I have recently come across an issue with Proxmox VE 8 and Proxmox Backup Server 3.
Every couple of hours, a log message will appear saying VE couldn't connect to the backup server.
After doing a packet capture, I saw that Proxmox is creating two connections at a time of about 22-27 packets every 10 seconds.
Each of these connections is closed with a RST packet by VE.
This leaves sockets in the time_wait state, which by default will not go away for a few minutes (in which many other connections will be started and stopped).
The source port used by VE will occasionally conflict with one of these dead connections, and the PBS server will respond with a RST packet after syn and cause the log message.
I would suggest that the frequency of connections be reduced. 10 Seconds seems excessive. I'm not entirely sure why this time period was chosen or if its even technically necessary for a proper functioning backup chain (if its just a heartbeat, why not a couple minutes between pings?).
I don't mean to nitpick here, but it seems a little sloppy to be creating these extremely short (both data and time) connections so often and not address the very real problem of filling each system with improperly closed sockets.
Every couple of hours, a log message will appear saying VE couldn't connect to the backup server.
After doing a packet capture, I saw that Proxmox is creating two connections at a time of about 22-27 packets every 10 seconds.
Each of these connections is closed with a RST packet by VE.
This leaves sockets in the time_wait state, which by default will not go away for a few minutes (in which many other connections will be started and stopped).
The source port used by VE will occasionally conflict with one of these dead connections, and the PBS server will respond with a RST packet after syn and cause the log message.
I would suggest that the frequency of connections be reduced. 10 Seconds seems excessive. I'm not entirely sure why this time period was chosen or if its even technically necessary for a proper functioning backup chain (if its just a heartbeat, why not a couple minutes between pings?).
I don't mean to nitpick here, but it seems a little sloppy to be creating these extremely short (both data and time) connections so often and not address the very real problem of filling each system with improperly closed sockets.