<div dir="ltr"><div>Thanks a lot for detailed write-up, this helps find the bottlenecks easily.<br></div><div>On a high level, to handle this directory hierarchy i.e. lots of directories with files, we need to improve healing</div><div>algorithms. Based on the data you provided, we need to make the following enhancements:</div><div><br></div><div>1) At the moment directories are healed one at a time, but files can be healed upto 64 in parallel per replica subvolume.</div><div>So if you have nX2 or nX3 distributed subvolumes, it can heal 64n number of files in parallel.</div><div><br></div><div>I raised <a href="https://github.com/gluster/glusterfs/issues/477">https://github.com/gluster/glusterfs/issues/477</a> to track this. In the mean-while you can use the following work-around:</div><div>a) Increase background heals on the mount:</div><div>gluster volume set <volname> cluster.background-self-heal-count 256</div><div>gluster volume set <volname> cluster.cluster.heal-wait-queue-length 10000<br></div><div>find <mnt> -type d | xargs stat</div><div><br></div><div>one 'find' will trigger 10256 directories. So you may have to do this periodically until all directories are healed.<br></div><div><br></div><div><div>2) Self-heal heals a file 128KB at a time(data-self-heal-window-size). I think for your environment bumping upto MBs is better. Say 2MB i.e. 16*128KB?</div><div><br></div><div>Command to do that is:<br></div><div>gluster volume set <volname> cluster.data-self-heal-window-size 16<br></div><div><br></div></div></div><div class="gmail_extra"><br><div class="gmail_quote">On Thu, Jul 26, 2018 at 10:40 AM, Hu Bert <span dir="ltr"><<a href="mailto:revirii@googlemail.com" target="_blank">revirii@googlemail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Hi Pranith,<br>
<br>
Sry, it took a while to count the directories. I'll try to answer your<br>
questions as good as possible.<br>
<span class=""><br>
> What kind of data do you have?<br>
> How many directories in the filesystem?<br>
> On average how many files per directory?<br>
> What is the depth of your directory hierarchy on average?<br>
> What is average filesize?<br>
<br>
</span>We have mostly images (more than 95% of disk usage, 90% of file<br>
count), some text files (like css, jsp, gpx etc.) and some binaries.<br>
<br>
There are about 190.000 directories in the file system; maybe there<br>
are some more because we're hit by bug 1512371 (parallel-readdir =<br>
TRUE prevents directories listing). But the number of directories<br>
could/will rise in the future (maybe millions).<br>
<br>
files per directory: ranges from 0 to 100, on average it should be 20<br>
files per directory (well, at least in the deepest dirs, see<br>
explanation below).<br>
<br>
Average filesize: ranges from a few hundred bytes up to 30 MB, on<br>
average it should be 2-3 MB.<br>
<br>
Directory hierarchy: maximum depth as seen from within the volume is<br>
6, the average should be 3.<br>
<br>
volume name: shared<br>
mount point on clients: /data/repository/shared/<br>
below /shared/ there are 2 directories:<br>
- public/: mainly calculated images (file sizes from a few KB up to<br>
max 1 MB) and some resouces (small PNGs with a size of a few hundred<br>
bytes).<br>
- private/: mainly source images; file sizes from 50 KB up to 30MB<br>
<br>
We migrated from a NFS server (SPOF) to glusterfs and simply copied<br>
our files. The images (which have an ID) are stored in the deepest<br>
directories of the dir tree. I'll better explain it :-)<br>
<br>
directory structure for the images (i'll omit some other miscellaneous<br>
stuff, but it looks quite similar):<br>
- ID of an image has 7 or 8 digits<br>
- /shared/private/: /(first 3 digits of ID)/(next 3 digits of ID)/$ID.jpg<br>
- /shared/public/: /(first 3 digits of ID)/(next 3 digits of<br>
ID)/$ID/$misc_formats.jpg<br>
<br>
That's why we have that many (sub-)directories. Files are only stored<br>
in the lowest directory hierarchy. I hope i could make our structure<br>
at least a bit more transparent.<br>
<br>
i hope there's something we can do to raise performance a bit. thx in<br>
advance :-)<br>
<div class="HOEnZb"><div class="h5"><br>
<br>
2018-07-24 10:40 GMT+02:00 Pranith Kumar Karampuri <<a href="mailto:pkarampu@redhat.com">pkarampu@redhat.com</a>>:<br>
><br>
><br>
> On Mon, Jul 23, 2018 at 4:16 PM, Hu Bert <<a href="mailto:revirii@googlemail.com">revirii@googlemail.com</a>> wrote:<br>
>><br>
>> Well, over the weekend about 200GB were copied, so now there are<br>
>> ~400GB copied to the brick. That's far beyond a speed of 10GB per<br>
>> hour. If I copied the 1.6 TB directly, that would be done within max 2<br>
>> days. But with the self heal this will take at least 20 days minimum.<br>
>><br>
>> Why is the performance that bad? No chance of speeding this up?<br>
><br>
><br>
> What kind of data do you have?<br>
> How many directories in the filesystem?<br>
> On average how many files per directory?<br>
> What is the depth of your directory hierarchy on average?<br>
> What is average filesize?<br>
><br>
> Based on this data we can see if anything can be improved. Or if there are<br>
> some<br>
> enhancements that need to be implemented in gluster to address this kind of<br>
> data layout<br>
>><br>
>><br>
>> 2018-07-20 9:41 GMT+02:00 Hu Bert <<a href="mailto:revirii@googlemail.com">revirii@googlemail.com</a>>:<br>
>> > hmm... no one any idea?<br>
>> ><br>
>> > Additional question: the hdd on server gluster12 was changed, so far<br>
>> > ~220 GB were copied. On the other 2 servers i see a lot of entries in<br>
>> > glustershd.log, about 312.000 respectively 336.000 entries there<br>
>> > yesterday, most of them (current log output) looking like this:<br>
>> ><br>
>> > [2018-07-20 07:30:49.757595] I [MSGID: 108026]<br>
>> > [afr-self-heal-common.c:1724:<wbr>afr_log_selfheal] 0-shared-replicate-3:<br>
>> > Completed data selfheal on 0d863a62-0dd8-401c-b699-<wbr>2b642d9fd2b6.<br>
>> > sources=0 [2] sinks=1<br>
>> > [2018-07-20 07:30:49.992398] I [MSGID: 108026]<br>
>> > [afr-self-heal-metadata.c:52:_<wbr>_afr_selfheal_metadata_do]<br>
>> > 0-shared-replicate-3: performing metadata selfheal on<br>
>> > 0d863a62-0dd8-401c-b699-<wbr>2b642d9fd2b6<br>
>> > [2018-07-20 07:30:50.243551] I [MSGID: 108026]<br>
>> > [afr-self-heal-common.c:1724:<wbr>afr_log_selfheal] 0-shared-replicate-3:<br>
>> > Completed metadata selfheal on 0d863a62-0dd8-401c-b699-<wbr>2b642d9fd2b6.<br>
>> > sources=0 [2] sinks=1<br>
>> ><br>
>> > or like this:<br>
>> ><br>
>> > [2018-07-20 07:38:41.726943] I [MSGID: 108026]<br>
>> > [afr-self-heal-metadata.c:52:_<wbr>_afr_selfheal_metadata_do]<br>
>> > 0-shared-replicate-3: performing metadata selfheal on<br>
>> > 9276097a-cdac-4d12-9dc6-<wbr>04b1ea4458ba<br>
>> > [2018-07-20 07:38:41.855737] I [MSGID: 108026]<br>
>> > [afr-self-heal-common.c:1724:<wbr>afr_log_selfheal] 0-shared-replicate-3:<br>
>> > Completed metadata selfheal on 9276097a-cdac-4d12-9dc6-<wbr>04b1ea4458ba.<br>
>> > sources=[0] 2 sinks=1<br>
>> > [2018-07-20 07:38:44.755800] I [MSGID: 108026]<br>
>> > [afr-self-heal-entry.c:887:<wbr>afr_selfheal_entry_do]<br>
>> > 0-shared-replicate-3: performing entry selfheal on<br>
>> > 9276097a-cdac-4d12-9dc6-<wbr>04b1ea4458ba<br>
>> ><br>
>> > is this behaviour normal? I'd expect these messages on the server with<br>
>> > the failed brick, not on the other ones.<br>
>> ><br>
>> > 2018-07-19 8:31 GMT+02:00 Hu Bert <<a href="mailto:revirii@googlemail.com">revirii@googlemail.com</a>>:<br>
>> >> Hi there,<br>
>> >><br>
>> >> sent this mail yesterday, but somehow it didn't work? Wasn't archived,<br>
>> >> so please be indulgent it you receive this mail again :-)<br>
>> >><br>
>> >> We are currently running a replicate setup and are experiencing a<br>
>> >> quite poor performance. It got even worse when within a couple of<br>
>> >> weeks 2 bricks (disks) crashed. Maybe some general information of our<br>
>> >> setup:<br>
>> >><br>
>> >> 3 Dell PowerEdge R530 (Xeon E5-1650 v3 Hexa-Core, 64 GB DDR4, OS on<br>
>> >> separate disks); each server has 4 10TB disks -> each is a brick;<br>
>> >> replica 3 setup (see gluster volume status below). Debian stretch,<br>
>> >> kernel 4.9.0, gluster version 3.12.12. Servers and clients are<br>
>> >> connected via 10 GBit ethernet.<br>
>> >><br>
>> >> About a month ago and 2 days ago a disk died (on different servers);<br>
>> >> disk were replaced, were brought back into the volume and full self<br>
>> >> heal started. But the speed for this is quite... disappointing. Each<br>
>> >> brick has ~1.6TB of data on it (mostly the infamous small files). The<br>
>> >> full heal i started yesterday copied only ~50GB within 24 hours (48<br>
>> >> hours: about 100GB) - with<br>
>> >> this rate it would take weeks until the self heal finishes.<br>
>> >><br>
>> >> After the first heal (started on gluster13 about a month ago, took<br>
>> >> about 3 weeks) finished we had a terrible performance; CPU on one or<br>
>> >> two of the nodes (gluster11, gluster12) was up to 1200%, consumed by<br>
>> >> the brick process of the former crashed brick (bricksdd1),<br>
>> >> interestingly not on the server with the failed this, but on the other<br>
>> >> 2 ones...<br>
>> >><br>
>> >> Well... am i doing something wrong? Some options wrongly configured?<br>
>> >> Terrible setup? Anyone got an idea? Any additional information needed?<br>
>> >><br>
>> >><br>
>> >> Thx in advance :-)<br>
>> >><br>
>> >> gluster volume status<br>
>> >><br>
>> >> Volume Name: shared<br>
>> >> Type: Distributed-Replicate<br>
>> >> Volume ID: e879d208-1d8c-4089-85f3-<wbr>ef1b3aa45d36<br>
>> >> Status: Started<br>
>> >> Snapshot Count: 0<br>
>> >> Number of Bricks: 4 x 3 = 12<br>
>> >> Transport-type: tcp<br>
>> >> Bricks:<br>
>> >> Brick1: gluster11:/gluster/bricksda1/<wbr>shared<br>
>> >> Brick2: gluster12:/gluster/bricksda1/<wbr>shared<br>
>> >> Brick3: gluster13:/gluster/bricksda1/<wbr>shared<br>
>> >> Brick4: gluster11:/gluster/bricksdb1/<wbr>shared<br>
>> >> Brick5: gluster12:/gluster/bricksdb1/<wbr>shared<br>
>> >> Brick6: gluster13:/gluster/bricksdb1/<wbr>shared<br>
>> >> Brick7: gluster11:/gluster/bricksdc1/<wbr>shared<br>
>> >> Brick8: gluster12:/gluster/bricksdc1/<wbr>shared<br>
>> >> Brick9: gluster13:/gluster/bricksdc1/<wbr>shared<br>
>> >> Brick10: gluster11:/gluster/bricksdd1/<wbr>shared<br>
>> >> Brick11: gluster12:/gluster/bricksdd1_<wbr>new/shared<br>
>> >> Brick12: gluster13:/gluster/bricksdd1_<wbr>new/shared<br>
>> >> Options Reconfigured:<br>
>> >> cluster.shd-max-threads: 4<br>
>> >> performance.md-cache-timeout: 60<br>
>> >> cluster.lookup-optimize: on<br>
>> >> cluster.readdir-optimize: on<br>
>> >> performance.cache-refresh-<wbr>timeout: 4<br>
>> >> performance.parallel-readdir: on<br>
>> >> server.event-threads: 8<br>
>> >> client.event-threads: 8<br>
>> >> performance.cache-max-file-<wbr>size: 128MB<br>
>> >> performance.write-behind-<wbr>window-size: 16MB<br>
>> >> performance.io-thread-count: 64<br>
>> >> cluster.min-free-disk: 1%<br>
>> >> performance.cache-size: 24GB<br>
>> >> nfs.disable: on<br>
>> >> transport.address-family: inet<br>
>> >> performance.high-prio-threads: 32<br>
>> >> performance.normal-prio-<wbr>threads: 32<br>
>> >> performance.low-prio-threads: 32<br>
>> >> performance.least-prio-<wbr>threads: 8<br>
>> >> performance.io-cache: on<br>
>> >> server.allow-insecure: on<br>
>> >> performance.strict-o-direct: off<br>
>> >> transport.listen-backlog: 100<br>
>> >> server.outstanding-rpc-limit: 128<br>
>> ______________________________<wbr>_________________<br>
>> Gluster-users mailing list<br>
>> <a href="mailto:Gluster-users@gluster.org">Gluster-users@gluster.org</a><br>
>> <a href="https://lists.gluster.org/mailman/listinfo/gluster-users" rel="noreferrer" target="_blank">https://lists.gluster.org/<wbr>mailman/listinfo/gluster-users</a><br>
><br>
><br>
><br>
><br>
> --<br>
> Pranith<br>
</div></div></blockquote></div><br><br clear="all"><br>-- <br><div class="gmail_signature" data-smartmail="gmail_signature"><div dir="ltr">Pranith<br></div></div>
</div>