is still a big. |
This is after 1.5w uptime on mercury:
|
This is really an infuriating bug.
This means that either:
If no multiplex session had been started, I would have expected something to call Otherwise, if the sessions had been closed, they should have called Is there anything in the logs about "failed to convert nconn.{Local,Remote}Addr: ..."? |
Maybe related to #21. Probably not.
So I have to keep checking this..., this is stuck waiting on the context, not |
Yes a few, it's all just earth dialing out though. Right now there's no significant leak happening though. We should look again when there is.
|
So, this appears to happen when a node connects, disconnects, and then reconnects. To reproduce:
IPFS node 1 will leak |
I should have fixed any leak there in #27. |
Otherwise, we leak them. fixes libp2p/go-ws-transport#21
So, this was actually a bug in go-peerstream. It was failing to close connections after failing to negotiate a stream muxer. I tried to also fix this in this package by modifying So, the new rule is, if you get an error on a connection, you must close it. |
ServeHTTP
go routines are getting left around.The text was updated successfully, but these errors were encountered: