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

[Pavel Emelyanov]

On 07/23/2015 03:33 PM, Yanbao Cui wrote:
> Hi Pavel,

Hi. Sorry for the late response. See my comments inline.

> my test env and test results are as follows
> 
> ENV:____
> 
> tcp client on host: 192.168.9.170____
> 
> tcp server in docker: 192.168.9.129____
> 
> udp client in docker: 192.168.9.129____
> 
> udp server on host: 192.168.9.170____
> 
> __ __
> 
> migrate docker container from host 192.168.9.170 to host 192.168.9.180____
> 
> __ __
> 
> Summarize:____
> 
> migration time: checkpoint+restore = 0.569+1.709 = 2.278s____
> 
> existing socket hang time: 3.6s, actually is 3.6-2.278 = *1.322s*, except migration downtime____
> 
> __ __
> 
> In tcpdump file timestamp is 3.941260 - 7.619094 for tcp, timestamp 4.012874 - 7.619208 for udp____

I don't see such timestamps in the dump file.

I see the only huge gap in tcpdump timestamps, here it is:

03:44:26.080341 IP 192.168.9.129.5925 > 192.168.9.170.38248: Flags [.], ack 81, win 227, options [nop,nop,TS val 578104050 ecr 578104050], length 0
03:44:26.151955 IP 192.168.9.129.46765 > 192.168.9.170.8557: UDP, length 19
03:44:29.758175 IP 192.168.9.129.5925 > 192.168.9.170.38248: Flags [.], ack 81, win 229, options [nop,nop,TS val 578104080 ecr 0], length 0
03:44:29.758239 IP 192.168.9.170.38248 > 192.168.9.129.5925: Flags [.], ack 3380, win 7378, options [nop,nop,TS val 578104970 ecr 578104050], length 0

The packet sent at 03:44:29.758175 seem to be the window probe. After this the communication seem to happen OK

03:44:29.758954 IP 192.168.9.129.5925 > 192.168.9.170.38248: Flags [P.], seq 3380:4080, ack 81, win 229, options [nop,nop,TS val 578104330 ecr 578104970], length 700
03:44:29.759179 IP 192.168.9.170.38248 > 192.168.9.129.5925: Flags [P.], seq 81:91, ack 4080, win 7401, options [nop,nop,TS val 578104970 ecr 578104330], length 10

Do the times 3:44:29... from tcpdump match your restore time?

-- Pavel