[Go-essp-tech] Replica support in Gateway 2.0
Eric Nienhouse
ejn at ucar.edu
Fri Nov 11 07:40:35 MST 2011
Hi All,
There is much to discuss here and I'd like to identify two notable
things have been raised in the thread:
* Stale replica datasets may appear in search results UI confusing users
(as Martin describes this, a "file share" issue.)
* QC/DOI applied "reference" datasets must be labeled as such and *may*
be fundamentally different than the original dataset supplied by the
data producers. This is represented as a "reference archive" issue.
We've often been using the term "replication" to describe many aspects
of this data management problem.
We're working on design to address the first issue above in the Gateway
UI. Namely directing users to the latest version of a dataset and
hiding/making difficult access to older/superseded replicated versions.
The feedback and suggestions offered in this thread are very useful
here. There is more to shine light on in this area, including end user
notifications regarding downloads of stale datasets as well as
administrative notification of new dataset versions requiring
replication. These are likely lower priority to making end user data
access clear, though they will help in scaling the system in the future.
There may be more at play regarding the QC/DOI assigned "reference
archive" datasets. I don't want to introduce undue complexity at this
point and feel we should strive to keep things as simple as we can.
However, a case can be made to identify a QC'ed dataset as a separate
"first class" entity with specific QC and DOI related attributes that
has been "derived from" the original, and perhaps less controlled,
dataset. The QC'ed dataset, though related to the original, lives a
live of its own and is replicated as such, forming the reference archive
version of this particular dataset. This would make "pre-print" data
distinctly different than a "published QC3" dataset.
I believe we need to pursue the latter issue in further detail. Bryan
and Martin have advocated this model which I believe we need in some
form. I'd like to get this more formalized quickly. It is notable that
we all agree that undue dataset copies and transfers are not desirable.
What do you think about this approach?
Thanks and regards,
-Eric
Bryan Lawrence wrote:
> ... and having replied to the first part of Jamie's message, and ignored the second ... clearly that was a mistake (on my behalf and the gateway) ... I agree that this is a major issue ... thanks Jamie!
>
> Bryan
>
>
>> Hi Jamie,
>>
>> I wasn't aware of the "stale" datasets. Well, that is a *major* Gateway
>> Bug, because it should know already about a newer version, so as it is
>> designed right now "should" point just to the original because there's
>> no replica for this.
>> [What now happens is that if you select BADC as the Gateway to download
>> to, you end up with the new version, although at DKRZ said the version
>> was the older one. - It looks like a major bug to me]
>>
>> The fact that the new version does not get immediately replicated it's
>> not a problem but a particularity of any distributed system. It's
>> impossible to have a replica of the all archive synchronized to the
>> minute. What we can do, is "hide" things that are outdated.
>> Basically:
>> 1) If the user searches for an id, that resolves to the latest version.
>> (If no version is given, we might safely presume the latest is
>> requested)
>> 2) Only replicas of that version should be displayed if available (this
>> should be already possible)
>> 3) If the user searches for a particular version, she/he should be
>> getting exactly this (feature not implemented)
>> 4) Same as 2, display only known replicas of that version.
>>
>> I don't think this is complicated. And regarding the "few years [from
>> now]" I don't think that's the case, and not only because of bandwidth
>> (which IS an issue already, as well as server load). I'm pretty sure
>> papers are getting written "as we speak", even before a DOI gets out
>> there. So these people "need" to cite something, right? The only thing
>> they can hold to is a bunch of URLs AND checksums. It's the only thing
>> you can "cite" at the moment, without this no papers could be written
>> and not only in the CMIP5 context (and thus a large AR5 share).
>> That's the reason the archives are there. You can cite what you find at
>> DKRZ, or BADC, or PCMDI. What's in there is guaranteed to remain exactly
>> where it was found for the next 10+ years, other institutions do not
>> have this commitment (nor can).
>>
>> Again 2c. Thanks,
>> Estani
>>
>> On 10.11.2011 09:14, Kettleborough, Jamie wrote:
>>
>>> Hello,
>>>
>>> I know this thread has moved on, but can we rewind just a bit. I
>>> think Nathan asked for use cases around this issue.
>>>
>>> As far as I'm aware there are two main user use cases here (there may
>>> be others obviously).
>>>
>>> 1. User wants to get data *now* (even if its only 'preprint' in
>>> Bryan's language) in a way that suites their particular needs
>>> (functional and non-functional)
>>>
>>> 2. In a few years (not sure how long - could be months) time user
>>> wants to get a copy of the data to verify or extend some previous
>>> analysis.
>>>
>>> I'm sure someone will correct me if I have this wrong, but I *think*
>>> most of the discussion so far has centered around the second of
>>> these.
>>> I think the first one needs some discusion too though doesn't it - it
>>> feels more urgent to me?
>>>
>>> I think it needs reviewing in the light of replication as when there
>>> is replication a user *may* choose to go to a particular data-node as
>>> it suites them better - they may see faster downloads because of the
>>> network route betweent them and the servers, or it might support some
>>> data service they want. Something in the system has to have the
>>> responsibility of choosing which replica to download. To be honest I
>>> think this is best left to the user. If this is the case then the
>>> user has to have a good view of *where* the data is through their
>>> interface. Nate - I think your proposed implementation would expose
>>> the information no?
>>>
>>> The other issue for replication I can think of coming out of use case
>>> 1 is versioning. Data will be revised by data providers (we have
>>> examples of this), so I think that the replication system has to keep
>>> up and the interfaces have to be able to communicate this to the
>>> user.
>>> A case I'm worried about is if a replica goes stale (sorry Estani, I
>>> *think* we have examples of these e.g.
>>>
>>> cmip5.output1.MOHC.HadGEM2-ES.historicalGHG.day.atmos.day.r1i1p1.v2011032
>>> at DKRZ is stale - we've resubmitted as v20111102. I only discovered
>>> this yesterday, and understood what *might* be happening today, I can
>>> send more examples if you need them). I think a user needs to be
>>> able
>>> to tell (without too much hard work) what really is a replica, and
>>> what is a replica of a previous version. Nate - can you cope with
>>> this situation in your proposed implementation?
>>>
>>> (Are there any issues around authorisation when it comes to replicas
>>> - would a new published version mean all replicas of previous
>>> versions
>>> are no longer 'authorisable' against, or would stale replicas be
>>> available?)
>>>
>>> I know that replication has started to happen - but is this the right
>>> thing now? Is everything in place to do this in a way that is *safe*
>>> and not going to confuse users?
>>>
>>> Jamie
>>>
>>> (I guess if you have a client that only uses the data nodes you know
>>> what node you were talking to, and see the full version information
>>> from the outset, so these aren't such big issues).
>>>
>>> ps yes I know I still have to answer some questions from Sebastien
>>> about our client.
>>>
>>>
>>>> -----Original Message-----
>>>> From: go-essp-tech-bounces at ucar.edu
>>>> [mailto:go-essp-tech-bounces at ucar.edu] On Behalf Of
>>>> martin.juckes at stfc.ac.uk
>>>> Sent: 10 November 2011 16:26
>>>> To: gonzalez at dkrz.de; go-essp-tech at ucar.edu;
>>>> esg-gateway-dev at earthsystemgrid.org
>>>> Subject: Re: [Go-essp-tech] Replica support in Gateway 2.0
>>>>
>>>> Hi Estani,
>>>>
>>>> You missed the start -- the bit which is not achievable is
>>>> publishing a replica to the same gateway used for the
>>>> original publication of that data. E.g. IPSL data published to BADC,
>>>>
>>>> Cheers,
>>>> Martin
>>>>
>>>>
>>>>>> -----Original Message-----
>>>>>> From: go-essp-tech-bounces at ucar.edu [mailto:go-essp-tech-
>>>>>> bounces at ucar.edu] On Behalf Of Estanislao Gonzalez
>>>>>> Sent: 10 November 2011 16:20
>>>>>> To: go-essp-tech at ucar.edu; esg-gateway-dev at earthsystemgrid.org
>>>>>> Subject: Re: [Go-essp-tech] Replica support in Gateway 2.0
>>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> this analogy seams perfect. Now regarding to the options
>>>>>>
>>>> we have at
>>>>
>>>>>> the
>>>>>> moment:
>>>>>> 1) How "unique" is the dataset id in a Gateway? Federation wide,
>>>>>> local gAteway or project unique?
>>>>>> 2) Depending on one, the procedure could involve "moving" the
>>>>>> published data to some other project, Gateway, Federation :-)
>>>>>>
>>>>>> I think this could be achievable:
>>>>>> 1) Data gets replicated to some Gateway (redundancy enforced)
>>>>>> 2) The originated Gateway, if it's also replicating,
>>>>>>
>>>> should replicate
>>>>
>>>>>> (just data, no publication yet) from the QC checked replica.
>>>>>> 3) The "pre-print" gets removed (which either mean move to a
>>>>>> different project, Gateway, etc or really completely
>>>>>>
>>>> delete it from
>>>>
>>>>>> the Gateway)
>>>>>> 4) The replica gets published.
>>>>>>
>>>>>> I might be omitting something, but it seams achievable right now.
>>>>>>
>>>>>> My 2c,
>>>>>> Estani
>>>>>>
>>>>>> Am 10.11.2011 07:15, schrieb Bryan Lawrence:
>>>>>>
>>>>>>> Martin has been quite vociferous (quite rightly) in
>>>>>>>
>>>> personal email
>>>>
>>>>>> to me that as far as QC goes, the dataset which gets
>>>>>>
>>>> through QC2 will
>>>>
>>>>>> *not* be the original dataset - we have no control over
>>>>>>
>>>> the original
>>>>
>>>>>> dataset's permanence and/or immutability.
>>>>>>
>>>>>>> This raises some interesting issues about the role of
>>>>>>>
>>>> ESGF ... and
>>>>
>>>>>> it's interaction with the data owner and the publication process
>>>>>> which is governed by DKRZ as the Publisher (and in the future
>>>>>> probably multiple publication processes and multiple
>>>>>>
>>>> Publishers). The
>>>>
>>>>>> correct analogy here, as I said on an earlier email today, is to
>>>>>> consider the original dataset as a preprint, of a
>>>>>>
>>>> Published dataset
>>>>
>>>>>> (at QC level 3).
>>>>>>
>>>>>>> Incidentally, this disctinction might offer us a possible
>>>>>>> (distinct)
>>>>>>>
>>>>>> future for two different types of gateways into ESGF: the
>>>>>>
>>>> Published
>>>>
>>>>>> datasets view (which makes pre-eminent the QC'd copy) and the
>>>>>> published view (which makes pre-eminenent whatever someone
>>>>>>
>>>> sticks on
>>>>
>>>>>> a data node).
>>>>>>
>>>>>>> But meanwhile, I think we can live with what you
>>>>>>>
>>>> proposed, as long
>>>>
>>>>>> as the QC status of the replicas is clearly visible - and the DOI
>>>>>> points to a landing page that somehow prioritises those versions,
>>>>>> which would be trivial if your page was organised in the same way
>>>>>> (prioritising the replicants of QC level 3, then replicants of QC
>>>>>> level 2, and then originals).
>>>>>>
>>>>>>> Cheers
>>>>>>> Bryan
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>> Hi Stephen,
>>>>>>>>
>>>>>>>> On 11/10/2011 05:23 AM, stephen.pascoe at stfc.ac.uk wrote:
>>>>>>>>
>>>>>>>>> Hi Eric,
>>>>>>>>>
>>>>>>>>> Replicas are beginning to show up in CMIP5 and this is
>>>>>>>>>
>>>> exposing
>>>>
>>>>>> some
>>>>>>
>>>>>>>>> gaps in what Gateway 1.x can do. I know you are
>>>>>>>>>
>>>> reimplementing
>>>>
>>>>>> replica
>>>>>>
>>>>>>>>> support in Gateway 2.0 so I'd like to raise these issues now.
>>>>>>>>>
>>>>>>>>> We need to be able to publish a replica to the same
>>>>>>>>>
>>>> Gateway that
>>>>
>>>>>> hosts
>>>>>>
>>>>>>>>> the original. I can't imagine this being possible with
>>>>>>>>>
>>>> Gateway
>>>>
>>>>>>>>> 1.x
>>>>>>>>>
>>>>>> since
>>>>>>
>>>>>>>>> the URL http://<GATEWAY>/dataset/<dataset-id>.html
>>>>>>>>>
>>>> only points to
>>>>
>>>>>> one
>>>>>>
>>>>>>>>> dataset on that Gateway. Either that page needs to link to
>>>>>>>>>
>>>> the
>>>>
>>>>>> original
>>>>>>
>>>>>>>>> and all replicas for that dataset or we need separate URLs
>>>>>>>>>
>>>> for
>>>>
>>>>>> each
>>>>>>
>>>>>>>>> replica/original, or both.
>>>>>>>>>
>>>>>>>> The current direction for the implementation would be
>>>>>>>>
>>>> to have a 1
>>>>
>>>>>> page
>>>>>>
>>>>>>>> for the original dataset and have that page list where
>>>>>>>>
>>>> replicas
>>>>
>>>>>>>> are located.
>>>>>>>>
>>>>>>>> If there are use cases for the other options we should get
>>>>>>>>
>>>> those
>>>>
>>>>>> identified.
>>>>>>
>>>>>>>> Thanks!
>>>>>>>> -Nate
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>> Is this part of your design for Gateway 2.0's replica
>>>>>>>>>
>>>> support?
>>>>
>>>>>>>>> Thanks,
>>>>>>>>>
>>>>>>>>> Stephen.
>>>>>>>>>
>>>>>>>>> ---
>>>>>>>>>
>>>>>>>>> Stephen Pascoe +44 (0)1235 445980
>>>>>>>>>
>>>>>>>>> Centre of Environmental Data Archival
>>>>>>>>>
>>>>>>>>> STFC Rutherford Appleton Laboratory, Harwell Oxford,
>>>>>>>>>
>>>> Didcot OX11
>>>>
>>>>>> 0QX, UK
>>>>>>
>>>>>>>>> --
>>>>>>>>> Scanned by iCritical.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> _______________________________________________
>>>>>>>>> GO-ESSP-TECH mailing list
>>>>>>>>> GO-ESSP-TECH at ucar.edu
>>>>>>>>> http://mailman.ucar.edu/mailman/listinfo/go-essp-tech
>>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> GO-ESSP-TECH mailing list
>>>>>>>> GO-ESSP-TECH at ucar.edu
>>>>>>>> http://mailman.ucar.edu/mailman/listinfo/go-essp-tech
>>>>>>>>
>>>>>>>>
>>>>>>> --
>>>>>>> Bryan Lawrence
>>>>>>> University of Reading: Professor of Weather and Climate
>>>>>>>
>>>> Computing.
>>>>
>>>>>>> National Centre for Atmospheric Science: Director of Models and
>>>>>>>
>>>>>> Data.
>>>>>>
>>>>>>> STFC: Director of the Centre for Environmental Data Archival.
>>>>>>> Ph: +44 118 3786507 or 1235 445012;
>>>>>>>
>>>> Web:home.badc.rl.ac.uk/lawrence
>>>>
>>>>>>> _______________________________________________
>>>>>>> GO-ESSP-TECH mailing list
>>>>>>> GO-ESSP-TECH at ucar.edu
>>>>>>> http://mailman.ucar.edu/mailman/listinfo/go-essp-tech
>>>>>>>
>>>>>> --
>>>>>> Estanislao Gonzalez
>>>>>>
>>>>>> Max-Planck-Institut für Meteorologie (MPI-M) Deutsches
>>>>>> Klimarechenzentrum (DKRZ) - German Climate Computing
>>>>>>
>>>> Centre Room 108
>>>>
>>>>>> - Bundesstrasse 45a, D-20146 Hamburg, Germany
>>>>>>
>>>>>> Phone: +49 (40) 46 00 94-126
>>>>>> E-Mail: gonzalez at dkrz.de
>>>>>>
>>>>>> _______________________________________________
>>>>>> GO-ESSP-TECH mailing list
>>>>>> GO-ESSP-TECH at ucar.edu
>>>>>> http://mailman.ucar.edu/mailman/listinfo/go-essp-tech
>>>>>>
>>>> --
>>>> Scanned by iCritical.
>>>> _______________________________________________
>>>> GO-ESSP-TECH mailing list
>>>> GO-ESSP-TECH at ucar.edu
>>>> http://mailman.ucar.edu/mailman/listinfo/go-essp-tech
>>>>
>>>>
>>
>
> --
> Bryan Lawrence
> University of Reading: Professor of Weather and Climate Computing.
> National Centre for Atmospheric Science: Director of Models and Data.
> STFC: Director of the Centre for Environmental Data Archival.
> Ph: +44 118 3786507 or 1235 445012; Web:home.badc.rl.ac.uk/lawrence
> _______________________________________________
> GO-ESSP-TECH mailing list
> GO-ESSP-TECH at ucar.edu
> http://mailman.ucar.edu/mailman/listinfo/go-essp-tech
>
More information about the GO-ESSP-TECH
mailing list