Do I have to balance periodically for single-ssd cache?

[testdasi]

My understanding from the help text is that balance is only required for multi-disk cache.

However, I read some other threads on the forum that (I think) suggest that balance should be run periodically in addition to trim, as a generic recommendation.

It appears to solve problem of having a false error of running out of space when there are still plenty of free space available.

 

[itimpi]

If you only have a single drive in the cache then XFS is likely to be more stable and performance file system to use.   If you use XFS then balance operations are not applicable.

[testdasi]

5 hours ago, itimpi said:

If you only have a single drive in the cache then XFS is likely to be more stable and performance file system to use.   If you use XFS then balance operations are not applicable.

 

I used btrfs, remnant from the time when I wanted RAID1 for cache. 

 

So for btrfs, do I still have to balance if only 1 disk?

[JorgeB]

It's a good idea to run a weekly balance, at least until we get to kernel 4.14, which is expected to fix this issue, it doesn't need to be a full balance, something like this:

 

btrfs balance start -dusage=75 /mnt/cache

 

[bbbilly05]

I take it we just do this manually - this command isn't baked into the scheduler > SSD TRIM function?  Not a big deal, took less than 5 minutes.

 

This is the first time I'd done balance since building my machine over six weeks ago - reported relocating 64 out of 105 chunks.

 

B.

 

[JorgeB]

35 minutes ago, bbbilly05 said:

I take it we just do this manually

 

You can use the user scripts plugin to schedule it.

[bbbilly05]

Great - thank you @johnnie.black.  That plug in makes it easy.  

 

Much appreciated.

[DZMM]

On 11/13/2017 at 10:26 AM, johnnie.black said:

 

You can use the user scripts plugin to schedule it.

I'm just running my first balance - it looks like I should have been doing frequently as so far it's balanced 24 out of 27 considered (469 in total).  I'm going to setup a script job as suggested - what commands do you recommend for a weekly job?

 

Has the only impact been the loss of space on my cache pool as a result of not balancing, or have I had a speed impact?  I'm guessing there's been a potential speed impact as I believe unRAID alternates which cache drive it reads from, so if a file only exists on one, I will have been a 'bit' slower??

 

Should I also setup a regular scrub job?

Edited November 24, 2017 by DZMM

[JorgeB]

1 minute ago, DZMM said:

what commands do you recommend for a weekly job?

 

btrfs balance start -dusage=75 /mnt/cache

 

2 minutes ago, DZMM said:

Has the only impact been the loss of space on my cache pool as a result of not balancing, or have I had a speed impact?

 

It doesn't cause loss of space, it can cause ENOSPC errors if the file system gets to be fully allocated.

It does have a speed impact since there's currently a bug where most or all allocated but free space is not trimmed.

 

4 minutes ago, DZMM said:

Should I also setup a regular scrub job?

If all is working well it's not really needed, but running say a monthly scrub can be good practice.

[DZMM]

7 minutes ago, johnnie.black said:

 

btrfs balance start -dusage=75 /mnt/cache

 

Thanks - setup weekly job.  Is this in a FAQ somewhere as I haven't come across this before?

 

8 minutes ago, johnnie.black said:

It doesn't cause loss of space, it can cause ENOSPC errors if the file system gets to be fully allocated.

 

I run mover automatically at 80%, so I don't think my cache has ever been full luckily

 

9 minutes ago, johnnie.black said:

If all is working well it's not really needed, but running say a monthly scrub can be good practice.

Will do this.  I''m guessing the code is:

 

btrfs scrub start

I've been doing this manually now and then (not balancing though!), so no errors but will do a job so can forget about this.

[JorgeB]

2 minutes ago, DZMM said:

Thanks - setup weekly job.  Is this in a FAQ somewhere as I haven't come across this before?

No, this is supposed to be fixed on kernel 4.14, if it's not I'll add a FAQ entry.

[DZMM]

Thanks - was I right with

btrfs scrub start

 

[JorgeB]

22 minutes ago, DZMM said:

Thanks - was I right with

btrfs scrub start

 

btrfs scrub start /mnt/cache

 

[DZMM]

36 minutes ago, johnnie.black said:

btrfs scrub start /mnt/cache

 

thanks

[DZMM]

ok balance over.  Should I be concerned by:

 

No balance found on '/mnt/cache'

or is that ok?

[JorgeB]

ok balance over.  Should I be concerned by:
 

No balance found on '/mnt/cache'

or is that ok?

That's normal, it just means that there's no balance running at the moment.

[DZMM]

6 minutes ago, johnnie.black said:

 

That's normal, it just means that there's no balance running at the moment.

 

Thanks - the wording could be a bit better!

[Lignumaqua]

On 11/13/2017 at 3:32 AM, johnnie.black said:

It's a good idea to run a weekly balance, at least until we get to kernel 4.14, which is expected to fix this issue, it doesn't need to be a full balance, something like this:

 

btrfs balance start -dusage=75 /mnt/cache

 

Resurrecting an old thread because I'm interested as to whether this issue has gone away now we are on kernel 4.18.10? I just ran balance and it did return a lot of space to my cache drive. Could that have been lingering since before the kernel change?  I've never run balance before, so perhaps it's built up over the last year or so.

 

Just wondering if I need to add balance to my maintenance schedule  with V6.6.1 and kernel 4.18.10?

[JorgeB]

1 minute ago, Lignumaqua said:

I just ran balance and it did return a lot of space to my cache drive. Could that have been lingering since before the kernel change?  I've never run balance before, so perhaps it's built up over the last year or so. 

If you never ran one it's good to run it on the newer kernel, and anyone coming from an earlier kernel, IIRC <4.16 should run a balance, after that there's no need to run regular balance to avoid the out of space errors, though it won't hurt to run one once in a while as regular maintenance.

[Lignumaqua]

Thank you, that’s very helpful. 

[DZMM]

I just thought - I have 3 ssds in a 2nd 'pool' using ud.  Should I be balancing these as well?

[JorgeB]

3 hours ago, DZMM said:

I just thought - I have 3 ssds in a 2nd 'pool' using ud.  Should I be balancing these as well?

Like mentioned above with newer kernels no need for regular balances.