[Gluster-devel] Glusterfs and Structured data

Sat Feb 17 05:42:45 UTC 2018

On Sat, Feb 17, 2018 at 2:58 AM, Amar Tumballi <atumball at redhat.com> wrote:

>
>
> On Thu, Feb 15, 2018 at 9:49 AM, Raghavendra Gowdappa <rgowdapp at redhat.com
> > wrote:
>
>>
>>
>> On Wed, Feb 14, 2018 at 10:31 PM, Amar Tumballi <atumball at redhat.com>
>> wrote:
>>
>>> Top posting as it is not exactly about in-consistency of perf layer.
>>>
>>> On the performance translators of Gluster, I am more interested to get
>>> work done on write-back caching layer, specially with using lease feature.
>>> Mainly because there are way too many usecases where a given directory
>>> would be used only by one client/application at the time.
>>>
>>
>> Can you explain a bit more in detail? Is it the problem of cached writes
>> reaching bricks after a lease is revoked (but application writes were done
>> when there was a valid lease)?
>>
>>
> I am not concerned on how quickly or slowly write reaches server if
> O_SYNC/O_DIRECT is not used. What I am expecting to see is, when there is
> just one client for the volume, for the below test case, no reads should
> ever reach bricks.
>
>   bash$ cd /mnt/glusterfs; cp /tmp/linux-tarball.tar.gz .; tar -xf
> linux-tarball.tar.gz; cd
>
> If above is properly achieved, totally happy. It may be already possible
> with current xlators, but if we can identify the options to achieve this,
> very happy. If not, lets fix that part of performance xlators.
>

This is not possible with just current perf xlator stack. The stack is
designed such that bricks are always considered as source of truth (for
both data and metadata). So, whatever that is cached is read a tleast once
from brick (either through pre-fetching like read-ahead etc or caching for
future access of same data). What you said is still possible by leveraging
VFS page-cache.

However, write-behind can be extended to do that, though it might be a bit
complex as currently it deals with lists of requests. But, to efficiently
implement this feature, write-behind should build the logic of "file"
(pages etc). Note that for a volume that is accessed just by a single
client, leases are of no value. But, I assume your intention was to achieve
the same goal even when multiple clients are accessing a volume, but only a
single client is active.

And you are aware that there is an issue for unified caching xlator :) [1].
One of the goals of this xlator is to achieve what you explained above.

[1] https://github.com/gluster/glusterfs/issues/218

> Regards,
> Amar
>
>
>>
>>> Regards,
>>> Amar
>>>
>>>
>>> On Wed, Feb 14, 2018 at 8:17 AM, Raghavendra G <raghavendra at gluster.com>
>>> wrote:
>>>
>>>> I've started marking "whiteboard" of bugs in this class with tag
>>>> "GLUSTERFS_METADATA_INCONSISTENCY". Please add the tag to any bugs
>>>> which you deem to fit in.
>>>>
>>>> On Fri, Feb 9, 2018 at 4:30 PM, Raghavendra Gowdappa <
>>>> rgowdapp at redhat.com> wrote:
>>>>
>>>>>
>>>>>
>>>>> ----- Original Message -----
>>>>> > From: "Pranith Kumar Karampuri" <pkarampu at redhat.com>
>>>>> > To: "Raghavendra G" <raghavendra at gluster.com>
>>>>> > Cc: "Gluster Devel" <gluster-devel at gluster.org>
>>>>> > Sent: Friday, February 9, 2018 2:30:59 PM
>>>>> > Subject: Re: [Gluster-devel] Glusterfs and Structured data
>>>>> >
>>>>> >
>>>>> >
>>>>> > On Thu, Feb 8, 2018 at 12:05 PM, Raghavendra G <
>>>>> raghavendra at gluster.com >
>>>>> > wrote:
>>>>> >
>>>>> >
>>>>> >
>>>>> >
>>>>> >
>>>>> > On Tue, Feb 6, 2018 at 8:15 PM, Vijay Bellur < vbellur at redhat.com >
>>>>> wrote:
>>>>> >
>>>>> >
>>>>> >
>>>>> >
>>>>> >
>>>>> > On Sun, Feb 4, 2018 at 3:39 AM, Raghavendra Gowdappa <
>>>>> rgowdapp at redhat.com >
>>>>> > wrote:
>>>>> >
>>>>> >
>>>>> > All,
>>>>> >
>>>>> > One of our users pointed out to the documentation that glusterfs is
>>>>> not good
>>>>> > for storing "Structured data" [1], while discussing an issue [2].
>>>>> >
>>>>> >
>>>>> > As far as I remember, the content around structured data in the
>>>>> Install Guide
>>>>> > is from a FAQ that was being circulated in Gluster, Inc. indicating
>>>>> the
>>>>> > startup's market positioning. Most of that was based on not wanting
>>>>> to get
>>>>> > into performance based comparisons of storage systems that are
>>>>> frequently
>>>>> > seen in the structured data space.
>>>>> >
>>>>> >
>>>>> > Does any of you have more context on the feasibility of storing
>>>>> "structured
>>>>> > data" on Glusterfs? Is one of the reasons for such a suggestion
>>>>> "staleness
>>>>> > of metadata" as encountered in bugs like [3]?
>>>>> >
>>>>> >
>>>>> > There are challenges that distributed storage systems face when
>>>>> exposed to
>>>>> > applications that were written for a local filesystem interface. We
>>>>> have
>>>>> > encountered problems with applications like tar [4] that are not in
>>>>> the
>>>>> > realm of "Structured data". If we look at the common theme across
>>>>> all these
>>>>> > problems, it is related to metadata & read after write consistency
>>>>> issues
>>>>> > with the default translator stack that gets exposed on the client
>>>>> side.
>>>>> > While the default stack is optimal for other scenarios, it does seem
>>>>> that a
>>>>> > category of applications needing strict metadata consistency is not
>>>>> well
>>>>> > served by that. We have observed that disabling a few performance
>>>>> > translators and tuning cache timeouts for VFS/FUSE have helped to
>>>>> overcome
>>>>> > some of them. The WIP effort on timestamp consistency across the
>>>>> translator
>>>>> > stack, patches that have been merged as a result of the bugs that you
>>>>> > mention & other fixes for outstanding issues should certainly help in
>>>>> > catering to these workloads better with the file interface.
>>>>> >
>>>>> > There are deployments that I have come across where glusterfs is
>>>>> used for
>>>>> > storing structured data. gluster-block & qemu-libgfapi overcome the
>>>>> metadata
>>>>> > consistency problem by exposing a file as a block device & by
>>>>> disabling most
>>>>> > of the performance translators in the default stack. Workloads that
>>>>> have
>>>>> > been deemed problematic with the file interface for the reasons
>>>>> alluded
>>>>> > above, function well with the block interface.
>>>>> >
>>>>> > I agree that gluster-block due to its usage of a subset of glusterfs
>>>>> fops
>>>>> > (mostly reads/writes I guess), runs into less number of consistency
>>>>> issues.
>>>>> > However, as you've mentioned we seem to disable perf xlator stack in
>>>>> our
>>>>> > tests/use-cases till now. Note that perf xlator stack is one of worst
>>>>> > offenders as far as the metadata consistency is concerned
>>>>> (relatively less
>>>>> > scenarios of data inconsistency). So, I wonder,
>>>>> > * what would be the scenario if we enable perf xlator stack for
>>>>> > gluster-block?
>>>>> > * Is performance on gluster-block satisfactory so that we don't need
>>>>> these
>>>>> > xlators?
>>>>> > - Or is it that these xlators are not useful for the workload
>>>>> usually run on
>>>>> > gluster-block (For random read/write workload, read/write caching
>>>>> xlators
>>>>> > offer less or no advantage)?
>>>>> >
>>>>> > Yes. They are not useful. Block/VM files are opened with O_DIRECT,
>>>>> so we
>>>>> > don't enable caching at any layer in glusterfs. md-cache could be
>>>>> useful for
>>>>> > serving fstat from glusterfs. But apart from that I don't see any
>>>>> other
>>>>> > xlator contributing much.
>>>>> >
>>>>> >
>>>>> >
>>>>> > - Or theoretically the workload is ought to benefit from perf
>>>>> xlators, but we
>>>>> > don't see them in our results (there are open bugs to this effect)?
>>>>> >
>>>>> > I am asking these questions to ascertain priority on fixing perf
>>>>> xlators for
>>>>> > (meta)data inconsistencies. If we offer a different solution for
>>>>> these
>>>>> > workloads, the need for fixing these issues will be less.
>>>>> >
>>>>> > My personal opinion is that both block and fs should work correctly.
>>>>> i.e.
>>>>> > caching xlators shouldn't lead to inconsistency issues.
>>>>>
>>>>> +1. That's my personal opinion too. We'll try to fix these issues.
>>>>> However, we need to qualify the fixes. It would be helpful if community can
>>>>> help here. We'll let community know when the fixes are in.
>>>>>
>>>>> > It would be better
>>>>> > if we are in a position where we choose a workload on block vs fs
>>>>> based on
>>>>> > their performance for that workload and nothing else. Block/VM
>>>>> usecases
>>>>> > change the workload of the application for glusterfs, so for small
>>>>> file
>>>>> > operations the kind of performance you see on block can never be
>>>>> achieved by
>>>>> > glusterfs with the current architecture/design.
>>>>> >
>>>>> >
>>>>> >
>>>>> >
>>>>> >
>>>>> >
>>>>> >
>>>>> > I feel that we have come a long way from the time the install guide
>>>>> was
>>>>> > written and an update for removing the "staleness of content" might
>>>>> be in
>>>>> > order there :-).
>>>>> >
>>>>> > Regards,
>>>>> > Vijay
>>>>> >
>>>>> > [4] https://bugzilla.redhat.com/show_bug.cgi?id=1058526
>>>>> >
>>>>> >
>>>>> >
>>>>> > [1] http://docs.gluster.org/en/latest/Install-Guide/Overview/
>>>>> > [2] https://bugzilla.redhat.com/show_bug.cgi?id=1512691
>>>>> > [3] https://bugzilla.redhat.com/show_bug.cgi?id=1390050
>>>>> >
>>>>> > regards,
>>>>> > Raghavendra
>>>>> > _______________________________________________
>>>>> > Gluster-devel mailing list
>>>>> > Gluster-devel at gluster.org
>>>>> > http://lists.gluster.org/mailman/listinfo/gluster-devel
>>>>> >
>>>>> >
>>>>> > _______________________________________________
>>>>> > Gluster-devel mailing list
>>>>> > Gluster-devel at gluster.org
>>>>> > http://lists.gluster.org/mailman/listinfo/gluster-devel
>>>>> >
>>>>> >
>>>>> >
>>>>> > --
>>>>> > Raghavendra G
>>>>> >
>>>>> > _______________________________________________
>>>>> > Gluster-devel mailing list
>>>>> > Gluster-devel at gluster.org
>>>>> > http://lists.gluster.org/mailman/listinfo/gluster-devel
>>>>> >
>>>>> >
>>>>> >
>>>>> > --
>>>>> > Pranith
>>>>> >
>>>>> > _______________________________________________
>>>>> > Gluster-devel mailing list
>>>>> > Gluster-devel at gluster.org
>>>>> > http://lists.gluster.org/mailman/listinfo/gluster-devel
>>>>> _______________________________________________
>>>>> Gluster-devel mailing list
>>>>> Gluster-devel at gluster.org
>>>>> http://lists.gluster.org/mailman/listinfo/gluster-devel
>>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Raghavendra G
>>>>
>>>> _______________________________________________
>>>> Gluster-devel mailing list
>>>> Gluster-devel at gluster.org
>>>> http://lists.gluster.org/mailman/listinfo/gluster-devel
>>>>
>>>
>>>
>>>
>>> --
>>> Amar Tumballi (amarts)
>>>
>>
>>
>
>
> --
> Amar Tumballi (amarts)
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.gluster.org/pipermail/gluster-devel/attachments/20180217/f17b9caf/attachment-0001.html>