Actually retrieve the batch script contents (PySlurm#258)

tazend · web-flow · commit ffc419f4cfc8 · 2023-02-04T23:58:20.000+01:00
* Fixes "incomplete type" complaints

cython always complaint about "incomplete type" for pthread_mutex_t
and sockaddr_storage.
They must be atleast defined with a "pass", which makes the complaint
go away.

Also reformat the imports a bit.

* Actually return the batch-script contents as a string.

Previously it was just printed to stdout, so it wasn't really
useful when wanting to process the contents any further in a script
for example.
diff --git a/UPGRADE_C_API.rst b/UPGRADE_C_API.rst
@@ -51,13 +51,8 @@ Then, simply generate the header definitions like in this example:
     scripts/pyslurm_bindgen.py -D /directoy/with/slurm/headers > pyslurm/slurm/header.pxi
 
 The script outputs everything to `stdout`. Simply redirect the output to the file: :code:`pyslurm/slurm/header.pxi`.
+The headers should now be fully translated.
 
-Now, 99% of the work is done for generating the headers. For the 1% left, you now need to open the generated file, search for the two follwowing statements and comment them out:
-
-- `slurm_addr_t control_addr`
-- `phtread_mutex_t lock`
-
-The compiler will otherwise complain that these are incomplete type definitions.
 
 Compiling, Updating, Testing
 ----------------------------
diff --git a/pyslurm/pyslurm.pyx b/pyslurm/pyslurm.pyx
@@ -2312,19 +2312,71 @@ cdef class job:
 
     def slurm_job_batch_script(self, jobid):
         """
-        Retrieve the batch script for a given jobid.
+        Return the contents of the batch-script for a Job.
 
-        :param str jobid: Job id key string to search
-        :returns: String output of a jobid's batch script
+        Note: The string returned also includes all the "\n" characters
+              (new-line).
+
+        :param jobid: ID of the Job for which the script should be retrieved.
+        :type jobid: Union[str, int]
+        :raises: [ValueError]: When retrieving the Batch-Script for the Job was
+            not successful.
+        :returns: The content of the batch script.
         :rtype: `str`
         """
+        # This reimplements the slurm_job_batch_script API call. Otherwise we
+        # would have to parse the FILE* ptr we get from it back into a
+        # char* which would be a bit silly. Source:
+        # https://github.com/SchedMD/slurm/blob/7162f15af8deaf02c3bbf940d59e818cdeb5c69d/src/api/job_info.c#L1319
+        cdef:
+            slurm.job_id_msg_t msg
+            slurm.slurm_msg_t req
+            slurm.slurm_msg_t resp
+            int rc = slurm.SLURM_SUCCESS
+            str script = None
+
         if isinstance(jobid, int) or isinstance(jobid, long):
             jobid = str(jobid).encode("UTF-8")
         else:
             jobid = jobid.encode("UTF-8")
 
         jobid_xlate = slurm.slurm_xlate_job_id(jobid)
-        return slurm.slurm_job_batch_script(slurm.stdout, jobid_xlate)
+
+        slurm.slurm_msg_t_init(&req)
+        slurm.slurm_msg_t_init(&resp)
+
+        memset(&msg, 0, sizeof(msg))
+        msg.job_id   = jobid_xlate
+        req.msg_type = slurm.REQUEST_BATCH_SCRIPT
+        req.data     = &msg
+
+        rc = slurm.slurm_send_recv_controller_msg(&req, &resp,
+                                                  slurm.working_cluster_rec)
+        if rc < 0:
+            err = slurm.slurm_get_errno()
+            raise ValueError(slurm.stringOrNone(slurm.slurm_strerror(err), ''),
+                             err)
+
+        if resp.msg_type == slurm.RESPONSE_BATCH_SCRIPT:
+            script = slurm.stringOrNone(<char*>resp.data, None)
+            slurm.xfree(resp.data)
+
+        elif resp.msg_type == slurm.RESPONSE_SLURM_RC:
+            rc = (<slurm.return_code_msg_t*> resp.data).return_code
+            slurm.slurm_free_return_code_msg(<slurm.return_code_msg_t*>resp.data)
+
+            if rc == slurm.SLURM_ERROR:
+                rc = slurm.slurm_get_errno()
+
+            raise ValueError(slurm.stringOrNone(slurm.slurm_strerror(rc), ''),
+                             rc)
+
+        else:
+            rc = slurm.slurm_get_errno()
+            raise ValueError(slurm.stringOrNone(slurm.slurm_strerror(rc), ''),
+                             rc)
+
+        return script
 
     cdef int fill_job_desc_from_opts(self, dict job_opts, slurm.job_desc_msg_t *desc):
         """
diff --git a/pyslurm/slurm/__init__.pxd b/pyslurm/slurm/__init__.pxd
@@ -1,19 +1,37 @@
 from libcpp cimport bool
-from posix.unistd cimport uid_t, pid_t, gid_t
-from libc.stdint cimport int8_t, int16_t, int32_t, int64_t
-from libc.stdint cimport uint8_t, uint16_t, uint32_t, uint64_t
 from cpython.version cimport PY_MAJOR_VERSION
-from libc.string cimport strlen, memcpy
 
-cdef extern from "<netinet/in.h>" nogil:
-    ctypedef struct sockaddr_in
-    ctypedef struct sockaddr_storage
+from posix.unistd cimport (
+    uid_t,
+    pid_t,
+    gid_t,
+)
+
+from libc.stdint cimport (
+    int8_t,
+    int16_t,
+    int32_t,
+    int64_t,
+    uint8_t,
+    uint16_t,
+    uint32_t,
+    uint64_t,
+)
+
+from libc.string cimport (
+    strlen,
+    memcpy,
+)
+
+cdef extern from '<netinet/in.h>' nogil:
+    ctypedef struct sockaddr_storage:
+        pass
 
 cdef extern from '<stdio.h>' nogil:
     ctypedef struct FILE
     cdef FILE *stdout
 
-cdef extern from 'time.h' nogil:
+cdef extern from '<time.h>' nogil:
     ctypedef long time_t
     double difftime(time_t time1, time_t time2)
     time_t time(time_t *t)
@@ -24,8 +42,15 @@ cdef extern from '<Python.h>' nogil:
     cdef int __LINE__
     char *__FUNCTION__
 
-cdef extern from "<pthread.h>" nogil:
-    ctypedef union pthread_mutex_t
+cdef extern from '<pthread.h>' nogil:
+    ctypedef struct pthread_mutex_t:
+        pass
+
+    ctypedef struct pthread_cond_t:
+        pass
+
+    ctypedef struct pthread_t:
+        pass
 
 cdef extern from *:
     ctypedef struct slurm_job_credential
diff --git a/pyslurm/slurm/extra.pxi b/pyslurm/slurm/extra.pxi
@@ -1,4 +1,132 @@
+#
+# Structs that are not in the Slurm headers, which need to be redefined
+# in order to implement certain features.
+#
+# For example: to communicate with the slurmctld directly in order
+# to retrieve the actual batch-script as a string.
+#
+
+# https://github.com/SchedMD/slurm/blob/26abe9188ea8712ba1eab4a8eb6322851f06a108/src/common/slurm_persist_conn.h#L51
+ctypedef enum persist_conn_type_t:
+    PERSIST_TYPE_NONE = 0
+    PERSIST_TYPE_DBD
+    PERSIST_TYPE_FED
+    PERSIST_TYPE_HA_CTL
+    PERSIST_TYPE_HA_DBD
+    PERSIST_TYPE_ACCT_UPDATE
+
+# https://github.com/SchedMD/slurm/blob/26abe9188ea8712ba1eab4a8eb6322851f06a108/src/common/slurm_persist_conn.h#L59
+ctypedef struct persist_msg_t:
+    void *conn
+    void *data
+    uint32_t data_size
+    uint16_t msg_type
+
+ctypedef int (*_slurm_persist_conn_t_callback_proc) (void *arg, persist_msg_t *msg, buf_t **out_buffer, uint32_t *uid)
+ctypedef void (*_slurm_persist_conn_t_callback_fini)(void *arg)
+
+# https://github.com/SchedMD/slurm/blob/26abe9188ea8712ba1eab4a8eb6322851f06a108/src/common/slurm_persist_conn.h#L66
+ctypedef struct slurm_persist_conn_t:
+    void *auth_cred
+    _slurm_persist_conn_t_callback_proc callback_proc
+    _slurm_persist_conn_t_callback_fini callback_fini
+    char *cluster_name
+    time_t comm_fail_time
+    uint16_t my_port
+    int fd
+    uint16_t flags
+    bool inited
+    persist_conn_type_t persist_type
+    uid_t r_uid
+    char *rem_host
+    uint16_t rem_port
+    time_t *shutdown
+    pthread_t thread_id
+    int timeout
+    slurm_trigger_callbacks_t trigger_callbacks;
+    uint16_t version
+
+# https://github.com/SchedMD/slurm/blob/20e2b354168aeb0f76d67f80122d80925c2ef32b/src/common/pack.h#L68
+ctypedef struct buf_t:
+    uint32_t magic
+    char *head
+    uint32_t size
+    uint32_t processed
+    bool mmaped
+
+# https://github.com/SchedMD/slurm/blob/20e2b354168aeb0f76d67f80122d80925c2ef32b/src/common/pack.h#L68
+ctypedef struct return_code_msg_t:
+    uint32_t return_code
+
+# https://github.com/SchedMD/slurm/blob/fe82218def7b57f5ecda9222e80662ebbb6415f8/src/common/slurm_protocol_defs.h#L650
+ctypedef struct job_id_msg_t:
+    uint32_t job_id
+    uint16_t show_flags
+
+# https://github.com/SchedMD/slurm/blob/fe82218def7b57f5ecda9222e80662ebbb6415f8/src/common/slurm_protocol_defs.h#L216
+# Only partially defined - not everything needed at the moment.
+ctypedef enum slurm_msg_type_t:
+    REQUEST_SHARE_INFO    = 2022
+    REQUEST_BATCH_SCRIPT  = 2051
+    RESPONSE_BATCH_SCRIPT = 2052
+    RESPONSE_SLURM_RC     = 8001
+
+# https://github.com/SchedMD/slurm/blob/fe82218def7b57f5ecda9222e80662ebbb6415f8/src/common/slurm_protocol_defs.h#L469
+ctypedef struct forward_t:
+    uint16_t cnt
+    uint16_t init
+    char *nodelist
+    uint32_t timeout
+    uint16_t tree_width
+
+# https://github.com/SchedMD/slurm/blob/fe82218def7b57f5ecda9222e80662ebbb6415f8/src/common/slurm_protocol_defs.h#L491
+ctypedef struct forward_struct_t:
+    char *buf
+    int buf_len
+    uint16_t fwd_cnt
+    pthread_mutex_t forward_mutex
+    pthread_cond_t notify
+    List ret_list
+    uint32_t timeout
+
+# https://github.com/SchedMD/slurm/blob/fe82218def7b57f5ecda9222e80662ebbb6415f8/src/common/slurm_protocol_defs.h#L514
+ctypedef struct slurm_msg_t:
+    slurm_addr_t address
+    void *auth_cred
+    int auth_index
+    uid_t auth_uid
+    bool auth_uid_set
+    uid_t restrict_uid
+    bool restrict_uid_set
+    uint32_t body_offset
+    buf_t *buffer
+    slurm_persist_conn_t *conn
+    int conn_fd
+    void *data
+    uint32_t data_size
+    uint16_t flags
+    uint8_t hash_index
+    uint16_t msg_type
+    uint16_t protocol_version
+    forward_t forward
+    forward_struct_t *forward_struct
+    slurm_addr_t orig_addr
+    List ret_list
+
+# https://github.com/SchedMD/slurm/blob/fe82218def7b57f5ecda9222e80662ebbb6415f8/src/common/slurm_protocol_defs.c#L865
+cdef extern void slurm_free_return_code_msg(return_code_msg_t *msg)
+
+# https://github.com/SchedMD/slurm/blob/2d2e83674b59410a7ed8ab6fc8d8acfcfa8beaf9/src/common/slurm_protocol_api.c#L2401 
+cdef extern int slurm_send_recv_controller_msg(slurm_msg_t *request_msg,
+                                        slurm_msg_t *response_msg,
+                                        slurmdb_cluster_rec_t *working_cluster_rec)
+
+# https://github.com/SchedMD/slurm/blob/fe82218def7b57f5ecda9222e80662ebbb6415f8/src/common/slurm_protocol_defs.c#L168
+cdef extern void slurm_msg_t_init(slurm_msg_t *msg)
+
+
 # Global Environment
+
 cdef extern char **environ
 
 #
