So I was able to test this branch. I setup the cluster to sync from the region every 10 seconds. With it syncing every 10 seconds with this change there where no CLOSE_WAIT sockets.
sudo lsof | grep maas | grep CLOSE_WAIT | wc -l
0
With this change disabled, so simplestreams using python-requests I got many CLOSE_WAITS
And it keeps growing. This fixes the problem lets land this now. So 1.8 is stable, without the cluster will need to keep being restarted. At some point people will reach a limit no matter the ulimit.
So I was able to test this branch. I setup the cluster to sync from the region every 10 seconds. With it syncing every 10 seconds with this change there where no CLOSE_WAIT sockets.
sudo lsof | grep maas | grep CLOSE_WAIT | wc -l
0
With this change disabled, so simplestreams using python-requests I got many CLOSE_WAITS
sudo lsof | grep maas | grep CLOSE_WAIT | wc -l
88
sudo lsof | grep maas | grep CLOSE_WAIT | wc -l
104
And it keeps growing. This fixes the problem lets land this now. So 1.8 is stable, without the cluster will need to keep being restarted. At some point people will reach a limit no matter the ulimit.