NET - deadlock during ReverseConnection

golemfactory / yagna

An open platform and marketplace for distributed computations

GNU General Public License v3.0

387 stars 61 forks source link

NET - deadlock during ReverseConnection #2501

Open nieznanysprawiciel opened 1 year ago

nieznanysprawiciel commented 1 year ago

NET can deadlock during establishing p2p connection using ReverseConnection message

Reproduction:

You need to have yagna ports open for Nodes to be able to connect to you
Increasing number of neighbors to 100: YA_NET_BROADCAST_SIZE=100
Wait for Nodes connecting and disconnecting
To check if yagna deadlocked, run yagna net sessions

It is non-deterministic bug. tt doesn't happen always in this scenario, but quite often.

nieznanysprawiciel commented 1 year ago

Interesting fact: when establishing connections, yagna net sessions command sometimes waits much longer than normally, but finishes correctly afterwards. It indicates, that somewhere we are holding lock for too long.

golmek commented 1 year ago

To be verified after blocking tickets are done

Blocked by:

https://github.com/golemfactory/ya-relay/issues/238

nieznanysprawiciel commented 4 months ago

Should be already solved. Requires verification