Update data-RecoParticleFlow-PFProducer.spec #4332
Conversation
Test new data files
|
A new Pull Request was created by @mrodozov (Mircho Rodozov) for branch IB/CMSSW_10_3_X/gcc700. @cmsbuild, @smuzaffar, @gudrutis, @mrodozov can you please review it and eventually sign? Thanks. |
|
please test |
|
The tests are being triggered in jenkins. |
|
if we are going to try to make sense out of the package tags, then V15-00-01 is not the most appropriate in this case. IIUC the tests with cmssw baseline will fail. |
|
Comparison job queued. |
|
Comparison is ready Comparison Summary:
|
|
ah, the tests actually passed. |
|
I am surprised! I certainly know from my own tests that PfElectrons23Jan_BDT.weights.xml.gz and pfConversionAug0411_BDT.weights.xml.gz are used in workflow 4.53, which cmsbot tested here. However, I don't know of a workflow that uses finalID_hzz-pions_BDT.weights.xml.gz. Maybe the tests are organized in such a way that the old files are still there? |
|
Here is the list of GBRForests constructed from weights files in 4.53: https://rembserj.web.cern.ch/rembserj/scratch/4.53_gbr.out Just a curiosity: notice I run my local tests all in single thread, so we would expect every file to only appear once in the log. It's interesting to see that some electron BDTs used in the reconstruction are loaded 2 or 3 times! |
|
In fact I found out some time ago that they are also used multiple times in each event, i.e. the MVA value is calculated redundantly. It's not much overhead since it's fast, but just to keep in mind. |
|
Just to repeat: the tests with results in #4332 (comment) were done on top of the baseline with this PR which corresponds to removal of the .txt files in one format and replacement of them with .xml.gz files with different names. The tests passed. Given a separate indications that the files are used, the explanation can be that there is a separate location of the same files. @smuzaffar @mrodozov |
They shouldn't be, consult with the jenkins console output to clarify this:
meaning - it tries to get data-RecoParticleFlow-PFProducer+V15-00-01, finds out it's not available |
|
Pull request #4332 was updated. |
|
@slava77, cmsdist PR tests do not use full release build strategy. The CMSSW_SEARCH_PATH for cmsdist PR tests contains the following directories. So if a file is not found in CMSSW_BASE/data then it will be looked in to CMSSW_RELEASE_BASE/data. We can update cmsdist PRs test to drop CMSSW_RELEASE_BASE/external/slc6_amd64_gcc700/data but this will not work for data files for src. |
|
On 9/13/18 6:17 AM, Malik Shahzad Muzaffar wrote:
@slava77
cmsdist PR tests do not use full release build strategy. The
CMSSW_SEARCH_PATH for cmsdist PR tests contains the following
directories. So if a file is not found in CMSSW_BASE/data then it will
be looked in to CMSSW_RELEASE_BASE/data. We can update cmsdist PRs test
to drop CMSSW_RELEASE_BASE/external/slc6_amd64_gcc700/data but this will
not work for data files for src.
|CMSSW_BASE/src CMSSW_BASE/external/slc6_amd64_gcc700/data
CMSSW_RELEASE_BASE/src CMSSW_RELEASE_BASE/external/slc6_amd64_gcc700/data |
—
Thank you for the clarification. This explains the test results.
This also then reminds that we should be really careful with removal of
the files fro the externals because we do not test it until the
deployment in the release.
About /src dependence of data files:
when we test a CMSSW PR, is there a "poison" or other method that
prevents the build from using the RELEASE_BASE files?
IIUC, we have it for py and headers.
It would be best to have the tests look as much as possible as what the
final release will be.
|
smuzaffar
changed the base branch from
IB/CMSSW_10_3_X/gcc700
to
IB/CMSSW_10_4_X/gcc700
|
The tests are being triggered in jenkins. |
|
Comparison job queued. |
|
The tests are being triggered in jenkins. |
|
+1 The following merge commits were also included on top of IB + this PR after doing git cms-merge-topic: |
|
Comparison job queued. |
|
The tests are being triggered in jenkins. |
|
Comparison job queued. |
|
+1 |
Test new data files