boegel commented 12 years ago

(old internal ticket 284, 324)

We should allow a way to specify a custom naming scheme for modules, instead of the default that EasyBuild uses now, i.e. name-prefix-version-toolkit-suffix.

This is badly needed for most big HPC sites.

Notes, as discussed during the 1st EasyBuild hackathon:

the module naming scheme should/could be defined by a 'standard', e.g. https://github.com/fgeorgatos/HPC-RFC/blob/master/0001/0001.md
create a symlink farm with given alternative names to modules files created by EasyBuild
- idea is to let EasyBuild keep it's current module scheme, and just provide a layer above that
- this easily allows later to change the module naming scheme without touching EasyBuild too much
- this requires that the current naming scheme that EasyBuild uses supports all possible naming schemes (e.g., see list below)
allow sysadmins to define function (or Python module) to map easyconfig parameters to this alternative namespace
- function can be defined in EasyBuild config file
try and come up with documentation that shows how to implement different namespaces
issues with dependency resolving by modules, conflicts between modules can be fixed by making modules more complex (if-else-if constructs for loading dependencies)
- (suggestion by @nudded): or maybe my 'alias' feature offered by environment modules?
nested levels of modules should also be supported
also release tool to merge from one namespace to another

The list below gives an overview of what this custom naming scheme should support (compiled by @fgeorgatos):

  ---------------------------------------------------------- --------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
  Example name                                               WHO                   Reference URL & Comments
  ---------------------------------------------------------- --------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
  netcdf/3.6.2                                               TACC                  [[http://services.tacc.utexas.edu/index.php/using-modules][]] \\ Lmod: automatic reloading of an entire module hierarchy when a single module anywhere in the hierarchy is changed.
  hpc/netcdf-3.6.3_intel-11.0.083                           HARVARD               [[http://rc.fas.harvard.edu/faq/modulelist][]] \\ very long and complete list
  misc-libs/netcdf/3.6.3_intel                              CLUMEQ                [[https://www.clumeq.ca/wiki/index.php/ModulesDisponiblesSurColosse][]] \\ nested layout
  netcdf/intel/64/4.0                                        SARA                  [[http://www.sara.nl/systems/shared/modules/][]] \\ [[http://www.sara.nl/systems/lisa/news/modules-ng][]] \\  [[http://www.sara.nl/systems/lisa/news/amd64-phase1][]] \\  [[http://www.sara.nl/systems/lisa/news/Modules-deprecated-2011-04-04][]]
[[https://www.clumeq.ca/wiki/index.php/ModulesDisponiblesSurColosse][]] \\ nested layout
  netcdf/gnu/64/4.0                                          SARA                  [[http://www.sara.nl/systems/shared/modules/][]] \\ [[http://www.sara.nl/systems/lisa/news/modules-ng][]] \\  [[http://www.sara.nl/systems/lisa/news/amd64-phase1][]] \\  [[http://www.sara.nl/systems/lisa/news/Modules-deprecated-2011-04-04][]]
  netcdf/4.0.1                                               UIBK                  [[http://www.uibk.ac.at/zid/systeme/hpc-systeme/common/tutorials/modules-howto.html][]] \\ PREFERRED_MC
  netcdf/X.Y.Z/gnu-4.1.2                                     UIBK, too             [[http://www.uibk.ac.at/th-physik/howto/hpc/modules.html][]]
  netcdf/X.Y.Z-gnu                                           CSC                   [[http://www.csc.fi/english/pages/hippu_guide/using_hippu/modules/index_html][]]
  ofed/qlogic/gcc/64/1.2.7                                   UCL                   [[http://www.ucl.ac.uk/isd/common/research-computing/services/legion-upgrade/userguide/userenvironment][]]
  netcdf/3.6.2                                               MIT                   [[http://www.darwinproject.mit.edu/wiki/index.php/Compute_cluster_hardware/software_overview][]]
  netcdf/X.Y.Z                                               MIT, too              [[http://coyote.mit.edu/mediawiki/index.php/Modules][]]
  netcdf/3.6.2                                               CAM                   [[http://www.hpc.cam.ac.uk/user/software.html][]]
  netcdf/4.0.1_nc3                                          UTORONTO              [[https://support.scinet.utoronto.ca/wiki/index.php/Software_and_Libraries][]]
  blas/?                                                     VLSCI@AU              [[http://www.vlsci.org.au/documentation/software-applications][]]
  fluent/?                                                   EPFL                  [[http://pleiades.epfl.ch/index.php?option=com_content&task=view&id=15&Itemid=30][]]
  netcdf/4.0.0.3.1jg64                                       CSCS                  [[http://user.cscs.ch/software_and_programming_environment/compilers_and_programming/rosa_cray_xt5/modules_framework/index.html][]]
  netcdf/4.1.1/pgi/10.3/64                                   TAMU                  [[http://brazos.tamu.edu/software/modules.html][]]
  sles10.1_gnu4.1.2_shared                                 UTENNESSEE            [[http://www.nics.tennessee.edu/computing-resources/kraken/software][]]
  netcdf/4.0-pgi, /4.0-gcc , /4.1.1                          UMICH                 [[http://cac.engin.umich.edu/resources/software/][]]
  netcdf-4.0.1                                               ARSC                  [[http://www.arsc.edu/support/news/systemnews/][]]
  NetCDF/4.1.3-gnu                                           Griffith University   [[http://confluence.rcs.griffith.edu.au:8080/display/GHPC/netcdf][]]
  HMMER                                                      iCER/HPCC             [[https://wiki.hpcc.msu.edu/display/Bioinfo/Module+Files][]]
  HMMER                                                      UPPNEX                [[https://www.uppnex.uu.se/installed-software][]]
  uberftp-client-2.6                                         NCSA                  [[http://www.ncsa.illinois.edu/UserInfo/Resources/Hardware/CommonDoc/module.html][]]
  netcdf                                                     LRZ                   [[http://www.lrz.de/services/software/utilities/modules/deisa_details.html][]] \\ Ref. on DEISA setup: [[http://www.lrz.de/services/software/utilities/modules/deisa_details.html][]]
  globus                                                     PRACE sites           [[http://www.prace-ri.eu/Interactive-Access-to-HPC][]]
  globus/5.0.4 & GLOBUS-5.0                                  TACC                  [[http://www.underworldproject.org/BuildRecipes/ranger.rst][]] See Teragrid below
  netcdf/4.1.2-gnu, /4.1.2-intel, /3.6.3-gnu, /3.6.3-intel   cyi/euclid@ls2        [[http://eniac.cyi.ac.cy/display/UserDoc/HPC+Baseline+Configuration]]
  netcdf/3.6.3-gcc, netcdf-4/4.0.1-gcclf                     cyi/planck   [[http://eniac.cyi.ac.cy/display/UserDoc/HPC+Baseline+Configuration]]         
  netcdf/3.6.2-intel, /3.6.2-gnu                             ba@ls2           [[http://eniac.cyi.ac.cy/display/UserDoc/HPC+Baseline+Configuration]]     
  netcdf/4.1.3-gcc(default)                                  cytera@ls2       [[http://eniac.cyi.ac.cy/display/UserDoc/HPC+Baseline+Configuration]]     
  netcdf/1.4.3-mpich2-intel-64bit                                  FZK       [[http://www.google.com]]     
  ---------------------------------------------------------- --------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

fgeorgatos commented 12 years ago

as regards this one, I can see 3 possible directions: 1) cultivate a symlink farm - yet, consider dependency resolution issues inside the modulefiles 2) use aliases - ie. dependent modules would come up with the easybuild namespace in their version strings 3) define a mapping function - this requires perfect unambiguous 1-1 mapping for proper functionality

I find that the 1st one is trivial to implement -without the deps- outside eb mechanisms; while with the deps, it will result in complex modulefiles. So, not too attractive.

The 2nd is kind of middleground to allow some pretty good compatibility with sites making the transition from legacy module namespace schemes to easybuild (which is far more rigorous)

The 3rd one is a more generic approach and should cover multiple scenarios, in effect putting the responsibility on the sysadmins to provide a consistent namespace.

The mapping function could also include retrieving the info from a "module mapping file", which allows infinite freedom and the possibility to open any can of worms you fancy...

fgeorgatos commented 11 years ago

Hi there,

after pondering a bit more on this one, here's some summary of what I think is the big picture:

The current functionality of easybuild should better remain as-is (flat under /all) - and yet extended
The provisioning of an alternate namespace should eventually allow all complicated namespace forms like:
ie. hierarchical categories + package + version + specializations

category1/category2/.../categoryN/package/version/spec1/spec2/.../specN

The current moduleclass implementation is quite interesting and useful, since it might allow (see #209) categories for an organization like this: https://www.clumeq.ca/wiki/index.php/ModulesDisponiblesSurColosse
some sites might resort to use "suffix" for lack of "/" specializations (eg 1.4.3-mpich2-intel-64bit)

Because we cannot have it all before v1.0 -due to time needed to implement the "best" solution-, one tractable approach to follow could perhaps be a "mapping" file (ie. a table of 2 columns) This becomes a python dictionary to assist with the production of the alternative namespace. (but not sure if this should rather be invoked as a separate secondary step)

Hopefully that covers the following scenarios relatively easily (read: before v1.0), and until we understand what the perfect solution is really about:

module load netcdf/intel/64/4.0 
module load netcdf/4.1.1/pgi/10.3/64 
module load netcdf/1.4.3-mpich2-intel-64bit      # doable right now with suffix?
module load misc-libs/netcdf/3.6.3_intel             # doable right now with suffix & moduleclass?
module load package/version/arch/fpaccuracy

In the ideal (long-term) situation, the full flexibility of modules should be available, with multiple categories being possible, for example:

bioinformatics/tools/MrBayes/3.2/intel/12.0/64

I have to admit I cannot easily grasp all the potential different specializations of environment modules configuration, therefor I suggest to go for now with the file-mapping approach described above, which should be the most future-proof and saves us from bothering with all the fancy (and often crippled) namespace schemes.

In essence, the proposed idea is option (3) of the previous email in this thread.

Fotis

itkovian commented 11 years ago

Considering my remark on #346, I do not think that the splitting should be done in the easyconfigs. That would mean each site with a different naming scheme would have to change all the easyconfigs they want. I hardly see the problem with offering a class that splits the module name into some format easybuild can handle and allow sites to offer a subclass that changes this to their needs.

boegel commented 11 years ago

No, not in the easyconfigs, that would be madness.

In the EasyBuild configuration file, where you also specify things like log_format and install_path, see https://github.com/hpcugent/easybuild/wiki/Configuration.

My current plan is to support a function like construct_module_name or something, that a user can define. That function would then take an EasyConfig instance, which has all the possible info, and can spit out a string/list/dictionary (not sure yet), that defines the corresponding module name. If such a function is defined, EasyBuild would use it when generating modules.

The current default would then be equivalent with something like (untested):

def construct_module_name(ec):
    return [ec.name, ec.get_installversion()]

You'd probably also need a function that does the reverse, i.e. deconstruct a given module name in parts and make sense of them.

fgeorgatos commented 10 years ago

shouldn't we close this one now that v1.8.0 is out?

boegel commented 10 years ago

I'll consider this fully fixed one #687 is handled, so let's leave it open for now.

boegel commented 9 years ago

687 is fixed, so closing this

easybuilders / easybuild-framework

support alternative naming schemes for modules #173

ie. hierarchical categories + package + version + specializations

687 is fixed, so closing this