Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

DataPlane Configuration changes #4121

Merged
merged 2 commits into from Apr 3, 2024
Merged

Conversation

eisenhauer
Copy link
Member

This PR does a couple of things:

  1. Change the RDMA transport verbosity in top-level CMake to print a list of possibly available SST DataPlanes rather than simply say "Available" or "Unconfigured"
  2. Allows specification of the CMake variable ADIOS2_SST_EXPECT_MPI_DP, which will cause CMake to terminate with an error if the tests for building the MPI dataplane show it unavailable.
  3. Allows specification of the CMake variable ADIOS2_SST_BUILD_MPI_DP, which will cause CMake to build the MPI DataPlane without checking to see if Client/Server functionality works.

Number 3 above is necessary on Frontier because MPI client/server works on compute nodes, but not on the login nodes where the user is likely to run cmake. Currently only doing this for the MPI dataplane because its use is the most conditional upon build conditions.

@guj
Copy link
Contributor

guj commented Apr 1, 2024

All three worked as expected! Thanks Greg

guj
guj previously approved these changes Apr 1, 2024
@eisenhauer eisenhauer enabled auto-merge (squash) April 1, 2024 17:15
@vicentebolea
Copy link
Collaborator

@eisenhauer the user can specify cmake -DADIOS2_HAVE_MPI_CLIENT_SERVER=ON to bypass/override the configure check. We do not need to add a new cmake variable for this. I should have documented this better.

@eisenhauer eisenhauer disabled auto-merge April 1, 2024 17:50
@guj
Copy link
Contributor

guj commented Apr 1, 2024

For me having MPI_DP in the name instead of MPI_CLIENT_SERVER is easier to guess and detect.

@vicentebolea
Copy link
Collaborator

I agree with @guj, while ADIOS2_HAVE_MPI_CLIENT_SERVER speaks of the mpi capabilities that we test, it should be named as what it enables since this is a user flag. I vote for renaming ADIOS2_HAVE_MPI_CLIENT_SERVER to ADIOS2_HAVE_MPI_DP we should also document this somewhere

pnorbert
pnorbert previously approved these changes Apr 1, 2024
@vicentebolea vicentebolea added this to the v2.10.0 milestone Apr 2, 2024
CMakeLists.txt Outdated Show resolved Hide resolved
@eisenhauer eisenhauer dismissed stale reviews from pnorbert and guj via 373e42c April 2, 2024 20:03
@eisenhauer eisenhauer merged commit 0a8cf8d into ornladios:master Apr 3, 2024
39 checks passed
@eisenhauer eisenhauer deleted the DPVerbose branch April 3, 2024 12:55
vicentebolea pushed a commit to vicentebolea/ADIOS2 that referenced this pull request Apr 3, 2024
* DataPlane Configuration changes

* Change name to include HAVE rather than BUILD
vicentebolea added a commit that referenced this pull request Apr 3, 2024
* origin/release_210: (124 commits)
  Bump version to v2.10.0
  Setting the derived variable support OFF by default
  Add the CURL function to derived variables (#4114)
  Add -f file1 [file2...] option to process adios files from a list instead of a campaign recording
  DataPlane Configuration changes (#4121)
  update doc
  Add attribute support to campaign
  Add a random string to each database name to avoid collision when running multiple applications in the same directory at the same time. Fixes issues with CI that runs ctest in parallel
  Initialize ADIOS::m_UserOption before tentaively calling ProcessUserConfig()
  - set ACA version to  0.1 - remove debug prints - add doc on Campaign Management in Advanced section - change static struct of UserOptions to class member of ADIOS class to make it work with gcc 4.8
  used size_t not int for map indexing to avoid type conversion
  fix remote server test, use the new binary name
  clang-format
  Use the name of the campaign in the cache path to avoid name collision
  rename remote_server to adios2_remote_server
  bug fix: the order of entries in bpdataset table is undefined but the campaign data reader relied on calculating the index as if it was sorted by the insertion order. Use a map instead to store the rowid and use that as index for the bpfile elements.
  Use yaml parser in campaign manager python script
  change a long variable to int64_t to avoid size confusion on windows
  do not include unistd.h
  Fix compiler error Remove extra file not needed
  ...
dmitry-ganyushin added a commit to dmitry-ganyushin/ADIOS2 that referenced this pull request Apr 16, 2024
* origin/adios-xrootd: (164 commits)
  Fixes for FreeBSD, including upstream (ornladios#4138)
  Bump version to v2.10.0
  Setting the derived variable support OFF by default
  Add the CURL function to derived variables (ornladios#4114)
  Add -f file1 [file2...] option to process adios files from a list instead of a campaign recording
  DataPlane Configuration changes (ornladios#4121)
  update doc
  Add attribute support to campaign
  Add a random string to each database name to avoid collision when running multiple applications in the same directory at the same time. Fixes issues with CI that runs ctest in parallel
  Initialize ADIOS::m_UserOption before tentaively calling ProcessUserConfig()
  - set ACA version to  0.1 - remove debug prints - add doc on Campaign Management in Advanced section - change static struct of UserOptions to class member of ADIOS class to make it work with gcc 4.8
  used size_t not int for map indexing to avoid type conversion
  fix remote server test, use the new binary name
  clang-format
  Use the name of the campaign in the cache path to avoid name collision
  rename remote_server to adios2_remote_server
  bug fix: the order of entries in bpdataset table is undefined but the campaign data reader relied on calculating the index as if it was sorted by the insertion order. Use a map instead to store the rowid and use that as index for the bpfile elements.
  Use yaml parser in campaign manager python script
  change a long variable to int64_t to avoid size confusion on windows
  do not include unistd.h
  Fix compiler error Remove extra file not needed
  Add support for user options in ~/.config/adios2/adios2.yaml Currently supported options:
  cmake: add sqlite3 and zlib dep in adios2 cmake pkg
  Different names for MPI and Serial tests (ornladios#4118)
  EVpath upstream to make NO_RDMA more robust (ornladios#4116)
  Warnings (ornladios#4113)
  Don't use assert() in tests (ornladios#4108)
  - Only add campaign store to file name if that is not absolute path - list command supports second argument as path
  flake8 fixes
  Update campaign manager script to handle config file, time in nanosecond format, and avoiding conflict when updating database
  dill 2024-03-12 (ebc98c4d) (ornladios#4091)
  Don't run derived test in MPI mode, it's not written for that (ornladios#4104)
  Fix static blosc2 build (ornladios#4093)
  ci: add ccache job summary (ornladios#4101)
  Fix typo in fortran.rst (ornladios#4102)
  WIP: Make Fortran tests fail with a non-zero exit code (ornladios#4097)
  Bison 3.8 Parser (ornladios#4062)
  Do not create adios-campaign/ unless there is something to record
  Add setup for Aurora (load adios2 as e4s package)
  Completely hide derived variables in C API if not enabled. Print warning inside Fortran F2C function.
  adios2_define_derived_variable C/Fortran API. C is compiled conditionally, like the C++ API. The Fortran function is always available, it will print a WARNING if derived variable is not supported. Added Fortran test for magnitude().
  Fix links to tutorial materials (ornladios#4086)
  BlockIndex.Evaluate() made branch based on BP5 vs BP4. To support CampaignReader engine, decision is made based on whether MinBlocksInfo is supported by engine.
  Update documentation for 2.10 changes to the GPU-backend (ornladios#4083)
  Add test for single string attribute vs string array attribute with a single element
  - Python: fix for scalar reading. If a global value has 1 step (i.e. always in streaming), read returns a 0-dim numpy array (single value). If the variable has multiple steps (only in ReadRandomAccess mode), read returns a 1-dim numpy array even if the step selection is a single step. This way, read of a certain variable always results in the same type of array no matter the number of steps selected. - Python: fix for string attributes: return a string, not a list of one element which is a string, to be consistent with string global values and with other APIs.
  format more
  format
  Python: add the same treatment to attributes as to variables before: return scalars (0-dim ndarray) for single value attributes.
  Raise an exception if remote open fails (ornladios#4069)
  Fortran bindings for memory space related functions (ornladios#4077)
  consolidate (ornladios#4078)
  Making the Detect memory space available regardless of the backend used
  Testing code for the C bindings with memory space API
  Adding c bindings for getting the shape of a variable based on a memory space
  Adding c bindings for setting and getting the memory space
  Small typo fixes
  - Restructure python API doc in separate main topic, add working example to it. - What's new for 2.10 - Usage on DOE machines
  Fix Reord to use MinBlocksInfo where appropriate (ornladios#4071)
  Using the correct flag to detect the CUDA backend in ZFP
  clang-format fix
  fixed warning
  added support to read back from H5T_STRING VARIABLES it turns out strings written out through h5py are all variable strings
  format
  fixes to still be able to build with gcc 4.8.2. Needed for OLCF DTN nodes
  Add minmax and shape functions to CampaignReader, so that per-block info is complete when listing campaign archives
  do not flush io and adios in read mode. BP5 reader does not like it.
  Campaign engine is recognized by file extension.
  WIP. Changed the name of the campaign config file.
  Added reading of configuration parametyers from ./config/adios2
  Use GetEstimatedSize in encryption operator plugin
  Allow plugin operators to take advantage of the estimated size API
  Setting the derived variable support OFF by default
  Add the CURL function to derived variables (ornladios#4114)
  Add -f file1 [file2...] option to process adios files from a list instead of a campaign recording
  DataPlane Configuration changes (ornladios#4121)
  update doc
  Add attribute support to campaign
  Add a random string to each database name to avoid collision when running multiple applications in the same directory at the same time. Fixes issues with CI that runs ctest in parallel
  Initialize ADIOS::m_UserOption before tentaively calling ProcessUserConfig()
  - set ACA version to  0.1 - remove debug prints - add doc on Campaign Management in Advanced section - change static struct of UserOptions to class member of ADIOS class to make it work with gcc 4.8
  used size_t not int for map indexing to avoid type conversion
  fix remote server test, use the new binary name
  clang-format
  Use the name of the campaign in the cache path to avoid name collision
  rename remote_server to adios2_remote_server
  bug fix: the order of entries in bpdataset table is undefined but the campaign data reader relied on calculating the index as if it was sorted by the insertion order. Use a map instead to store the rowid and use that as index for the bpfile elements.
  Use yaml parser in campaign manager python script
  change a long variable to int64_t to avoid size confusion on windows
  do not include unistd.h
  Fix compiler error Remove extra file not needed
  Add support for user options in ~/.config/adios2/adios2.yaml Currently supported options:
  cmake: add sqlite3 and zlib dep in adios2 cmake pkg
  Different names for MPI and Serial tests (ornladios#4118)
  EVpath upstream to make NO_RDMA more robust (ornladios#4116)
  Warnings (ornladios#4113)
  Don't use assert() in tests (ornladios#4108)
  - Only add campaign store to file name if that is not absolute path - list command supports second argument as path
  flake8 fixes
  Update campaign manager script to handle config file, time in nanosecond format, and avoiding conflict when updating database
  ...
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

4 participants