[Go-essp-tech] Visibility of old versions, was... RE: Fwd: Re: Publishing dataset with option --update
Kettleborough, Jamie
jamie.kettleborough at metoffice.gov.uk
Tue Jan 10 05:00:54 MST 2012
Hello,
I'm not sure how to say this: but I'm not sure its just down to DOI's to determine whether a data set should always be visible. I think data needs to be visible where its sufficiently important that a user might want to download it. e.g they want to check or extend someone elses study (and I think there are other reasons). Its not clear to me that all data of this kind will have a DOI - for instance how many of the datasets referenced in papers being written now for the summer deadline of AR5 have (or will have in time) DOIs?
I know its tempting to say - any dataset referenced in a paper should have a DOI. But I think you need to be realistic about the prospects of this happening on the right timescales.
If the DOI is used as the determinent of whether data is always visible then should users be made aware of the risk they are carrying now? For instance, so they know to have local backups of data that is really important to them. (With the possible implication too that they may need to be prepared to 'reshare' this data with others.)
For what its worth my personal preference is with the BADC/DKRZ (and I'm sure others) philosophy of keeping all versions - though I realise there are costs in doing this, like getting DRSlib sufficiently bug free and getting it to work in all the contexts it needs to (hard links/soft links), getting it deployed, getting the update mechanism in place for when new bugs are found etc. If you used DRSlib doesn't Estanis use case that caused user grief become easier too - the wget scripts do not need regenerating, you should instead be able to replace the version strings in the url (though I may be assuming things about load balancing etc in saying this).
Jamie
________________________________
From: go-essp-tech-bounces at ucar.edu [mailto:go-essp-tech-bounces at ucar.edu] On Behalf Of Estanislao Gonzalez
Sent: 10 January 2012 10:21
To: Karl Taylor
Cc: Drach, Bob; go-essp-tech at ucar.edu; serguei.nikonov at noaa.gov
Subject: Re: [Go-essp-tech] Fwd: Re: Publishing dataset with option --update
Well to be honest I do agree this is a decision each institution has to make, but for us I'd prefer offering everything we have and let the systems decide what to do with this information. I.e. I've used it to generate some comments (I might have already show you this), just go here: http://ipcc-ar5.dkrz.de/dataset/cmip5.output1.NCC.NorESM1-M.sstClim.mon.land.Lmon.r1i1p1.html and click on history.
That information could be generated only because we store the metadata to the previous version.
By the way, The only way of inhibiting the user from getting an older version, if that's what it's wanted, is by either removing the files from the TDS served directory, or changing the access restriction at the Gateway. Because of a well-known TDS bug (or feature) files present at that directory and not found in any catalog are served without any restriction (AFAIK no certificate is required for this). So, normally the wget script would work even if the files where unpublished.
It really depends on the use-case... but e.g. I had to explain all this to a couple of people in the help-desk since the wget script they've downloaded wasn't working anymore (files were removed). They weren't thrilled to know they had to re issue the search again (there's no workaround for this) and they wanted to know what was changed in the new version, and there's where we can't help our users any more since we don't have that information...
I don't know what our users prefer, but I think they have more important problems to cope with at this time... if they could reliably get one version they could start worrying about others. From my perspective as a data manager, it's worth the tiny additional effort, if there's any.
Cheers,
Estani
Am 09.01.2012 20:05, schrieb Karl Taylor:
Hi Estani,
I agree that a new version number should (I'd say must) be assigned when any changes are made. However, except for DOI datasets, most groups will not want older versions to be visible or downloadable.
Do you agree?
cheers,
Karl
On 1/9/12 10:37 AM, Estanislao Gonzalez wrote:
Hi Karl,
It is indeed a good point, but I must add that we are not talking about preserving a version (although we do it here at DKRZ) but of signaling that a version has been changed. So the version is a key to find a specific dataset which changes in time.
Even before a DOI assignment I'd encourage all to create a new version every time the dataset changes in any way. Institutions have the right to preserve whatever version they want (they may even delete DOI-assigned versions, on the other hand archives can't, that's why archives are for).
But altering the dataset preserving the version just bring chaos for the users and for us at the help-desk as we have to explain why something has changed (or rather answer that we don't know why...). It means that the same key now points to a different dataset.
The only benefits I can see for preserving the same version is that publishing using the same version seems to be easier to some (for our workflow it's not, it's exactly the same) and that if only new files are added this seems to work fine for publication at both the data-node and the gateway as it's properly supported.
If anything else changes, this does not work as expected (wrong checksums, ghost files at the gateway, etc). And changing a version contents makes no sense to the user IMHO (e.g. it's as if you might sometimes get more files from a tarred file... how often should you extract it to be sure you got "all of them")
If old versions were preserved (which take almost no resources if using hardlinks), a simple comparison would tell that the only changes were the addition of some specific files.
Basically, reusing the version ends in a non-recoverable loss of information. That's why I discourage it.
My 2c,
Estani
Am 09.01.2012 17:25, schrieb Karl Taylor:
Dear all,
I do not have time to read this thoroughly, so perhaps what I'll mention here is irrelevant. There may be some miscommunication about what is meant by "version". There are two cases to consider:
1. Before a dataset has become official (i.e., assigned a DOI), a group may choose to remove all record of it from the database and publish a replacement version.
2. Alternatively, if a group wants to preserve a previous version (as is required after a DOI has been assigned), then the new version will not "replace" the previous version, but simply be added to the archive.
It is possible that different publication procedures will apply in these different cases.
best,
Karl
On 1/9/12 4:26 AM, Estanislao Gonzalez wrote:
Just to mentioned that we do the same thing. We use directly
--new-version and a map file containing all files for the new version,
but we do create hard-links to the files being reused, so they are
indeed all "new" as their paths always differ from those of previous
versions. (In any case for the publisher they are the same and thus
encode them with the nc_0 name if I recall correctly)
Thanks,
Estani
Am 09.01.2012 12:15, schrieb stephen.pascoe at stfc.ac.uk:
Hi Bob,
This "unpublish first" requirement is news to me. We've been publishing new versions without doing this for some time. Now, we have come across difficulties with a few datasets but it's generally worked.
We don't use the --update option though. Each time we publish a new version we provide a mapfile of all files in the dataset(s). I'd recommend Sergey try doing this before removing a previous version.
If you unpublish from the Gateway first you'll loose the information in the "History" tab. For instance http://cmip-gw.badc.rl.ac.uk/dataset/cmip5.output2.MOHC.HadGEM2-ES.rcp85.mon.aerosol.aero.r1i1p1.html shows 2 versions.
Stephen.
---
Stephen Pascoe +44 (0)1235 445980
Centre of Environmental Data Archival
STFC Rutherford Appleton Laboratory, Harwell Oxford, Didcot OX11 0QX, UK
-----Original Message-----
From: go-essp-tech-bounces at ucar.edu [mailto:go-essp-tech-bounces at ucar.edu] On Behalf Of Drach, Bob
Sent: 06 January 2012 20:53
To: Serguei Nikonov; Eric Nienhouse
Cc: go-essp-tech at ucar.edu
Subject: Re: [Go-essp-tech] Fwd: Re: Publishing dataset with option --update
Hi Sergey,
When updating a dataset, it's also important to unpublish it before publishing the new version. E.g, first run
esgunpublish<dataset_id>
The reason is that, when you publish to the gateway, the gateway software tries to *add* the new information to the existing dataset entry, rather that replace it.
--Bob
________________________________________
From: Serguei Nikonov [serguei.nikonov at noaa.gov]
Sent: Friday, January 06, 2012 10:45 AM
To: Eric Nienhouse
Cc: Bob Drach; go-essp-tech at ucar.edu
Subject: Re: [Go-essp-tech] Fwd: Re: Publishing dataset with option --update
Hi Eric,
thanks for you help. I have no any objections about any adopted versioning
policy. What I need is to know how to apply it. The ways I used did not work for
me. Hopefully, the reasons is bad things in thredds and database you pointed
put. I am cleaning them right now, then will see...
Just for clarification, if I need to update dataset (with changing version) I
create map file containing full set of files (old and new ones) and then use
this map file in esgpublish script with option --update, is it correct? Will it
be enough for creating dataset of new version? BTW, there is nothing about
version for option 'update' in esgpublish help.
Thanks,
Sergey
On 01/04/2012 04:27 PM, Eric Nienhouse wrote:
Hi Serguei,
Following are a few more suggestions to diagnose this publishing issue. I agree
with others on this thread that adding new files (or changing existing ones)
should always trigger a new dataset version.
It does not appear you are receiving a final "SUCCESS" or failure message when
publishing to the Gateway (with esgpublish --publish). Please try increasing
your "polling" levels in your $ESGINI file. Eg:
hessian_service_polling_delay = 10
hessian_service_polling_iterations = 500
You should see a final "SUCCESS" or "ERROR" with Java trace output at the
termination of the command.
I've reviewed the Thredds catalog for the dataset you note below:
http://esgdata.gfdl.noaa.gov/thredds/esgcet/1/cmip5.output1.NOAA-GFDL.GFDL-CM3.historical.mon.atmos.Amon.r1i1p1.v2.xml
There appear to be multiple instances of certain files within the catalog which
is a problem. The Gateway publish will fail if a particular file (URL) is
referenced multiple times with differing metadata. An example is:
*/gfdl_dataroot/NOAA-GFDL/GFDL-CM3/historical/mon/atmos/Amon/r1i1p1/v20110601/rtmt/rtmt_Amon_GFDL-CM3_historical_r1i1p1_186001-186412.nc
This file appears as two separate file versions in the Thredds catalog (one with
id ending in ".nc" and another with ".nc_0"). There should be only one reference
to this file URL in the catalog.
The previous version of the dataset in the publisher/node database may be
leading to this issue. You may need to add "--database-delete" to your
esgunpublish command to clean things up. Bob can advise on this. Note that the
original esgpublish command shown in this email thread included "--keep-version".
After publishing to the Gateway successfully, you can check the dataset details
by URL with the published dataset identifier. For example:
http://pcmdi3.llnl.gov/esgcet/dataset/cmip5.output1.NOAA-GFDL.GFDL-CM3.historical.mon.atmos.Amon.r1i1p1.html
I hope this helps.
Regards,
-Eric
Serguei Nikonov wrote:
Hi Bob,
I still can not do anything about updating datasets. The commands you
suggested executed successfully but datasets did not appear on gateway. I
tried it several times for different datasets but result is the same.
Do you have any idea what to undertake in such situation.
Here it is some details about what I tried.
I needed to add file to dataset
cmip5.output1.NOAA-GFDL.GFDL-CM3.historical.mon.atmos.Amon.r1i1p1.
As you advised I unpublished it (esgunpublish
cmip5.output1.NOAA-GFDL.GFDL-CM3.historical.mon.atmos.Amon.r1i1p1) and then
created full mapfile (with additional file) and then publised it:
esgpublish --read-files --map new_mapfile --project cmip5 --thredd --publish
As I told there were no any errors. Dataset is in database and in thredds but
not in gateway.
The second way I tried is using mapfile containing only files to update. I
needed to substitute several existing files in dataset for new ones. I created
mapfile with only files needed to substitute:
esgscan_directory --read-files --project cmip5 -o mapfile.txt
/data/CMIP5/output1/NOAA-GFDL/GFDL-ESM2M/historical/mon/ocean/Omon/r1i1p1/v20111206
and then published it with update option:
esgpublish --update --map mapfile.txt --project cmip5 --thredd --publish.
The result is the same as in a previous case - all things are fine locally but
nothing happened on gateway.
Thanks,
Sergey
-------- Original Message --------
Subject: Re: [Go-essp-tech] Publishing dataset with option --update
Date: Thu, 29 Dec 2011 11:02:05 -0500
From: Serguei Nikonov<Serguei.Nikonov at noaa.gov> <mailto:Serguei.Nikonov at noaa.gov>
Organization: GFDL
To: Drach, Bob<drach1 at llnl.gov> <mailto:drach1 at llnl.gov>
CC: Nathan Wilhelmi<wilhelmi at ucar.edu> <mailto:wilhelmi at ucar.edu> , "Ganzberger, Michael"
<Ganzberger1 at llnl.gov> <mailto:Ganzberger1 at llnl.gov> , "go-essp-tech at ucar.edu" <mailto:go-essp-tech at ucar.edu> <go-essp-tech at ucar.edu> <mailto:go-essp-tech at ucar.edu>
Hi Bob,
I tried the 1st way you suggested and it worked partially - the dataset was
created om datanode with version 2 but it was not popped up on gateway. To make
sure that it's not occasional result I repeated it with another datasets with
the same result.
Now I have 2 datasets on datanode (visible in thredds server) but they are
absent on gateway:
cmip5.output1.NOAA-GFDL.GFDL-CM3.historical.mon.atmos.Amon.r1i1p1.v2
cmip5.output1.NOAA-GFDL.GFDL-CM3.historical.mon.atmos.Amon.r2i1p1.v2.
Does it make sense to repeat esgpublish with 'publish' option?
Thanks and Happy New Year,
Sergey
On 12/21/2011 08:41 PM, Drach, Bob wrote:
Hi Sergey,
The way I would recommend adding new files to an existing dataset is as
follows:
- Unpublish the previous dataset from the gateway and thredds
% esgunpublish
cmip5.output1.NOAA-GFDL.GFDL-CM3.historical.mon.atmos.Amon.r1i1p1
- Add the new files to the existing mapfile for the dataset they are being
added to.
- Republish with the expanded mapfile:
% esgpublish --read-files --map newmap.txt --project cmip5 --thredds
--publish
The publisher will:
- not rescan existing files, only the new files
- create a new version to reflect the additional files
Alternatively you can create a mapfile with *only* the new files (Using
esgscan_directory), then republish using the --update command.
--Bob
On 12/21/11 8:40 AM, "Serguei Nikonov"<serguei.nikonov at noaa.gov> <mailto:serguei.nikonov at noaa.gov> wrote:
Hi Nate,
unfortunately this is not the only dataset I have a problem - there are at
least
5 more. Should I unpublish them locally (db, thredds) and than create new
version containing full set of files? What is the official way to update
dataset?
Thanks,
Sergey
On 12/20/2011 07:06 PM, Nathan Wilhelmi wrote:
Hi Bob/Mike,
I believe the problem is that when files were added the timestamp on the
dataset
wasn't updated.
The triple store will only harvest datasets that have files and an updated
timestamp after the last harvest.
So what likely happened is the dataset was created without files, so it
wasn't
initially harvested. Files were subsequently added, but the timestamp wasn't
updated, so it was still not a candidate for harvesting.
Can you update the date_updated timestamp for the dataset in question and
then
trigger the RDF harvesting, I believe the dataset will show up then.
Thanks!
-Nate
On 12/20/2011 11:49 AM, Serguei Nikonov wrote:
Hi Mike,
I am a member of data publishers group. I have been publishing considerable
amount of data without such kind of troubles but this one occurred only when
I
tried to add some files to existing dataset. Publishing from scratch works
fine
for me.
Thanks,
Sergey
On 12/20/2011 01:29 PM, Ganzberger, Michael wrote:
Hi Serguei,
That task is on a scheduler and will re-run every 10 minutes. If your data
does not appear after that time then perhaps there is another issue. One
issue could be that publishing to the gateway requires that you have the
role
of "Data Publisher";
"check that the account is member of the proper group and has the special
role of Data Publisher"
http://esgf.org/wiki/ESGFNode/FAQ
Mike
-----Original Message-----
From: Serguei Nikonov [mailto:serguei.nikonov at noaa.gov]
Sent: Tuesday, December 20, 2011 10:12 AM
To: Ganzberger, Michael
Cc: StИphane Senesi; Drach, Bob; go-essp-tech at ucar.edu
Subject: Re: [Go-essp-tech] Publishing dataset with option --update
Hi Mike,
thansk for suggestion but I don't have any privileges to do anything on
gateway.
I am just publishing data on GFDL data node.
Regards,
Sergey
On 12/20/2011 01:05 PM, Ganzberger, Michael wrote:
Hi Serguei,
I'd like to suggest this that may help you from
http://esgf.org/wiki/Cmip5Gateway/FAQ
"The search does not reflect the latest DB changes I've made
You have to manually trigger the 3store harvesting. Logging as root and go
to Admin->"Gateway Scheduled Tasks"->"Run tasks" and restart the job named
RDFSynchronizationJobDetail"
Mike Ganzberger
-----Original Message-----
From: go-essp-tech-bounces at ucar.edu [mailto:go-essp-tech-bounces at ucar.edu]
On Behalf Of StИphane Senesi
Sent: Tuesday, December 20, 2011 9:42 AM
To: Serguei Nikonov
Cc: Drach, Bob; go-essp-tech at ucar.edu
Subject: Re: [Go-essp-tech] Publishing dataset with option --update
Serguei
We have for some time now experienced similar problems when publishing
to the PCMDI gateway, i.e. not getting a "SUCCESS" message when
publishing . Sometimes, files are actually published (or at least
accessible through the gateway, their status being actually
"START_PUBLISHING", after esg_list_datasets report) , sometimes not. An
hypothesis is that the PCMDI Gateway load do generate the problem. We
havn't yet got a confirmation by Bob.
In contrast to your case, this happens when publishing a dataset from
scratch (I mean, not an update)
Best regards (do not expect any feeback from me since early january, yet)
S
Serguei Nikonov wrote, On 20/12/2011 18:11:
Hi Bob,
I needed to add some missed variables to existing dataset and I found in
esgpublish command an option --update. When I tried it I've got normal
message like
INFO 2011-12-20 11:21:00,893 Publishing:
cmip5.output1.NOAA-GFDL.GFDL-CM3.historical.mon.atmos.Amon.r1i1p1, parent
=
pcmdi.GFDL
INFO 2011-12-20 11:21:07,564 Result: PROCESSING
INFO 2011-12-20 11:21:11,209 Result: PROCESSING
....
but nothing happened on gateway - new variables are not there. The files
corresponding to these variables are in database and in THREDDS catalog
but
apparently were not published on gateway.
I used command line
esgpublish --update --keep-version --map<map_file> --project cmip5
--noscan
--publish.
Should map file be of some specific format to make it works in mode I
need?
Thanks,
Sergey Nikonov
GFDL
_______________________________________________
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
_______________________________________________
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
--
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
--
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
--
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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/go-essp-tech/attachments/20120110/4bdafb65/attachment-0001.html
More information about the GO-ESSP-TECH
mailing list