Re: SSD overheating during scrub in laptop

[Qu Wenruo <quwenruo.btrfs@gmx.com>]

在 2025/9/14 21:27, Martin Steigerwald 写道:
> Hi!
> 
> Just to share an experience and ask whether others have experienced this.
> 
> Already during summer it happened, that the Samsung 990 Pro 4 TB SSD in
> this ThinkPad T14 AMD Gen 5 said goodbye during scrubbing a 2 TB BTRFS
> filesystem with almost 2 TB of data in big files running at about 3,5 to
> 4,2 GiB/s. I concluded this being due to excessive heat.
> 
> However it still succeeded with the 500 GiB /home with about 300-350 GiB
> of data in it back then. I worked around it, by scrubbing two minutes,
> then canceling, waiting to cool down, resuming for two minutes until the
> 7-9 minutes of scrubbing were complete. I tried to work around with
> lowered speed, but even then the temperature slowly rose till the SSD said
> goodbye. I think I went down to 1 GiB/s of speed maximum, maybe even below
> that. But then the scrubbing takes longer so more time for SSD to heat up.
> However… I would have thought it would not heat that much with a slower
> speed. Maybe it would have worked with 50 or 100 MiB/s. But this takes
> long.
> 
> Now I had these SSD goodbyes during regular use in times of heavy I/O and
> in the end it could not even scrub that /home partition anymore in one go.
> Linux hung and only way to recover was to forcefully power off the laptop.

Can you setup netconsole and catch the dying message?

I doubt if it's really the drive dying.


> 
> I opened the laptop and removed dust with high pressure air can while
> holding the fan still so it would not generate current. And with disabled
> laptop battery.
> 
> This fixed the SSDs goodbye issue and I could even scrub that 2 TiB
> filesystem again. However, according to sensors command it still had about
> 80,8 °C composite temperature and even 100,8 °C for sensor 2 for the NVME
> SSD at "nvme-pci-0300" shortly before the end of the scrub, with critical
> for composite at 84,8 °C, but no critical set for sensor 2. That is still
> quite high. Granted, it took about 7-8 minutes of scrubbing at 3,5 to 4,2
> GiB/s in one go for it to heat up like this. But on the other hand it is
> not summer anymore and room is not as hot as in summer.

I have hit similar situation, but the symptom is very different, the 
death come silently at boot, the drive is no longer recognized by the 
BIOS thus no longer bootable, and Linux kernel from liveUSB will not 
recognize it either.

That's why I'm asking if it's really dying caused by the heat.


> 
> Anyone had similar experiences?
> 
> My solution to this will be to remove the dust inside laptop about every
> half year. However… I was a bit surprised that the NVME SSD would not
> throttle itself before saying goodbye. Or maybe it tried and it was not
> enough or to late? The laptop is a bit less than 15 months old. So I
> conclude that it takes less than a year for the cooling system to become
> quite a bit less effective due to dust. Good old ThinkPad T520 took way
> longer for that. But it is way larger on the other hand with more space to
> distribute heat.

You can refer to man page of btrfs-scrub, it provides the way to limit 
the bandwidth of scrub using cgroup or even the btrfs sysfs interface.

Thanks,
Qu

> 
> I bet it is out of scope for btrfs scrub command to monitor NVME SSD
> temperature and pause when to high. And I conclude the NVME layer of the
> kernel is not throttling to hot NVME flash either.
> 
> I checked my data filesystems, two BTRFS and one BCacheFS by scrubbing and
> fsck'ing without changing anything. They are okay.
> 
> Best,