Communication issues between NBD driver and NBDKit server
Hi, I am observing some strange errors when using the Kernel's NBD driver with NBDkit. On the kernel side, I see: May 15 16:16:11 vostro.rath.org kernel: nbd0: detected capacity change from 0 to 104857600 May 15 16:16:11 vostro.rath.org kernel: nbd1: detected capacity change from 0 to 104857600 May 15 16:18:23 vostro.rath.org kernel: block nbd0: Possible stuck request 00000000ae5feee7: control (write@4836316160,32768B). Runtime 30 seconds May 15 16:18:25 vostro.rath.org kernel: block nbd0: Possible stuck request 000000007094eddc: control (write@5372947456,10240B). Runtime 30 seconds May 15 16:18:27 vostro.rath.org kernel: block nbd0: Suspicious reply 89 (status 0 flags 0) May 15 16:18:31 vostro.rath.org kernel: block nbd0: Possible stuck request 0000000075f8b9bc: control (write@8057764864,32768B). Runtime 30 seconds May 15 16:18:41 vostro.rath.org kernel: block nbd0: Possible stuck request 000000002d1b3e8b: control (write@14499979264,32768B). Runtime 30 seconds [...] And userspace ('zfs snapshot" in this instance) is stuck afterwards. On the NBDkit side, there seemingly are write errors when replying back to the kernel: $ nbdkit --unix /tmp/tmpi5o59_y_/nbd_socket_sb --foreground --filter=exitlast --filter=stats --threads 16 S3 size=50G bucket=nikratio-backup key=sb statsfile=/tmp/tmpi5o59_y_/stats_sb.txt object-size=32K & $ nbd-client -unix /tmp/tmpi5o59_y_/nbd_socket_sb /dev/nbd2 Warning: the oldstyle protocol is no longer supported. This method now uses the newstyle protocol with a default export Negotiation: ..size = 51200MB Connected /dev/nbd0 [....] nbdkit: python.10: error: write reply: NBD_CMD_WRITE: Broken pipe What's the best way to narrow down who's the culprit here (kernel vs NBD server)? Best, -Nikolaus -- GPG Fingerprint: ED31 791B 2C5C 1613 AF38 8B8A D113 FCAC 3C4E 599F »Time flies like an arrow, fruit flies like a Banana.«
On Sun, May 15, 2022 at 11:45 AM Nikolaus Rath <nikolaus@rath.org> wrote:
Hi,
I am observing some strange errors when using the Kernel's NBD driver with NBDkit.
On the kernel side, I see:
May 15 16:16:11 vostro.rath.org kernel: nbd0: detected capacity change from 0 to 104857600 May 15 16:16:11 vostro.rath.org kernel: nbd1: detected capacity change from 0 to 104857600 May 15 16:18:23 vostro.rath.org kernel: block nbd0: Possible stuck request 00000000ae5feee7: control (write@4836316160,32768B). Runtime 30 seconds May 15 16:18:25 vostro.rath.org kernel: block nbd0: Possible stuck request 000000007094eddc: control (write@5372947456,10240B). Runtime 30 seconds
The server isn't responding to the request fast enough, and you don't have a timeout set so it'll just hang until you disconnect.
May 15 16:18:27 vostro.rath.org kernel: block nbd0: Suspicious reply 89 (status 0 flags 0)
The server double replied to a request. Thanks, Josef
On May 15 2022, Josef Bacik <josef@toxicpanda.com> wrote:
On Sun, May 15, 2022 at 11:45 AM Nikolaus Rath <nikolaus@rath.org> wrote:
Hi,
I am observing some strange errors when using the Kernel's NBD driver with NBDkit.
On the kernel side, I see:
May 15 16:16:11 vostro.rath.org kernel: nbd0: detected capacity change from 0 to 104857600 May 15 16:16:11 vostro.rath.org kernel: nbd1: detected capacity change from 0 to 104857600 May 15 16:18:23 vostro.rath.org kernel: block nbd0: Possible stuck request 00000000ae5feee7: control (write@4836316160,32768B). Runtime 30 seconds May 15 16:18:25 vostro.rath.org kernel: block nbd0: Possible stuck request 000000007094eddc: control (write@5372947456,10240B). Runtime 30 seconds
The server isn't responding to the request fast enough, and you don't have a timeout set so it'll just hang until you disconnect.
Are you saying it will continue to hang even after the request has been responded to? Because NBDkit claims to have sent a response (I'm not sure if within 30 seconds, but definitely at some point). Best, -Nikolaus -- GPG Fingerprint: ED31 791B 2C5C 1613 AF38 8B8A D113 FCAC 3C4E 599F »Time flies like an arrow, fruit flies like a Banana.«
On Sun, May 15, 2022 at 04:45:11PM +0100, Nikolaus Rath wrote:
Hi,
I am observing some strange errors when using the Kernel's NBD driver with NBDkit.
On the kernel side, I see:
May 15 16:16:11 vostro.rath.org kernel: nbd0: detected capacity change from 0 to 104857600 May 15 16:16:11 vostro.rath.org kernel: nbd1: detected capacity change from 0 to 104857600 May 15 16:18:23 vostro.rath.org kernel: block nbd0: Possible stuck request 00000000ae5feee7: control (write@4836316160,32768B). Runtime 30 seconds May 15 16:18:25 vostro.rath.org kernel: block nbd0: Possible stuck request 000000007094eddc: control (write@5372947456,10240B). Runtime 30 seconds May 15 16:18:27 vostro.rath.org kernel: block nbd0: Suspicious reply 89 (status 0 flags 0) May 15 16:18:31 vostro.rath.org kernel: block nbd0: Possible stuck request 0000000075f8b9bc: control (write@8057764864,32768B). Runtime 30 seconds May 15 16:18:41 vostro.rath.org kernel: block nbd0: Possible stuck request 000000002d1b3e8b: control (write@14499979264,32768B). Runtime 30 seconds [...]
Does it really take over 30 seconds for nbdkit to respond? You might want to insert some debugging into the S3 plugin to see what stage of the request cycle is taking so long, although I'm going to guess it's the remote Amazon server itself. It seems like you can adjust this timeout using the nbd-client -t flag (it calls ioctl(NBD_SET_TIMEOUT) in the kernel). If I understand the logic correctly, the nbd timeout is currently set to 0, which causes the default socket timeout to be used. Using the -t flag overrides this. So I guess try setting it larger and see if the problem goes away. Rich.
And userspace ('zfs snapshot" in this instance) is stuck afterwards.
On the NBDkit side, there seemingly are write errors when replying back to the kernel:
$ nbdkit --unix /tmp/tmpi5o59_y_/nbd_socket_sb --foreground --filter=exitlast --filter=stats --threads 16 S3 size=50G bucket=nikratio-backup key=sb statsfile =/tmp/tmpi5o59_y_/stats_sb.txt object-size=32K & $ nbd-client -unix /tmp/tmpi5o59_y_/nbd_socket_sb /dev/nbd2 Warning: the oldstyle protocol is no longer supported. This method now uses the newstyle protocol with a default export Negotiation: ..size = 51200MB Connected /dev/nbd0 [....] nbdkit: python.10: error: write reply: NBD_CMD_WRITE: Broken pipe
What's the best way to narrow down who's the culprit here (kernel vs NBD server)?
Best, -Nikolaus
-- GPG Fingerprint: ED31 791B 2C5C 1613 AF38 8B8A D113 FCAC 3C4E 599F
»Time flies like an arrow, fruit flies like a Banana.«
_______________________________________________ Libguestfs mailing list Libguestfs@redhat.com https://listman.redhat.com/mailman/listinfo/libguestfs
-- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com libguestfs lets you edit virtual machines. Supports shell scripting, bindings from many languages. http://libguestfs.org
On May 15 2022, "Richard W.M. Jones" <rjones@redhat.com> wrote:
On Sun, May 15, 2022 at 04:45:11PM +0100, Nikolaus Rath wrote:
Hi,
I am observing some strange errors when using the Kernel's NBD driver with NBDkit.
On the kernel side, I see:
May 15 16:16:11 vostro.rath.org kernel: nbd0: detected capacity change from 0 to 104857600 May 15 16:16:11 vostro.rath.org kernel: nbd1: detected capacity change from 0 to 104857600 May 15 16:18:23 vostro.rath.org kernel: block nbd0: Possible stuck request 00000000ae5feee7: control (write@4836316160,32768B). Runtime 30 seconds May 15 16:18:25 vostro.rath.org kernel: block nbd0: Possible stuck request 000000007094eddc: control (write@5372947456,10240B). Runtime 30 seconds May 15 16:18:27 vostro.rath.org kernel: block nbd0: Suspicious reply 89 (status 0 flags 0) May 15 16:18:31 vostro.rath.org kernel: block nbd0: Possible stuck request 0000000075f8b9bc: control (write@8057764864,32768B). Runtime 30 seconds May 15 16:18:41 vostro.rath.org kernel: block nbd0: Possible stuck request 000000002d1b3e8b: control (write@14499979264,32768B). Runtime 30 seconds [...]
Does it really take over 30 seconds for nbdkit to respond? You might want to insert some debugging into the S3 plugin to see what stage of the request cycle is taking so long, although I'm going to guess it's the remote Amazon server itself.
It seems unlikely, but it is possible - especially since I'm serializing requests for debugging.
It seems like you can adjust this timeout using the nbd-client -t flag (it calls ioctl(NBD_SET_TIMEOUT) in the kernel). If I understand the logic correctly, the nbd timeout is currently set to 0, which causes the default socket timeout to be used. Using the -t flag overrides this. So I guess try setting it larger and see if the problem goes away.
Well, my concern is more about the "suspicious reply" message which - according to Josef - means that NBDkit replied twice to the same request. If that is the case, that might explain why another request seemingly remained unanswered. Do you see any way for this to happen? Best, Nikolaus -- GPG Fingerprint: ED31 791B 2C5C 1613 AF38 8B8A D113 FCAC 3C4E 599F »Time flies like an arrow, fruit flies like a Banana.«
On Sun, May 15, 2022 at 08:12:59PM +0100, Nikolaus Rath wrote:
Do you see any way for this to happen?
I think it's impossible. A more likely explanation follows. If you look at the kernel code, the NBD_CMD_INFLIGHT command flag is cleared when a command times out: https://github.com/torvalds/linux/blob/0cdd776ec92c0fec768c7079331804d3e52d4... That's the place where it would have printed the "Possible stuck request" message. Some time later, nbdkit actually replies to the message (for the first and only time) and in that code the flag is checked and found to be clear already, causing the "Suspicious reply" message to be printed: https://github.com/torvalds/linux/blob/0cdd776ec92c0fec768c7079331804d3e52d4... I'd say you need to increase the timeout and/or work out why the S3 plugin is taking so long to respond. Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com libguestfs lets you edit virtual machines. Supports shell scripting, bindings from many languages. http://libguestfs.org
On 2022/05/16 3:25, Richard W.M. Jones wrote:
On Sun, May 15, 2022 at 08:12:59PM +0100, Nikolaus Rath wrote:
Do you see any way for this to happen?
I think it's impossible. A more likely explanation follows.
If you look at the kernel code, the NBD_CMD_INFLIGHT command flag is cleared when a command times out:
https://github.com/torvalds/linux/blob/0cdd776ec92c0fec768c7079331804d3e52d4...
That's the place where it would have printed the "Possible stuck request" message.
Some time later, nbdkit actually replies to the message (for the first and only time) and in that code the flag is checked and found to be clear already, causing the "Suspicious reply" message to be printed: Hi,
You are right, can you try the following patch? https://lists.debian.org/nbd/2022/04/msg00212.html Thanks, Kuai
https://github.com/torvalds/linux/blob/0cdd776ec92c0fec768c7079331804d3e52d4...
I'd say you need to increase the timeout and/or work out why the S3 plugin is taking so long to respond.
Rich.
On May 16 2022, "yukuai (C)" <yukuai3@huawei.com> wrote:
On 2022/05/16 3:25, Richard W.M. Jones wrote:
On Sun, May 15, 2022 at 08:12:59PM +0100, Nikolaus Rath wrote:
Do you see any way for this to happen? I think it's impossible. A more likely explanation follows. If you look at the kernel code, the NBD_CMD_INFLIGHT command flag is cleared when a command times out: https://github.com/torvalds/linux/blob/0cdd776ec92c0fec768c7079331804d3e52d4... That's the place where it would have printed the "Possible stuck request" message. Some time later, nbdkit actually replies to the message (for the first and only time) and in that code the flag is checked and found to be clear already, causing the "Suspicious reply" message to be printed: Hi,
You are right, can you try the following patch?
Ah, I guess that makes sense. I thought that not specifying a timeout meant "never time out". I'll set this to a large value and report back if the problem recurs. Thanks! -Nikolaus -- GPG Fingerprint: ED31 791B 2C5C 1613 AF38 8B8A D113 FCAC 3C4E 599F »Time flies like an arrow, fruit flies like a Banana.«
participants (5)
-
Josef Bacik -
Nikolaus Rath -
Nikolaus Rath
-
Richard W.M. Jones -
yukuai (C)