Re: Unusual crash -- data rolled back ~2 weeks?

[Timothy Pearson <tpearson@raptorengineering.com>]

----- Original Message -----
> From: "Qu Wenruo" <quwenruo.btrfs@gmx.com>
> To: "Timothy Pearson" <tpearson@raptorengineering.com>
> Cc: "linux-btrfs" <linux-btrfs@vger.kernel.org>
> Sent: Sunday, November 10, 2019 7:45:14 AM
> Subject: Re: Unusual crash -- data rolled back ~2 weeks?

> On 2019/11/10 下午3:18, Timothy Pearson wrote:
>> 
>> 
>> ----- Original Message -----
>>> From: "Qu Wenruo" <quwenruo.btrfs@gmx.com>
>>> To: "Timothy Pearson" <tpearson@raptorengineering.com>
>>> Cc: "linux-btrfs" <linux-btrfs@vger.kernel.org>
>>> Sent: Sunday, November 10, 2019 6:54:55 AM
>>> Subject: Re: Unusual crash -- data rolled back ~2 weeks?
>> 
>>> On 2019/11/10 下午2:47, Timothy Pearson wrote:
>>>>
>>>>
>>>> ----- Original Message -----
>>>>> From: "Qu Wenruo" <quwenruo.btrfs@gmx.com>
>>>>> To: "Timothy Pearson" <tpearson@raptorengineering.com>, "linux-btrfs"
>>>>> <linux-btrfs@vger.kernel.org>
>>>>> Sent: Saturday, November 9, 2019 9:38:21 PM
>>>>> Subject: Re: Unusual crash -- data rolled back ~2 weeks?
>>>>
>>>>> On 2019/11/10 上午6:33, Timothy Pearson wrote:
>>>>>> We just experienced a very unusual crash on a Linux 5.3 file server using NFS to
>>>>>> serve a BTRFS filesystem.  NFS went into deadlock (D wait) with no apparent
>>>>>> underlying disk subsystem problems, and when the server was hard rebooted to
>>>>>> clear the D wait the BTRFS filesystem remounted itself in the state that it was
>>>>>> in approximately two weeks earlier (!).
>>>>>
>>>>> This means during two weeks, the btrfs is not committed.
>>>>
>>>> Is there any hope of getting the data from that interval back via btrfs-recover
>>>> or a similar tool, or does the lack of commit mean the data was stored in RAM
>>>> only and is therefore gone after the server reboot?
>>>
>>> If it's deadlock preventing new transaction to be committed, then no
>>> metadata is even written back to disk, so no way to recover metadata.
>>> Maybe you can find some data written, but without metadata it makes no
>>> sense.
>> 
>> OK, I'll just assume the data written in that window is unrecoverable at this
>> point then.
>> 
>> Would the commit deadlock affect only one btrfs filesystem or all of them on the
>> machine?  I take it there is no automatic dmesg spew on extended deadlock?
>> dmesg was completely clean at the time of the fault / reboot.
> 
> It should have some kernel message for things like process hang for over
> 120s.
> If you could recover that, it would help us to locate the cause.
> 
> Normally such deadlock should only affect the unlucky fs which meets the
> condition, not all filesystems.
> But if you're unlucky enough, it may happen to other filesystems.
> 
> Anyway, without enough info, it's really hard to say.

Agreed.  I'll check to see if there's some chance of a dmesg rotated file still hanging around from ~2 weeks prior when I get back tomorrow.

>> 
>>>>
>>>> If the latter, I'm somewhat surprised given the I/O load on the disk array in
>>>> question, but it would also offer a clue as to why it hard locked the
>>>> filesystem eventually (presumably on memory exhaustion -- the server has
>>>> something like 128GB of RAM, so it could go quite a while before hitting the
>>>> physical RAM limits).
>>>>
>>>>>
>>>>>>  There was also significant corruption of certain files (e.g. LDAP MDB and MySQL
>>>>>>  InnoDB) noted -- we restored from backup for those files, but are concerned
>>>>>>  about the status of the entire filesystem at this point.
>>>>>
>>>>> Btrfs check is needed to ensure no metadata corruption.
>>>>>
>>>>> Also, we need sysrq+w output to determine where we are deadlocking.
>>>>> Otherwise, it's really hard to find any clue from the report.
>>>>
>>>> It would have been gathered if we'd known the filesystem was in this bad state.
>>>> At the time, the priority was on restoring service and we had assumed NFS had
>>>> just wedged itself (again).  It was only after reboot and remount that the
>>>> damage slowly came to light.
>>>>
>>>> Do the described symptoms (what little we know of them at this point) line up
>>>> with the issues fixed by https://patchwork.kernel.org/patch/11141559/ ?  Right
>>>> now we're hoping that this particular issue was fixed by that series, but if
>>>> not we might consider increasing backup frequency to nightly for this
>>>> particular array and seeing if it happens again.
>>>
>>> That fix is already in v5.3, thus I don't think that's the case.
>>>
>>> Thanks,
>>> Qu
>> 
>> Looking more carefully, the server in question had been booted on 5.3-rc3
>> somehow.  It's possible that this was because earlier versions were showing
>> driver problems with the other hardware, but somehow this machine was running
>> 5.3-rc3 and the patch was created *after* rc3 release.
> 
> If that's the case, just upgrade the kernel should prevent such problem
> from happening.
> And it's a relief that we don't need to face another deadly deadlock.

Yes, absolutely, the kernel was updated immediately once this was discovered, but it was too late for the unfortunate filesystem in question.  At least we had a good DR test and backup check. :)

I'll keep an eye on this over the coming months to make sure it doesn't recur.

Thanks!