[CRIU] [Users] socket will take at least 0.5 seconds to recovery after docker restore done

Pavel Emelyanov xemul at parallels.com
Fri Jul 31 05:23:36 PDT 2015


On 07/31/2015 09:54 AM, Yanbao Cui wrote:
> HI,
> 
> Sorry for give the relative timestamp.
> 
> Yeah, the gap cover my checkpoint/restore time. And larger than the whole migration time.
> 
> It seems we can not find why the connection hang from the dump file

Well, it's possible, but it would require more advanced debugging :) I would
patch/kprobe the kernel to printk() messages at these 2 points:

1. tcp_send_window_probe()
2. tcp_v4_rcv()

for the connection in question and checked the gap between the real repair-off
and the response.

-- Pavel


More information about the CRIU mailing list