[pbs-devel] [RFC 0/2] backup client: implement some HTTP timeouts

Stefan Reiter s.reiter at proxmox.com
Mon Dec 21 16:49:22 CET 2020

On 12/21/20 4:36 PM, Dietmar Maurer wrote:
>> Not a comprehensive overhaul, but should fix the most common hangs to at least
>> finish *sometime*.
> We already have TCP timeouts. So why exactly do we need those short timeouts?

It's not a TCP timeout if the server hangs. This prevents the case where 
the client has a connection established but the server fails to send 
data within the given time.

Came up during discussion of this report:

where high load on the server (from too many verification tasks, which 
is a different problem) causes VM clients to hang for unreasonable 
amounts of time.

(Note that with QEMU 5.2 we can easily do the 'connect' async in the 
background as well, preventing the full VM to hang, but we still need 
some way to timeout the connection attempt, lest it stays active in the 

More information about the pbs-devel mailing list