smartctl --all /dev/sda make reads and no spin down

[prchal]

Hi,
I use command periodically (60s) to get temperature of disk:
smartctl --all /dev/sda | awk '/Temperature_Celsius/ {print $10}'
but hd-idle reports reads and thus won't spin down:
disk=sda command=scsi spunDown=false reads=39880915 writes=33187968 idleTime=77 idleDuration=8 spindown=0001-01-01T00:00:00 spinup=2022-03-04T16:28:55 lastIO=2022-03-04T16:32:00 disk=sda command=scsi spunDown=false reads=39880923 writes=33187968 idleTime=77 idleDuration=0 spindown=0001-01-01T00:00:00 spinup=2022-03-04T16:28:55 lastIO=2022-03-04T16:32:16 disk=sda command=scsi spunDown=false reads=39880923 writes=33187968 idleTime=77 idleDuration=8 spindown=0001-01-01T00:00:00 spinup=2022-03-04T16:28:55 lastIO=2022-03-04T16:32:16 disk=sda command=scsi spunDown=false reads=39880923 writes=33187968 idleTime=77 idleDuration=15 spindown=0001-01-01T00:00:00 spinup=2022-03-04T16:28:55 lastIO=2022-03-04T16:3
If I spun disk down manually it stays down and reading temp works too.
Is there any solution of this?
Thanks Jiri

[adelolmo]

Hi @prchal
Would you mind to try out version 1.17 and see if it makes any difference?

[prchal]

Hi adelolmo,
I'm confused, looks like it didn't help:
disk=sda command=scsi spunDown=true reads=109727262 writes=170619104 idleTime=360 idleDuration=3604 spindown=2022-08-02T09:26:03 spinup=2022-08-02T09:20:03 lastIO=2022-08-02T09:20:03
run command:
smartctl --all /dev/sda | awk '/Temperature_Celsius/ {print $10}'
and then:
sda spinup disk=sda command=scsi spunDown=false reads=109727270 writes=170619104 idleTime=360 idleDuration=0 spindown=2022-08-02T09:26:03 spinup=2022-08-02T10:20:43 lastIO=2022-08-02T10:20:43
but disk is not spinning.
BUT if I run hd-idle with long timeout and reading temp is stopped, it won't spin down:
disk=sda command=scsi spunDown=false reads=109727256 writes=170619104 idleTime=3600 idleDuration=1741 spindown=0001-01-01T00:00:00 spinup=2022-08-02T07:33:57 lastIO=2022-08-02T07:59:59 disk=sda command=scsi spunDown=false reads=109727259 writes=170619104 idleTime=3600 idleDuration=0 spindown=0001-01-01T00:00:00 spinup=2022-08-02T07:33:57 lastIO=2022-08-02T08:30:00
Something reads 3B every 1/2 hour?

[adelolmo]

Something reads 3B every 1/2 hour?

That's possible.
Could you please check #59 (comment)?

[prchal]

Not my situation, udisks2 never installed.

[Daryes]

You need to change --all to implement --nocheck=standby with smartctl

This should do the trick: --info --health --attributes --tolerance=verypermissive -nocheck=standby --format=brief
(taken from the telegraf monitoring agent => smart plugin)

Keep in mind smartctl will return an error (2) due to the drive sleeping.

[prchal]

Did not help.

[prchal]

Now I've got version 1.18 and it looks like any smartctl make reads.

[pimlie]

Can confirm what @prchal mentioned above, even when the drive is already in standby (done manually by calling hdparm -y) it still counts reads. This means that you cannot use any smart monitoring (smartmontools or munin etc) with hd-idle at the moment:

Unfortunately Im not even sure how hd-idle could workaround this behaviour, seems to be either an upstream issue or just an unavoidable caveat. Maybe a setting like ignore_read_threshold=N to forcibly put a disk to idle if there are no writes and less then N reads within $idle_time?

~# cat /sys/block/sda/stat | awk '{ print $3 }'
32594596872
~# /usr/sbin/smartctl -n standby -i /dev/sda
smartctl 7.2 2020-12-30 (local build)
Copyright (C) 2002-20, Bruce Allen, Christian Franke, www.smartmontools.org

Device is in STANDBY mode, exit(2)
~# cat /sys/block/sda/stat | awk '{ print $3 }'
32594596873
~# hdparm -C /dev/sda

/dev/sda:
 drive state is:  standby

[chrishoage]

I'm not sure if this is the exact issue here but I wanted to share what fixed it for me.

I was having this issue (smartctl causing reads on the device and not allowing spindown). I poked around the issues in the repo and came across this #38

I was not using partitions for my devices. I partitioned and formatted the partition of these devices and then hd-idle was able to detect partition reads only and it would splin down devices while smartctl would probe the device.

Now this obviously won't be an option if you have a LUKS encrypted device, but something to consider for anyone reading this and having similar issues.

[asquelt]

same here - partitionless drives are not spun down automatically, but when spinning them down (with -t) they stay in standby. something is incorrectly accounted as disk activity by hd-idle.

root@kkakdugi:~# lsb_release -a ; uname -a
No LSB modules are available.
Distributor ID:	Debian
Description:	Debian GNU/Linux 11 (bullseye)
Release:	11
Codename:	bullseye
Linux kkakdugi 5.15.60-1-pve #1 SMP PVE 5.15.60-1 (Mon, 19 Sep 2022 17:53:17 +0200) x86_64 GNU/Linux

other system with same configuration, but disks partitioned - hd-idle is correctly spinning them down.

root@kimchi:~# lsb_release -a ; uname -a 
No LSB modules are available.
Distributor ID:	Debian
Description:	Debian GNU/Linux 11 (bullseye)
Release:	11
Codename:	bullseye
Linux kimchi 5.15.60-1-pve #1 SMP PVE 5.15.60-1 (Mon, 19 Sep 2022 17:53:17 +0200) x86_64 GNU/Linux

[adelolmo]

First of all I have to apologize for not being responsive at all:
I'm sorry!

After reading this issue (and some others) many times, I'm starting to grasp what's this about.
This is my current understanding:

• Monitoring disk tools (like smartmontools) increase reads in /proc/diskstats.
• If the disk has partitions the reads go to the disk level. hd-idle relays on the partitions' activity to determine activity. So all good here.
• If the disk has no partitions the reads go also to the disk level, but hd-idle relays on the disk to determine activity. Therefore hd-idle is mislead and believes there's disk activity.

[prchal]

Hi, still not solved, but postponed to "sometimes". You might be right, I don't have a partition on affected disk. So, is there only solution to reformat disk with at least one partition?

…

On 02. 11. 23 9:16, Andoni del Olmo Elizari wrote: First of all I have to apologize for not being responsive at all: I'm sorry! After reading this issue (and some others) many times, I'm starting to grasp what's this about. This is my current understanding: * Monitoring disk tools (like smartmontools) increase reads in |/proc/diskstats|. * If the disk has partitions the reads go to the disk level. hd-idle relays on the partitions' activity to determine activity. So all good here. * If the disk has *no* partitions the reads go also to the disk level, but hd-idle relays on the disk to determine activity. Therefore hd-idle is mislead and believes there's disk activity. — Reply to this email directly, view it on GitHub <#70 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABIS2CREOZ2WSI6FJ3B75ADYCNJFFAVCNFSM5P54LJ32U5DIOJSWCZC7NNSXTN2JONZXKZKDN5WW2ZLOOQ5TCNZZGAZDMMBZGI3A>. You are receiving this because you were mentioned.Message ID: ***@***.***>

[adelolmo]

Currently I only see two options to proceed here:

1. Document that disk without partitions are not supported (known shortcoming)
2. Introduce a threshold to ignore reads

This is all mentioned here: #70 (comment)

Where option 1 would be the easiest to implement from my side, I see the value for users to implement option 2.
The option 2 requires introducing the concept of threshold.
If hd-idle puts defining the threshold on the user side, I see how questions/issues will pile here (I don't like giving support to niche scenarios).
On the other hand, if hd-idle determines the threshold... it better is a damn good algorithm, because it might have to be calculated dynamically for every case 🤷‍♂️

So, is there only solution to reformat disk with at least one partition?

@prchal Currently yes, that's the only solution if you want to keep disk monitoring tools.