Correct, best-effort. There is no guarantee or time boxing on cross-region
replication. The best way to manage cross site replication is by tuning
your replica count to ensure you have primary copies in each region -
eventually. Possibly evaluate if you need write_affinity at all (you can
That's a pretty high rate of failure, and really needs investigation.
That's a great point, did you look into the logs of any of those jobs?
Thanks for bringing it to my attention.
I saw a few swift tests that would pop, I'll open bugs to look into those.
But the cardinality of the
OH yeah that's much better. I had found those eventually but had to dig
through all that other stuff :'(
Moving forward I think we can keep an eye on that page, open bugs for those
tests causing issue and dig in.
On Fri, Jan 24, 2014 at 11:37 AM, Sean Dague s...@dague.net
Well... What results did you get? What did you expect? What do you hope
How are you balancing your client requests across the five nodes? I'm not
sure you're going to get anywhere near 2000^H^H requests a second from a
single thread (!?) - Swift's performs best against many
At the HK summit, the topic of hot content came up and seemed to broken
into two parts.
1) developing a caching storage tier for hot content that would allow
proxies to more quickly serve small data requests with even higher rates of
2) developing a mechanism to
On Mon, Nov 11, 2013 at 11:02 AM, John Griffith john.griff...@solidfire.com
Seems to me there's a middle ground here, but honestly if you're value
add to the review process is catching grammatical or spelling errors
in comments and commit messages I'd argue that in most cases it would
409 on DELETE (object?) is a pretty specific error. That should mean that
the timestamp assigned to the delete is earlier than the timestamp of the
Most likely mean that you're getting some time-drift on your proxies (but
that assumes multi-node) or maybe that you're reusing names
I thought those tracebacks only showed up with old versions of eventlet or
and eventlet_debug = true?
In my experience that normally indicates a client disconnect on a chucked
encoding transfer request (request w/o a content-length). Do you know if
your clients are using transfer encoding
I also have limited experience with Ceph and rados bench - but it looks
like you're setting the number of worker threads to only 1? (-t 1)
I think the default is 16, and most storage distributed storage systems
designed for concurrency are going to do a bit better if you exercise more
On Mon, Sep 29, 2014 at 2:53 PM, Chmouel Boudjnah chmo...@enovance.com
eventual consistency will only affect container listing and I don't think
there is a need for container listing in that driver.
well now hold on...
if you're doing an overwrite in the face of server failures you
On Mon, Sep 29, 2014 at 4:15 PM, Clint Byrum cl...@fewbar.com wrote:
It would, however, be bad to get a 404 for something that is otherwise
present.. as that will result in an erroneous failure for the client.
That almost never happens, but is possible if all the primaries are down*,
I've never read that paper before and can not find a free copy online.
Based on the abstract it seems to be a parity based algorithm (erasure
codes) and would not be directly applicable to replication based
There is current work to enable an erasure code scheme for data
Did you find out anything more on this?
There's lots on places in Swift organized around concurrent access to
objects - so I think it's probably good that you have that 423 response;
your clients will probably see it...
When you have multiple replicas the proxy's PUT will return shortly after
You might check if the swift-recon tool has the data you're looking for.
It can report the last completed replication pass time across nodes in the
On Thu, Nov 20, 2014 at 1:28 AM, Matsuda, Kenichiro
I would like to know about a way of checking
From: Clay Gerrard [mailto:clay.gerr...@gmail.com]
I'm not sure if I agree with the giving justifications (harder to bill,
confusing to users), but I *do* think it could simplify some of the
implementation (since users can't create accounts withe the wrong storage
policy willy nilly).
I think eventually some use cases may want mixed accounts
OpenStack-dev mailing list
I think delay_denial will have to be maintained for awhile for backwards
compatibility no matter what happens.
I think existing auth middlewares can and often do reject requests outright
without forwarding them to swift (no x-auth-token?).
I think get_info and the env caching is relatively new,
1) make your runs last longer than a few seconds, a sustained load over a
longer period will average out some of jitter..
2) You don't have enough log lines. Is this a single replica cluster? How
does that work?
3) If you're benchmarking with the background daemons on, you might
add-apt-repository -y ppa:swift-core/release
^ is that a thing?
How sure are you that you're running 184.108.40.206.g3b48a71 ?
python -c 'import swift; print swift.__version__'
python -c 'import swift.common.middleware.catch_errors; print
to not included failure in recon?
On Tuesday, November 25, 2014 5:53 AM, Clay Gerrard [mailto:
On Friday, November 21, 2014 4:22 AM, Clay Gerrard [mailto:
You might check if the swift-recon tool has the data
https://github.com/cschwede/django-swiftbrowser is done by a swift core dev
You should browse:
On Mon, Jan 26, 2015 at 11:50 AM, Adam Lawson alaw...@aqorn.com wrote:
I'm researching for a web-based
On Fri, Feb 13, 2015 at 2:15 PM, David Kranz dkr...@redhat.com wrote:
Swift is different in that most interesting data is in the headers except
for GET methods, and applying the same methodology as the others does not
make sense to me. There are various ways the swift client could be changed
Can the bits that make those devices invalid and udev out of date call udev
admin --settle to just block till things are upto date and hopefully the
subseqent vg and pv scans quicker?
On Monday, March 16, 2015, John Griffith john.griffi...@gmail.com wrote:
Thought I'd reach out
On Wed, Mar 11, 2015 at 1:16 PM, Weidong Shao weidongs...@gmail.com wrote:
the url is encoded in the object hash! This somehow entangles the data
storage/validity with its account and makes it difficult to migrate the
data. I guess it is too late to debate on the design of this. Do you know
On Mon, Mar 9, 2015 at 12:27 PM, Weidong Shao weidongs...@gmail.com wrote:
I noticed swauth project is not actively maintained. In my local testing,
swauth did not work after I upgraded swift to latest.
Hrm... I think gholt would be open to patches/support, I know of a number
of deployers of
On Mon, Mar 2, 2015 at 8:07 AM, Duncan Thomas duncan.tho...@gmail.com
Why do you say auto-abandon is the wrong tool? I've no problem with the 1
week warning if somebody wants to implement it - I can see the value. A
change-set that has been ignored for X weeks is pretty much the
So, Swift doesn't enforce H302 - and our imports are sorta messy frankly -
but it doesn't really bother me, and I do rather enjoy the terseness of not
having to spell out the module name. It's not really a chore to maintain,
if you don't know where a name came from split the window (or drop a
On Thu, May 7, 2015 at 3:48 PM, Clint Byrum cl...@fewbar.com wrote:
I'm still very curious to hear if anybody has been willing to try to
make Swift work on pypy.
yeah, Alex Gaynor was helping out with it for awhile. It worked. And it
helped. A little bit.
Probably still worth looking at
On Thu, May 7, 2015 at 5:05 PM, Adam Lawson alaw...@aqorn.com wrote:
what sort of limitations have you discovered that had to do specifically
with the fact we're using Python?
Python is great. Conscious decision to optimize for developer wall time
over cpu cycles has made it a great language
Can you give an example of an Object HEAD request returning 204? I tried a
HEAD of an object with a body and also a HEAD of an object of length 0 and
I seem to get 200's...
Container's and accounts are a little more interesting story... 
how is the top pick not the author of the book of five rings 
On Mon, Jun 22, 2015 at 7:07 AM, Monty Taylor mord...@inaugust.com wrote:
Is Swift the only project that uses the ceilometermiddleware - or just the
only project that uses ceilometermiddleware that doesn't already have a
oslo.config instance handy?
FWIW There's a WIP patch that's trying to bring a *bit* of oslo.config love
to the keystone middleware for policy.json
On Fri, Jun 26, 2015 at 8:16 PM, Michael Barton m...@weirdlooking.com
What's the logical difference between having object data in memory on a
memcache server and having it in page cache on an object server?
+1 - about a syscall - i.e. not much - I think memcache does it's own heap
FWIW, as a nod to the great people we've had the privilege of working with
as Swift Core Maintainers - we've taking to promoting people who have moved
on to Core Emeritus:
On Fri, May 22, 2015 at 4:34 PM, Mike Perez thin...@gmail.com wrote:
This is long
I believe our glance friends are not the only project with some open
questions on dealing with the required dependency for optional plugin
use-case. You've made a recommendation to leverage some python tooling
functionality that I'm not familiar with. I was hoping I could probe you
So helpful! Thank you.
On Wed, Jul 29, 2015 at 7:48 AM, Doug Hellmann d...@doughellmann.com
There is some documentation in the pbr manual
feature is implemented throughout the packaging tool chain now.
I agree an error message is better than breaking for insane reasons.
But... maybe as an aside... what about not breaking?
How come the openstack ecosystem doesn't have wait for PEP 426 to be
approved and for setuptools 17.1 to be widely deployed before it can
require/depend on it? Is there no
On Wed, Jul 22, 2015 at 12:37 PM, Luse, Paul E paul.e.l...@intel.com
Wrt why the replication code seems to work if you delete just a .data
no it doesn't - https://gist.github.com/clayg/88950d77d25a441635e6
forces a listing every 10 passes for some reason. Clay?
IIRC the every
On Wed, Jul 22, 2015 at 12:24 PM, Changbin Liu email@example.com
But now I wonder: is it by design that EC does not handle an accidental
deletion of just the data file?
Well, the design goal was not do not handle the accidental deletion of
just the data file - it was make
How did you deleted one data fragment?
Like replication the EC consistency engine uses some sub directory hashing
to accelerate replication requests in a consistent system - so if you just
rm a file down in an hashdir somewhere you also need to delete the
hashes.pkl up in the part dir (or call
On Tue, Aug 25, 2015 at 8:45 AM, Kevin L. Mitchell
On Mon, 2015-08-24 at 22:53 -0700, Clay Gerrard wrote:
So, I know that hacking has H301 (one import per line) - but say maybe
you wanted to import *more* that one thing on a line (there's some
A lot of these deficiencies are drastically improved with static large
objects - and non-trivial to address (impossible?) with DLO's because of
their dynamic nature. It's unfortunate, but DLO's don't really serve your
use-case very well - and you should find a way to transition to SLO's .
So, I know that hacking has H301 (one import per line) - but say maybe you
wanted to import *more* that one thing on a line (there's some exceptions
right? sqlalchemy migrations or something?)
Anyway - I'm sure there could be a pep8 plugin rule that enforces use of
parenthesis for multi line
I've used it before in a limited capacity; and still recommend it to java
developers who want to use a language binding to connect to swift.
As best I can tell the primary maintainers for the project still work at
the same company, but there are some pending issues than't haven't received
There's probably some minimal gain in cross compatibility testing to
sticking with the status quo. The Swift API is old and stable, but I
believe there was some bug in recent history where some return value in
swiftclient changed from a iterable to a generator or something and some
... really more like 1999, but when OpenStack started back in '10 - RFC
2616 was the boss.
Since then (circa '14) we've got 7230 et. al. - a helpful attempt to
disambiguate things! Hooray progress!
But when someone recently opened this bug I got confused:
On Mon, Aug 8, 2016 at 8:31 AM, Matthew Treinish
> When we EOL a branch all of the infrastructure for running any ci against
> it goes away.
But... like... version control? I mean I'm sure it's more complicated than
that or you wouldn't have said this - but I
On Wed, Aug 10, 2016 at 7:42 AM, Ben Swartzlander
> A big source of problems IMO is that tempest doesn't have stable branches.
> We use the master branch of tempest to test stable branches of other
> projects, and tempest regularly adds new features.
On Thu, Aug 11, 2016 at 2:25 PM, Ed Leafe wrote:
> Overall this looks good, although it seems a bit odd to have
> ALL_CAPS_STRINGS to represent all:caps:strings throughout. The example you
> >>> print os_caps.HW_CPU_X86_SSE42
Just to be clear,
On Wed, Aug 10, 2016 at 10:21 AM, Matthew Treinish
> But, to keep the gate running
> involves a lot of coordination between multiple projects that are tightly
> coupled. Things like an entire extra set of job definitions in zuul, a
> branch on
> global requirements,
On Wed, Aug 10, 2016 at 10:57 AM, Matthew Treinish
> We also test every incoming
> tempest change on all the stable branches, and nothing can land unless it
> on all supported branches.
Did not know that, pretty awesome!
On Tue, Aug 9, 2016 at 11:54 AM, Hayes, Graham wrote:
> It might not make a difference to deployers / packagers who only deploy
> one project from OpenStack, but they are in the minority - having a
> known good minimum for requirements helps deployers who have multiple
On Wed, Aug 10, 2016 at 10:57 AM, Matthew Treinish
This was actually a *great* read, thanks for that link!
On Thu, Aug 11, 2016 at 7:14 AM, Erno Kuvaja wrote:
> Lets say I was ops evaluating different options as storage vendor for
> my cloud and I get told that "Here is the list of supported drivers
> for different OpenStack Cinder back ends delivered by Cinder team", I
option sounds good! The experiment would be then if it's still enough of a
stick to keep 3rd party drivers pony'd up on their commitment to the Cinder
team to consistently ship quality releases?
I'd noticed other-requirements.txt around, but figured it needed a bunch of
custom tooling to actually make it useful.
And... it's a subprocess wrapper to a handful of package management tools
(surprised to see emerge and pacmac - Kudos!) and a custom format for
describing package requirements...
On Fri, Aug 12, 2016 at 11:52 AM, Andreas Jaeger <a...@suse.com> wrote:
> On 08/12/2016 08:37 PM, Clay Gerrard wrote:
> > ... but ... it doesn't have a --install option? Do you know if that is
> > strictly out-of-scope or roadmap or ... ?
On Thu, Feb 2, 2017 at 12:50 PM, Sean Dague wrote:
> This is one of the reasons to get the wsgi stack off of eventlet and
> into a real webserver, as they handle HTTP request backups much much
To some extent I think this is generally true for *many* common
On Thu, Sep 8, 2016 at 6:13 AM, Chris Dent wrote:
> That is, it thinks of itself as an existing truth to be ratified.
Gah! YES!! Exactly this! Well said!
And this attitude keeps getting echoed again and again from the current
oligarchy TC! "We know what OpenStack
There's a note in the "for development" section  that notes the
development instructions don't include anything that puts kolla in your
sys.path or any bin scripts copied out anywhere into the PATH - i.e. it's
That seems less than ideal for a developer - did I miss a `pip install
On Sun, Sep 11, 2016 at 11:53 PM, Thierry Carrez
> FWIW I agree with Jay that the wording "a product" is definitely
> outdated and does not represent the current reality. "Product"
> presupposes a level of integration that we never achieved, and which is,
> in my
I'm interested to hear how this works out.
I thought upper-constraints was somehow supposed to work to prevent this?
Like maybe don't install a brand new shiny upstream version on the gate
infrastructure test jobs until it passes all our tests? Prevent a fire
drill? That bug was active back in
FWIW, No, this is *not* just an problem for OpenStack
^ Raymond Hettinger
Ultimately the problem is mis-aligned goals between the individual and the
project maintainers. They want to "do stuff" and get a change landed; we
want to maximize the positive results
On Thu, Sep 29, 2016 at 8:34 PM, John Griffith
> I think what's more important and critical is the future and where
> OpenStack is going over the course of the next few years.
I think this is a really important topic right now! Do you see any
On Mon, Oct 3, 2016 at 9:46 AM, Edward Leafe wrote:
> After the nominations close, the election officials will assign each
> candidate a non-identifying label, such as a random number, and those
> officials will be the only ones who know which candidate is associated with
I just re-read your announcement - and I couldn't be happier you're running
I was so surprised at the fallout from your suggestion that the TC should
actively engage in more broadcasting of important topics  to bring in
more voices from the community early!? Links to IRC logs and Gerrit
On Tue, Sep 27, 2016 at 2:52 PM, Emilien Macchi wrote:
> - Make sure it works outside Devstack.
> There is a huge gap between what is tested by Devstack gate and what
> deploy on the field. This gap tends to stretch the feedback loop between
> developers and
This was a riveting soapbox missive - and I agree with what you said -
particularly about the focus on breaking down the barriers to building out
and supporting the OpenStack contributor base.
But I don't have a good sense for how you want to apply that focus in
action on the TC? I went back and
On Wed, Nov 9, 2016 at 3:14 AM, Chris Dent wrote:
> As a community we don't want to be bound together by rules, we want
> to be enabled by processes that support making and doing things
> effectively. The things that we make and do is what binds us
On Tue, Oct 11, 2016 at 10:52 AM, Anita Kuno wrote:
> On 2016-10-11 01:40 PM, Ed Leafe wrote:
>> On Oct 11, 2016, at 12:17 PM, Anita Kuno wrote:
>> There really needs to be a period when a) we know who all the candidates
are, and b) voting
On Tue, Oct 11, 2016 at 4:02 AM, Davanum Srinivas wrote:
> Apologies for the top post.
Oh goodness, none needed my friend!
Eternal wells of gratitude for that read!
On Tue, Oct 11, 2016 at 5:40 AM, Ian Cordasco
> So, as a core developer
> of Requests, I would endorse requests-mock for this category of
Excellent; exactly the kind of feedback I was hoping to solicit. If you're
looking to add this catagory of
On Tue, Oct 11, 2016 at 9:57 AM, Thiago da Silva wrote:
> it would also be nice to have a better place for the questions/answers to
> be stored. During last week there was a ton of great discussion, but when
> it came to voting time (towards end of the week) it was
On Tue, Oct 11, 2016 at 10:44 AM, Clay Gerrard <clay.gerr...@gmail.com>
> I'm not really sure what a fixture is this context?
Answered my own question in this case!
It's one of *these* of course:
On Tue, Oct 11, 2016 at 4:24 PM, Jamie Lennox wrote:
> So I'm not going to comment too much on the quality of the library as i
> obviously think it's good
acahcpahch, no worries.
Your insights are invaluable - thanks for taking the time to connect some
dots for me -
Anyone have any experience to share positive or negative using
requests-mock? I see it's been used to replace another dependency that had
some problems in many of the OpenStack python client libraries:
Added to global requirements -> https://review.openstack.org/#/c/104067
On Tue, Dec 20, 2016 at 1:00 PM, Hongbin Lu wrote:
> $ openstack objectstore container
> $ openstack container container
> $ openstack secret container
This is the closest thing I can see that's somewhat reasonable - with the
On Tue, Dec 20, 2016 at 2:11 PM, Dean Troyer wrote:
> This is exactly how it should work. I do want to make an additional
> important but subtle point: while it looks like those are namespaced
> commands, we used 'server' not 'compute' because it is not a
I hate this stuff.
Not just pbr (tho I do have a long history of being kicked in the nuts by
pbr for no good reason I can ascertain). But when suddenly some process
OpenStack invented I've never *heard of in two years* breaks - and
overnight me and 100's of other folks have to stop what their
On Thu, Mar 2, 2017 at 10:41 AM, Paul Belanger
> In fact, the openstack-infra team does mirror a
> lot of things today
I bumped into this the other day:
... but so far haven't
On Wed, Apr 5, 2017 at 1:30 PM, Andrea Frittoli
> I just want to say thank you! to you clarkb clayg and everyone involved :)
> This is so much better!
Sean is throwing credit at me where none is due. IIRC I was both in the
room and in a
On Wed, Jun 28, 2017 at 7:50 AM, Thierry Carrez
> 2- it lets us host things that are not openstack but which we work on
> (like abandoned Python libraries or GPL-licensed things) in a familiar
Do we no longer think openstack hosted infra holds a
On Wed, Jun 28, 2017 at 7:50 AM, Thierry Carrez
> It's hard for newcomers to the OpenStack world to see what
> is a part of OpenStack and what's not.
Just an aside, this Perception problems works in our favor sometimes too.
I know in the past some BigCorp
Probably the "devices" option in the object server is misconfigured?
On my lab and production servers I configure the object-server.conf with
devices = /srv/node
And then I make sure my mounted devices appear at:
The path in the error
On Wed, May 17, 2017 at 3:47 PM, Bruno L wrote:
> I see multiple bugs in launchpad that are related to this issue.
AFAIK, only one bug for this issue is still open, and has the most recent
thoughts added from the Forum
This sounds amazing and Swift could definitely use some [automated]
assistance here. It would help if you could throw out a WIP somewhere.
First thought that comes to mind tho storyboard.o.o :\
On Fri, Jun 23, 2017 at 9:52 AM, Sean Dague wrote:
> The Nova
On Fri, May 26, 2017 at 4:06 PM, John Dickinson wrote:
> The new meeting is at a reasonable time for just about everyone, other
> than those who live in New York to San Francisco time zones.
define *un*-reasonable ;) Regardless we'll have the logs.
> I'd like to thank
Can we make this (at least) two (community?) goals?
#1 Make a thing that is not paste that is better than paste (i.e. > works,
ie >= works & is maintained)
#2 Have some people/projects "migrate" to it
If the goal is just "take over paste maintenance" that's maybe ok - but is
that an "OpenStack
On Fri, Jun 2, 2017 at 12:47 PM, John Griffith
> What I'm wondering is, even though I certainly think this is a FAR
> SUPERIOR design to what we had, I don't like having both code-paths and
> designs in the code base.
Might be useful to enumerate those?
On Fri, Jun 2, 2017 at 1:21 PM, Matt Riedemann wrote:
> I don't think the maintenance issue is the prime motivator, it's the fact
> paste is in /etc which makes it a config file and therefore an impediment
> to smooth upgrades. The more we can move into code, like default
On Thu, Jun 15, 2017 at 8:07 AM, Sean Dague wrote:
> I do kind of wonder if we returned the stackforge or
> friends-of-openstack or whatever to the github namespace when we
> mirrored if it would clear a bunch of things up for people. It would
> just need to be an extra piece
On Fri, Jun 16, 2017 at 7:19 AM, Doug Hellmann
> Excerpts from Thierry Carrez's message of 2017-06-16 11:17:30 +0200:
> > == Need for a TC meeting next Tuesday ==
> > In order to make progress on the Pike goal selection, I think a
> > dedicated IRC meeting
There's a couple of options in the object server that are related to how
object data is cached (or generally *not*)
At scale on dense nodes it becomes challenging to keep all the filesystem
metdata in the page cache, so
I like a representative democracy. It mostly means I get a say in which
other people I have to trust to think deeply about issues which effect me
and make decisions which I agree (more or less) are of benefit to the
social groups in which I participate. When I vote IRL I like to consider
On Thu, Oct 12, 2017 at 2:48 PM, Emilien Macchi wrote:
> The vision exercise was, in my opinion, one of the more exciting
> things we have done in 2017.
Yeah for sure, that was a big goings-on.
It's not an easy thing to do because of our diverses opinions, but
I ment to include reference back to (what I believe) was the original work:
On Thu, Oct 12, 2017 at 3:17 PM, Clay Gerrard <clay.gerr...@gmail.com>
> On Thu, Oct 12, 2017 at 2:48 PM, Emilien Macchi <emil...@redhat.com>
On Thu, Oct 12, 2017 at 3:20 PM, Clay Gerrard <clay.gerr...@gmail.com>
> I ment to include reference back to (what I believe) was the original work:
Sounds interesting! I'd be *very* interested to see *any* work that's been
done to make use of SPDK functionality for faster or more efficient HTTP ->
rotational media storage. Couple of thoughts...
1) I think the memcache or redis API would be a better target for SPDK
research than Swift's
I'm pretty sure that would only be possible with a code change in glance to
move the consumption of the swiftclient abstraction up a layer from the
client/connection objects to swiftclient's service objects . I'm not
sure if that'd be something that would make a lot of sense to the Image
1 - 100 of 108 matches
Mail list logo