Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
* 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 <[email protected]> * 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 <[email protected]> * 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: SchedMD/slurm@e35a6e3 On the other side, slurm_fini() cleanup the configuration data structures in memory. Co-authored-by: Nicholas Carriero <[email protected]> * 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 <[email protected]> Co-authored-by: Giovanni Torres <[email protected]>
- Loading branch information