BTRFS maintenance on SSD

English Forums Feedback
1

I wrote the following script to perform the recommended maintenance on Tumbleweed with btrfs running on a SSD. This is for a couple new installs so I don’t have much history of running it.

I decided to go for the manual route since I can follow what’s going on. It runs in about 25 sec here, so not much of a problem, at least until the drive goes way past the current 6% usage. Running once a fortnight seems a good schedule, and I can review everything in a single place.

Another relevant aspect is that individual tasks won’t interfere with each other, nor with a send/receive.

Currently it handles a single btrfs file system, but I plan on supporting any additional ones.

#!/bin/bash

COLOR1='\0330;33m'
COLOR2='\0330;34m'
NC='\0330m'

printf "${COLOR1}> Disk state before maintenance${NC}

"
btrfs filesystem usage /
printf "
"

printf "${COLOR1}> Cleaning old snapshots${NC}

"
/usr/lib/snapper/systemd-helper --cleanup
printf "
"

printf "${COLOR1}> Checking data integrity${NC}

"
btrfs scrub start -BR /
printf "
"

printf "${COLOR1}> Compacting data blocks${NC}

"
btrfs balance start --verbose -dusage=5 /
printf "
"

printf "${COLOR1}> Discarding unused blocks${NC}

"
fstrim --fstab --verbose
printf "
"

printf "${COLOR1}> Disk state after maintenance${NC}

"
btrfs filesystem usage /
printf "
"

printf "${COLOR1}> S.M.A.R.T.${NC}

"
smartctl --scan | awk '{ print $1 } ' | sort | xargs -n1 -I {} sh -c 'printf "${COLOR2}>> {}${NC}

" && smartctl -a {}'
printf "
"
2

Hey Admin, Thanks for sharing this blog. I hope you will share more blogs like this. Keep blogging.You get here all the steps to fix 123.hp.setup issues.

3

Thanks for sharing the script, good to have this set of commands.

Did some code to measure how long the script takes:


START=`date +%s`

<body>

END=`date +%s`
RUNTIME=$((END-START))
printf "${COLOR1}> Script to ${RUNTIME} seconds${NC}

"

Running this script after every “zypper dup” and today it took 998 seconds.

4

Thanks for sharing!

Looking at your script and at this openSUSE SDB article https://en.opensuse.org/SDB:Disable_btrfsmaintenance, the switches and calls to balance as well as the scrub switch are different. Would you care to give a look and comment? Cheers.

5

Sure!

Balance

btrfs balance start -dusage=0 /
btrfs balance start -v -dusage=5 /
btrfs balance start -v -dusage=10 /
btrfs balance start -musage=0 /
btrfs balance start -v -musage=3 /

Remove -(d/m)usage=0, because btrfs already handles this automatically since v3.18.
Remove -musage=3, because I’ve read somewhere this shouldn’t be done (not that it would crash on you, just don’t).
Remove -dusage=10, because I’ve plenty of space atm. YMMV

Scrub

btrfs scrub start -Bd /

This is simply a matter of how I want to consume the report. I have a single device, so no need for -d, and I want to see the result as soon as it’s done, otherwise a call to btrfs scrub status is necessary, so the -R.

On another note, I saw linux 5.6 brings the mount option discard=async, which is awesome and that’s when I’m gonna remove the fstrim call from the script.

Hope this helps!

6

Great. Thanks a lot.
Cheers, and stay healthy.

7

Ah, another one, please: Your script, the btrfsmaintenance github https://github.com/kdave/btrfsmaintenance/blob/master/README.md, and the Arch Wiki mention scrub first, then balance. Only the openSUSE SDB has it the other way round. Haven’t checked the btrfs Wiki yet. Is there a rationale or recommendation as to the ordering?

8

Scrub checks if the file system can read from disk the same data that was written to it. I think it makes more sense running it before moving data around (balance). Every time a block is read from disk, its checksum is computed and compared. I understand that’s still the case if the block is to be relocated during a balance. If there’s a bit-flip in there, I don’t know how balance will deal with that, maybe it stops, or ignore the block. I’d prefer to handle it during a scrub. Good that you mention though, because I think my script will continue with the balance even in case scrub reports a checksum mismatch.