[CRIU] Do anyone knows how to migrate a container with web app

Tue May 3 09:06:32 PDT 2016

I know what you mean. But I just want to migrate some containers to another
host, all of those application can deal tcp reconnection.
You say if there are some packets cannot deliver into new container's
location, the restore will broken.
So, If there is* a method to skip this progress, just to bind the old ip
address to the container on the new host.*
Everytime I checkpoint a container with tcp connection, scp to another
host, then restore, the restore log always say that

(00.294516)      1: Found id pipe:[6392267] (fd 1) in inherit fd list
(00.294520)      1: File pipe:[6392267] will be restored from fd 3 dumped
from inherit fd 1
(00.294535)      1: Found fd 1 (id pipe:[6392267]) in inherit fd list
(caller inherit_fd_resolve_clash)
(00.294538)      1: Inherit fd 1 moved to 3 to resolve clash
(00.294541)      1:             Create fd for 1
(00.294544)      1:     Restoring fd 2 (state -> create)
(00.294545)      1:             Creating pipe pipe_id=0x6189cc id=0x3
(00.294547)      1: Found id pipe:[6392268] (fd 2) in inherit fd list
(00.294548)      1: File pipe:[6392268] will be restored from fd 4 dumped
from inherit fd 2
(00.294554)      1: Found fd 2 (id pipe:[6392268]) in inherit fd list
(caller inherit_fd_resolve_clash)
(00.294555)      1: Inherit fd 2 moved to 4 to resolve clash
(00.294558)      1:             Create fd for 2
(00.294560)      1:     Restoring fd 4 (state -> create)
(00.294569)      1:     Restore: family 10 type 1 proto 6 port 6379 state
10 src_addr ::
(00.294592)      1: 5 restore sndbuf 16384 rcv buf 87380
(00.294595)      1:     restore priority 0 for socket
(00.294596)      1:     restore rcvlowat 1 for socket
(00.294598)      1:     restore mark 0 for socket
(00.294611)      1: Found fd 4 (id pipe:[6392268]) in inherit fd list
(caller inherit_fd_resolve_clash)
(00.294616)      1: Inherit fd 4 moved to 6 to resolve clash
(00.294619)      1:             Create fd for 4
(00.294622)      1:     Restoring fd 5 (state -> create)
(00.294625)      1:     Restore: family 2 type 1 proto 6 port 6379 state 10
src_addr 0.0.0.0
(00.294633)      1: 5 restore sndbuf 16384 rcv buf 87380
(00.294635)      1:     restore priority 0 for socket
(00.294636)      1:     restore rcvlowat 1 for socket
(00.294638)      1:     restore mark 0 for socket
(00.294640)      1:             Create fd for 5
(00.294642)      1:     Restoring fd 6 (state -> create)
(00.294644)      1:     Restore: family 2 type 1 proto 6 port 6379 state 1
src_addr 172.17.128.2
(00.294646)      1: Restoring TCP connection
(00.294649)      1: Restoring TCP connection id 7 ino 619526
(00.294657)      1:     Setting 1 queue seq to 2786935679
(00.294659)      1:     Setting 2 queue seq to 919486667
(00.294679)      1: Error (sk-inet.c:721): Can't bind inet socket (id 7):
Cannot assign requested address

You see, 172.17.128.2 is the old ip, criu tried to bind this ip to new
container, but failed.

Regards,

David Sun 孙炜程
School of Software Engineering, Shanghai Jiaotong University

Mobile:  15921638078

E-mail:  SunWeicheng0001 at gmail.com

2016-04-28 18:33 GMT+08:00 Pavel Emelyanov <xemul at virtuozzo.com>:

> On 04/28/2016 08:01 AM, Weicheng SUN wrote:
> > Hi Everyone,
> >
> > The latest release of cr-combined said that it supports network. Then I
> do some experiments. I use
> > 1. docker run -d -v demo:/data --name=redis redis:latest
> > 2. docker run -d -e USE_REDIS_HOST=redis --link redis:redis -p 80:80
> --name=app binocarlos/moby-counter:latest
> > which is a demo of simple website and redis is its storage. I first run
> them on host A
> >
> > I try to checkpoint them and copy to another host(host B). I create two
> containers with same configuration
> > of them and applied the change of file system to the new created
> container. Then I tried to restore them,
> > which sometimes came out a wrong result.
> > I notice that when redis and app have same ip address on host A and B,
> then the restore will success. But,
> > when they have different IP on two hosts, the restore progress will
> certainly broken.
> >
> > The reason I guess is that the state of tcp connection in checkpoint
> image didn't match the new environment
> > in host B. For example, redis in host A is 17.172.20.1 and app in host A
> is 17.172.20.2, then for app there
> > is a connection whose remote host is 17.172.20.1. Then come to host B,
> if redis in host B is 17.172.20.3 and
> > when I try to restore app, criu will confuse where is 17.172.20.1(redis)?
>
> After migration all the sockets get restored with the IP addresses they
> had on source node,
> so if for some reason packets to former container IP cannot be delivered
> into container's
> new location, the container would get broken. This is unavoidable
> limitation of TCP protocol
> itself. There can be workarounds to this (DNAT-like for example), but this
> is out of CRIU's
> scope and should be done on Docker level.
>
> > Generally, is there always a problem to checkpoint and restore a
> container, who has a link with other
> > containers, on the another host?
> >
> > And who knows that how to use a specific IP address while start a new
> container? For the latest release of
> > cr-combined doesn't support docker run --ip. Even --ip is supported, it
> requires a user defined network,
> > default network is not supported.
> >
> > Hopes for your kindly response & wish all of you a great day!
> >
> > Regards,
> >
> > David Sun
> >
> >
> >
> > _______________________________________________
> > CRIU mailing list
> > CRIU at openvz.org
> > https://lists.openvz.org/mailman/listinfo/criu
> >
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openvz.org/pipermail/criu/attachments/20160504/a77f4c91/attachment-0001.html>