experience with drobo gen2 and ext4: just don't!
314 views
Skip to first unread message
Andreas Leitgeb
Jun 17, 2016, 7:12:38 AM6/17/16
to drobo-talk
I had some apparent issues with space for deleted files not getting reclaimed, now I'm doing experiments with test data.
my drobo is filled with about 4.5 TB of disks, and I set LUN size to 1tb, giving me three luns. i formatted the luns with ext4, ext3 and vfat, and filled it with each-1gb files.
now, I've got 800 of these files in each partition, that makes about 2.4tb in total and just doesn't fit the 4.5tb of disks. how is this possible? it is possible, because the drobo happily reclaims blocks from the ext4 filesystem, leaving broken files with blocks apparently zeroed out.
it seems like ext4 is similar enough to ext3 for the drobo to attempt to reclaim blocks, but different enough that it does it file-destroyingly wrong!
tl;dr - dont ever use ext4 with a drobo that doesn't explicitly support it!
Peter Silva
Jun 17, 2016, 7:36:37 AM6/17/16
to drobo...@googlegroups.com
I wasn't aware that any drobos support ext4. It's ext3 only afaik.
yes, drobo is file system aware, and using an unsupported one will end
very badly.
To understand your space issue, need to know size of the underlying disks.
Basically subract the size of the biggest disk in your array from
total capacity, then again if dual-disk redundancy is set...
e.g. 2 x 2 Tib + 1 x 0.5 TB .... have to remove 1 2 TiB for parity,
so only have 2.5 TiB usable, matches your description, and totally
correct.
> --
> You received this message because you are subscribed to the Google Groups "drobo-talk" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to drobo-talk+...@googlegroups.com.
> To post to this group, send email to drobo...@googlegroups.com.
> Visit this group at https://groups.google.com/group/drobo-talk.
> For more options, visit https://groups.google.com/d/optout.
yes, drobo is file system aware, and using an unsupported one will end
very badly.
To understand your space issue, need to know size of the underlying disks.
Basically subract the size of the biggest disk in your array from
total capacity, then again if dual-disk redundancy is set...
e.g. 2 x 2 Tib + 1 x 0.5 TB .... have to remove 1 2 TiB for parity,
so only have 2.5 TiB usable, matches your description, and totally
correct.
> You received this message because you are subscribed to the Google Groups "drobo-talk" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to drobo-talk+...@googlegroups.com.
> To post to this group, send email to drobo...@googlegroups.com.
> Visit this group at https://groups.google.com/group/drobo-talk.
> For more options, visit https://groups.google.com/d/optout.
Peter Silva
Jun 17, 2016, 7:38:52 AM6/17/16
to drobo...@googlegroups.com
oh... I forgot to mention... need to keep in mind TB (base 10) vs. TiB
(base 2).. disks are rated in base 10, and Drobo (and operating
systems in general) report in base 2. so 2.4 TiB would likely match
2.5TB.
(base 2).. disks are rated in base 10, and Drobo (and operating
systems in general) report in base 2. so 2.4 TiB would likely match
2.5TB.
Andreas Leitgeb
Jun 17, 2016, 4:33:49 PM6/17/16
to drobo-talk
about "supporting ext4" - my old drobo certainly doesn't, but I didn't want to exclude the mere possiblity of later models supporting it...
about size: yes, you're right, the layout of my disks is 2,1,1,½, so 2.5tb wouldn't be impossible.
nevertheless, the test files on the ext4 are definitely broken (while those on ext3 and vfat are ok), and meanwhile the drobo shows 73% while df still shows 838...... used on each of the mounted luns. the sum of "used" is 2517554356 kilobytes, thus more than 2.5 decimal terabytes.
ext4 might be the sweet spot of a no-no. I'll do further experiments with even less supported filesystems, like jfs, xfs. I'd expect that it it won't even try to reclaim blocks on those and thus not accidentally destroy data. I'll eventually followup on that.
yes, I'm aware of decimal versus binary terabytes (10^12 versus 2^40). the 4.5tb are decimal tb, whereas my 800 of each-1gb testfiles are binary gb.
I read in older posts here that you no longer update drobo-utils for newer drobos - primarily because you don't own any and DRI won't donate you any to get them supported. Would you still accept patches (e.g. for marginal issues) ?
Thanks a lot for drobo-utils as it is. Since I have neither mac nor windoze, my drobo would essentially be a doorstop without drobo-utils.
about size: yes, you're right, the layout of my disks is 2,1,1,½, so 2.5tb wouldn't be impossible.
nevertheless, the test files on the ext4 are definitely broken (while those on ext3 and vfat are ok), and meanwhile the drobo shows 73% while df still shows 838...... used on each of the mounted luns. the sum of "used" is 2517554356 kilobytes, thus more than 2.5 decimal terabytes.
ext4 might be the sweet spot of a no-no. I'll do further experiments with even less supported filesystems, like jfs, xfs. I'd expect that it it won't even try to reclaim blocks on those and thus not accidentally destroy data. I'll eventually followup on that.
yes, I'm aware of decimal versus binary terabytes (10^12 versus 2^40). the 4.5tb are decimal tb, whereas my 800 of each-1gb testfiles are binary gb.
I read in older posts here that you no longer update drobo-utils for newer drobos - primarily because you don't own any and DRI won't donate you any to get them supported. Would you still accept patches (e.g. for marginal issues) ?
Thanks a lot for drobo-utils as it is. Since I have neither mac nor windoze, my drobo would essentially be a doorstop without drobo-utils.
Andreas Leitgeb
Jun 17, 2016, 7:34:48 PM6/17/16
to drobo-talk
First update: after unmounting and remounting the vfat filesystem, it now appears empty. Looking at the device with hexdump shows a big zeroed-out block after the 2nd copy of the fat extending to somewhere in the middle of one of the testfiles. the main directory seems gone. so, unless my drobo is actually broken, it seems that (like ext4) vfat isn't a recommendable choice, either.
Peter Silva
Jun 17, 2016, 8:52:37 PM6/17/16
to drobo...@googlegroups.com
about accepting patches...
Before I release a package, there is a QA process, that involves
installing the oldest firmware I can find, then testing the software with
each firmware version to confirm we can get to the latest one.
Not having any free disks lying around (though I still have one gen1
on a shelf) I can't do the QA. So I can look over your patches, I can
even accept them into mainline, but I won't make a release unless
proper QA
is possible.
Before I release a package, there is a QA process, that involves
installing the oldest firmware I can find, then testing the software with
each firmware version to confirm we can get to the latest one.
Not having any free disks lying around (though I still have one gen1
on a shelf) I can't do the QA. So I can look over your patches, I can
even accept them into mainline, but I won't make a release unless
proper QA
is possible.
Reply all
Reply to author
Forward
0 new messages