Re: [Qemu-devel] 答复: Re: 答复: Re: [BUG]COLO failover hang

["Dr. David Alan Gilbert" <dgilbert@redhat.com>]

* Hailiang Zhang (zhang.zhanghailiang@huawei.com) wrote:
> On 2017/3/21 19:56, Dr. David Alan Gilbert wrote:
> > * Hailiang Zhang (zhang.zhanghailiang@huawei.com) wrote:
> > > Hi,
> > > 
> > > Thanks for reporting this, and i confirmed it in my test, and it is a bug.
> > > 
> > > Though we tried to call qemu_file_shutdown() to shutdown the related fd, in
> > > case COLO thread/incoming thread is stuck in read/write() while do failover,
> > > but it didn't take effect, because all the fd used by COLO (also migration)
> > > has been wrapped by qio channel, and it will not call the shutdown API if
> > > we didn't qio_channel_set_feature(QIO_CHANNEL(sioc), QIO_CHANNEL_FEATURE_SHUTDOWN).
> > > 
> > > Cc: Dr. David Alan Gilbert <dgilbert@redhat.com>
> > > 
> > > I doubted migration cancel has the same problem, it may be stuck in write()
> > > if we tried to cancel migration.
> > > 
> > > void fd_start_outgoing_migration(MigrationState *s, const char *fdname, Error **errp)
> > > {
> > >      qio_channel_set_name(QIO_CHANNEL(ioc), "migration-fd-outgoing");
> > >      migration_channel_connect(s, ioc, NULL);
> > >      ... ...
> > > We didn't call qio_channel_set_feature(QIO_CHANNEL(sioc), QIO_CHANNEL_FEATURE_SHUTDOWN) above,
> > > and the
> > > migrate_fd_cancel()
> > > {
> > >   ... ...
> > >      if (s->state == MIGRATION_STATUS_CANCELLING && f) {
> > >          qemu_file_shutdown(f);  --> This will not take effect. No ?
> > >      }
> > > }
> > 
> > (cc'd in Daniel Berrange).
> > I see that we call qio_channel_set_feature(ioc, QIO_CHANNEL_FEATURE_SHUTDOWN); at the
> > top of qio_channel_socket_new;  so I think that's safe isn't it?
> > 
> 
> Hmm, you are right, this problem is only exist for the migration incoming fd, thanks.


Yes, and I don't think we normally do a cancel on the incoming side of a migration.

Dave

> > Dave
> > 
> > > Thanks,
> > > Hailiang
> > > 
> > > On 2017/3/21 16:10, wang.guang55@zte.com.cn wrote:
> > > > Thank you。
> > > > 
> > > > I have test aready。
> > > > 
> > > > When the Primary Node panic,the Secondary Node qemu hang at the same place。
> > > > 
> > > > Incorrding http://wiki.qemu-project.org/Features/COLO ，kill Primary Node qemu will not produce the problem,but Primary Node panic can。
> > > > 
> > > > I think due to the feature of channel does not support QIO_CHANNEL_FEATURE_SHUTDOWN.
> > > > 
> > > > 
> > > > when failover,channel_shutdown could not shut down the channel.
> > > > 
> > > > 
> > > > so the colo_process_incoming_thread will hang at recvmsg.
> > > > 
> > > > 
> > > > I test a patch:
> > > > 
> > > > 
> > > > diff --git a/migration/socket.c b/migration/socket.c
> > > > 
> > > > 
> > > > index 13966f1..d65a0ea 100644
> > > > 
> > > > 
> > > > --- a/migration/socket.c
> > > > 
> > > > 
> > > > +++ b/migration/socket.c
> > > > 
> > > > 
> > > > @@ -147,8 +147,9 @@ static gboolean socket_accept_incoming_migration(QIOChannel *ioc,
> > > > 
> > > > 
> > > >        }
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 
> > > >        trace_migration_socket_incoming_accepted()
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 
> > > >        qio_channel_set_name(QIO_CHANNEL(sioc), "migration-socket-incoming")
> > > > 
> > > > 
> > > > +    qio_channel_set_feature(QIO_CHANNEL(sioc), QIO_CHANNEL_FEATURE_SHUTDOWN)
> > > > 
> > > > 
> > > >        migration_channel_process_incoming(migrate_get_current(),
> > > > 
> > > > 
> > > >                                           QIO_CHANNEL(sioc))
> > > > 
> > > > 
> > > >        object_unref(OBJECT(sioc))
> > > > 
> > > > 
> > > > 
> > > > 
> > > > My test will not hang any more.
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 原始邮件
> > > > 
> > > > 
> > > > 
> > > > 发件人： ＜zhangchen.fnst@cn.fujitsu.com＞
> > > > 收件人：王广10165992 ＜zhang.zhanghailiang@huawei.com＞
> > > > 抄送人： ＜qemu-devel@nongnu.org＞ ＜zhangchen.fnst@cn.fujitsu.com＞
> > > > 日 期 ：2017年03月21日 15:58
> > > > 主 题 ：Re: [Qemu-devel]  答复: Re:  [BUG]COLO failover hang
> > > > 
> > > > 
> > > > 
> > > > 
> > > > 
> > > > Hi,Wang.
> > > > 
> > > > You can test this branch:
> > > > 
> > > > https://github.com/coloft/qemu/tree/colo-v5.1-developing-COLO-frame-v21-with-shared-disk
> > > > 
> > > > and please follow wiki ensure your own configuration correctly.
> > > > 
> > > > http://wiki.qemu-project.org/Features/COLO
> > > > 
> > > > 
> > > > Thanks
> > > > 
> > > > Zhang Chen
> > > > 
> > > > 
> > > > On 03/21/2017 03:27 PM, wang.guang55@zte.com.cn wrote:
> > > > ＞
> > > > ＞ hi.
> > > > ＞
> > > > ＞ I test the git qemu master have the same problem.
> > > > ＞
> > > > ＞ (gdb) bt
> > > > ＞
> > > > ＞ #0  qio_channel_socket_readv (ioc=0x7f65911b4e50, iov=0x7f64ef3fd880,
> > > > ＞ niov=1, fds=0x0, nfds=0x0, errp=0x0) at io/channel-socket.c:461
> > > > ＞
> > > > ＞ #1  0x00007f658e4aa0c2 in qio_channel_read
> > > > ＞ (ioc=ioc@entry=0x7f65911b4e50, buf=buf@entry=0x7f65907cb838 "",
> > > > ＞ buflen=buflen@entry=32768, errp=errp@entry=0x0) at io/channel.c:114
> > > > ＞
> > > > ＞ #2  0x00007f658e3ea990 in channel_get_buffer (opaque=＜optimized out＞,
> > > > ＞ buf=0x7f65907cb838 "", pos=＜optimized out＞, size=32768) at
> > > > ＞ migration/qemu-file-channel.c:78
> > > > ＞
> > > > ＞ #3  0x00007f658e3e97fc in qemu_fill_buffer (f=0x7f65907cb800) at
> > > > ＞ migration/qemu-file.c:295
> > > > ＞
> > > > ＞ #4  0x00007f658e3ea2e1 in qemu_peek_byte (f=f@entry=0x7f65907cb800,
> > > > ＞ offset=offset@entry=0) at migration/qemu-file.c:555
> > > > ＞
> > > > ＞ #5  0x00007f658e3ea34b in qemu_get_byte (f=f@entry=0x7f65907cb800) at
> > > > ＞ migration/qemu-file.c:568
> > > > ＞
> > > > ＞ #6  0x00007f658e3ea552 in qemu_get_be32 (f=f@entry=0x7f65907cb800) at
> > > > ＞ migration/qemu-file.c:648
> > > > ＞
> > > > ＞ #7  0x00007f658e3e66e5 in colo_receive_message (f=0x7f65907cb800,
> > > > ＞ errp=errp@entry=0x7f64ef3fd9b0) at migration/colo.c:244
> > > > ＞
> > > > ＞ #8  0x00007f658e3e681e in colo_receive_check_message (f=＜optimized
> > > > ＞ out＞, expect_msg=expect_msg@entry=COLO_MESSAGE_VMSTATE_SEND,
> > > > ＞ errp=errp@entry=0x7f64ef3fda08)
> > > > ＞
> > > > ＞     at migration/colo.c:264
> > > > ＞
> > > > ＞ #9  0x00007f658e3e740e in colo_process_incoming_thread
> > > > ＞ (opaque=0x7f658eb30360 ＜mis_current.31286＞) at migration/colo.c:577
> > > > ＞
> > > > ＞ #10 0x00007f658be09df3 in start_thread () from /lib64/libpthread.so.0
> > > > ＞
> > > > ＞ #11 0x00007f65881983ed in clone () from /lib64/libc.so.6
> > > > ＞
> > > > ＞ (gdb) p ioc-＞name
> > > > ＞
> > > > ＞ $2 = 0x7f658ff7d5c0 "migration-socket-incoming"
> > > > ＞
> > > > ＞ (gdb) p ioc-＞features        Do not support QIO_CHANNEL_FEATURE_SHUTDOWN
> > > > ＞
> > > > ＞ $3 = 0
> > > > ＞
> > > > ＞
> > > > ＞ (gdb) bt
> > > > ＞
> > > > ＞ #0  socket_accept_incoming_migration (ioc=0x7fdcceeafa90,
> > > > ＞ condition=G_IO_IN, opaque=0x7fdcceeafa90) at migration/socket.c:137
> > > > ＞
> > > > ＞ #1  0x00007fdcc6966350 in g_main_dispatch (context=＜optimized out＞) at
> > > > ＞ gmain.c:3054
> > > > ＞
> > > > ＞ #2  g_main_context_dispatch (context=＜optimized out＞,
> > > > ＞ context@entry=0x7fdccce9f590) at gmain.c:3630
> > > > ＞
> > > > ＞ #3  0x00007fdccb8a6dcc in glib_pollfds_poll () at util/main-loop.c:213
> > > > ＞
> > > > ＞ #4  os_host_main_loop_wait (timeout=＜optimized out＞) at
> > > > ＞ util/main-loop.c:258
> > > > ＞
> > > > ＞ #5  main_loop_wait (nonblocking=nonblocking@entry=0) at
> > > > ＞ util/main-loop.c:506
> > > > ＞
> > > > ＞ #6  0x00007fdccb526187 in main_loop () at vl.c:1898
> > > > ＞
> > > > ＞ #7  main (argc=＜optimized out＞, argv=＜optimized out＞, envp=＜optimized
> > > > ＞ out＞) at vl.c:4709
> > > > ＞
> > > > ＞ (gdb) p ioc-＞features
> > > > ＞
> > > > ＞ $1 = 6
> > > > ＞
> > > > ＞ (gdb) p ioc-＞name
> > > > ＞
> > > > ＞ $2 = 0x7fdcce1b1ab0 "migration-socket-listener"
> > > > ＞
> > > > ＞
> > > > ＞ May be socket_accept_incoming_migration should
> > > > ＞ call qio_channel_set_feature(ioc, QIO_CHANNEL_FEATURE_SHUTDOWN)??
> > > > ＞
> > > > ＞
> > > > ＞ thank you.
> > > > ＞
> > > > ＞
> > > > ＞
> > > > ＞
> > > > ＞
> > > > ＞ 原始邮件
> > > > ＞ *发件人：*＜zhangchen.fnst@cn.fujitsu.com＞
> > > > ＞ *收件人：*王广10165992＜qemu-devel@nongnu.org＞
> > > > ＞ *抄送人：*＜zhangchen.fnst@cn.fujitsu.com＞＜zhang.zhanghailiang@huawei.com＞
> > > > ＞ *日 期 ：*2017年03月16日 14:46
> > > > ＞ *主 题 ：**Re: [Qemu-devel] COLO failover hang*
> > > > ＞
> > > > ＞
> > > > ＞
> > > > ＞
> > > > ＞ On 03/15/2017 05:06 PM, wangguang wrote:
> > > > ＞ ＞   am testing QEMU COLO feature described here [QEMU
> > > > ＞ ＞ Wiki](http://wiki.qemu-project.org/Features/COLO).
> > > > ＞ ＞
> > > > ＞ ＞ When the Primary Node panic,the Secondary Node qemu hang.
> > > > ＞ ＞ hang at recvmsg in qio_channel_socket_readv.
> > > > ＞ ＞ And  I run  { 'execute': 'nbd-server-stop' } and { "execute":
> > > > ＞ ＞ "x-colo-lost-heartbeat" } in Secondary VM's
> > > > ＞ ＞ monitor,the  Secondary Node qemu still hang at recvmsg .
> > > > ＞ ＞
> > > > ＞ ＞ I found that the colo in qemu is not complete yet.
> > > > ＞ ＞ Do the colo have any plan for development?
> > > > ＞
> > > > ＞ Yes, We are developing. You can see some of patch we pushing.
> > > > ＞
> > > > ＞ ＞ Has anyone ever run it successfully? Any help is appreciated!
> > > > ＞
> > > > ＞ In our internal version can run it successfully,
> > > > ＞ The failover detail you can ask Zhanghailiang for help.
> > > > ＞ Next time if you have some question about COLO,
> > > > ＞ please cc me and zhanghailiang ＜zhang.zhanghailiang@huawei.com＞.
> > > > ＞
> > > > ＞
> > > > ＞ Thanks
> > > > ＞ Zhang Chen
> > > > ＞
> > > > ＞
> > > > ＞ ＞
> > > > ＞ ＞
> > > > ＞ ＞
> > > > ＞ ＞ centos7.2+qemu2.7.50
> > > > ＞ ＞ (gdb) bt
> > > > ＞ ＞ #0  0x00007f3e00cc86ad in recvmsg () from /lib64/libpthread.so.0
> > > > ＞ ＞ #1  0x00007f3e0332b738 in qio_channel_socket_readv (ioc=＜optimized out＞,
> > > > ＞ ＞ iov=＜optimized out＞, niov=＜optimized out＞, fds=0x0, nfds=0x0, errp=0x0) at
> > > > ＞ ＞ io/channel-socket.c:497
> > > > ＞ ＞ #2  0x00007f3e03329472 in qio_channel_read (ioc=ioc@entry=0x7f3e05110e40,
> > > > ＞ ＞ buf=buf@entry=0x7f3e05910f38 "", buflen=buflen@entry=32768,
> > > > ＞ ＞ errp=errp@entry=0x0) at io/channel.c:97
> > > > ＞ ＞ #3  0x00007f3e032750e0 in channel_get_buffer (opaque=＜optimized out＞,
> > > > ＞ ＞ buf=0x7f3e05910f38 "", pos=＜optimized out＞, size=32768) at
> > > > ＞ ＞ migration/qemu-file-channel.c:78
> > > > ＞ ＞ #4  0x00007f3e0327412c in qemu_fill_buffer (f=0x7f3e05910f00) at
> > > > ＞ ＞ migration/qemu-file.c:257
> > > > ＞ ＞ #5  0x00007f3e03274a41 in qemu_peek_byte (f=f@entry=0x7f3e05910f00,
> > > > ＞ ＞ offset=offset@entry=0) at migration/qemu-file.c:510
> > > > ＞ ＞ #6  0x00007f3e03274aab in qemu_get_byte (f=f@entry=0x7f3e05910f00) at
> > > > ＞ ＞ migration/qemu-file.c:523
> > > > ＞ ＞ #7  0x00007f3e03274cb2 in qemu_get_be32 (f=f@entry=0x7f3e05910f00) at
> > > > ＞ ＞ migration/qemu-file.c:603
> > > > ＞ ＞ #8  0x00007f3e03271735 in colo_receive_message (f=0x7f3e05910f00,
> > > > ＞ ＞ errp=errp@entry=0x7f3d62bfaa50) at migration/colo.c:215
> > > > ＞ ＞ #9  0x00007f3e0327250d in colo_wait_handle_message (errp=0x7f3d62bfaa48,
> > > > ＞ ＞ checkpoint_request=＜synthetic pointer＞, f=＜optimized out＞) at
> > > > ＞ ＞ migration/colo.c:546
> > > > ＞ ＞ #10 colo_process_incoming_thread (opaque=0x7f3e067245e0) at
> > > > ＞ ＞ migration/colo.c:649
> > > > ＞ ＞ #11 0x00007f3e00cc1df3 in start_thread () from /lib64/libpthread.so.0
> > > > ＞ ＞ #12 0x00007f3dfc9c03ed in clone () from /lib64/libc.so.6
> > > > ＞ ＞
> > > > ＞ ＞
> > > > ＞ ＞
> > > > ＞ ＞
> > > > ＞ ＞
> > > > ＞ ＞ --
> > > > ＞ ＞ View this message in context: http://qemu.11.n7.nabble.com/COLO-failover-hang-tp473250.html
> > > > ＞ ＞ Sent from the Developer mailing list archive at Nabble.com.
> > > > ＞ ＞
> > > > ＞ ＞
> > > > ＞ ＞
> > > > ＞ ＞
> > > > ＞
> > > > ＞ --
> > > > ＞ Thanks
> > > > ＞ Zhang Chen
> > > > ＞
> > > > ＞
> > > > ＞
> > > > ＞
> > > > ＞
> > > > 
> > > 
> > --
> > Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK
> > 
> > .
> > 
> 
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK