Skip to content

v21.8.0

Compare
Choose a tag to compare
@giovtorres giovtorres released this 01 Mar 13:38
· 67 commits to main since this release
8febf21
Support for Slurm 21.08 (#227)

* Expect slurm version 21.08

* Search slurm version in slurm_version.h

Starting from Slurm >= 21.08, slurm version is declared in this
dedicated header file.

* Port templates to Slurm 21.08

These templates have been generated using autopxd2, installed with:

  $ pip install autopxd2

Then, *.pxd files have been generated using these commands:

  $ cd /usr/include
  $ autopxd --include-dir . slurm/slurm_errno.h > \
    ~/pyslurm/jinja2/slurm_errno.h.pxd
  $ autopxd --include-dir . slurm/slurm.h > \
    ~/pyslurm/jinja2/slurm.h.pxd
  $ autopxd --include-dir . slurm/slurmdb.h > \
    ~/pyslurm/jinja2/slurmdb.h.pxd

Then, jinja2/slurm.h.pxd and jinja2/slurmdb.h.pxd have been manually
modified to:

- Remove libc.stdint import
- Remove duplicated slurm_errno symbols
- Include defines from dedicated subdir

Additionally:

- in jinja2/slurm.h.pxd:
  - symbols SLURM_ERROR, SLURM_SUCCESS and SLURM_VERSION have been
    restored,
  - slurm_addr_t control_addr and pthread_mutex_t lock are commented
    out, just like before, to avoid compilation error with these types
    (they are not used by pyslurm).
- in jinja2/slurmdb.h.pxd, all symbols duplicated in
  jinja2/slurm.h.pxd have been removed.

This finally produces this commit.

* Transpose e79981c in template

* Update slurmdb CLUSTER_FLAGS defs for 21.08

* Update slurm CPU_FREQ defs for 21.08

* Update NODE_STATE defs for slurm 21.08

* Update DEBUG_FLAG defs for slurm 21.08

* Update READY_* defs for slurm 21.08

* Generate pyslurm/slurm.pxd based on template

To generate this files, I used j2cli:

  $ pip install j2cli
  $ j2 jinja2/slurm.j2 > pyslurm/slurm.pxd

* Update python structures for slurm 21.08

There are some new parameters, some have vanished, some have been
renamed.

* Update *.pxi following latest defs changes

* Add new sibling arg to slurm_kill_job2()

Slurm 21.08 slurm_kill_job2() now expects a fourth char* sibling
argument.

* Bump __version__

* Typo fix in slurmdb job dict key

* Update slurm version in README

* Update NODE_STATE in examples

* Use docker image tag 21.08.0 for tests

* Update test_slurm_api_version expected value

* Use NULL slurmdb_connection_get persist_conn_flags

Use hard-coded NULL slurm.slurmdb_connection_get() persist_conn_flags
consistently. Persistent connections are not used in PySlurm. The
variable was declared to NULL and not used elsewhere, the pointer was
not preallocated in class slurmdb_jobs.

I propose NULL to be used consistently in both cases.

Co-authored-by: Nicholas Carriero <njcarriero@gmail.com>

* Initialize and dealloc slurmdb_clusters db_conn

Class slurmdb_clusters db_conn attribute is declared and used in get()
but it was not initialized with a proper connection. Also close and
free the allocation in __dealloc__().

Co-authored-by: Nicholas Carriero <njcarriero@gmail.com>

* Add missing slurm_init() and slurm_fini()

Starting from Slurm 20.11, slurm_init() must be called prior to any
other Slurm library API calls. For the moment, its load Slurm
configuration structure. For reference:

https://github.com/SchedMD/slurm/commit/e35a6e3

On the other side, slurm_fini() cleanup the configuration data
structures in memory.

Co-authored-by: Nicholas Carriero <njcarriero@gmail.com>

* Call slurm_init() at pyslurm module load

This way, PySlurm consumers do not have to do it explicitely.

* Bump to docker-centos7-slurm:21.08.6

* fix: update tests to use pyenv

Co-authored-by: Nicholas Carriero <njcarriero@gmail.com>
Co-authored-by: Giovanni Torres <giovanni.torres@gmail.com>