/****** execd/loadsensor/sge_ls_start_ls() ************************************
* NAME
* sge_ls_start_ls -- starts a loadsensor
*
* SYNOPSIS
* static void sge_ls_start_ls(const char *qualified_hostname, lListElem *this_ls)
*
* FUNCTION
* An additional loadsensor process will be started. The name
* of the script has to be stored in the LS_command entry of
* 'this_ls' before this function will be called.
*
* The process environment of the loadsensor will contain
* the HOST variable. This variable contains the hostname
* of the execution daemon which calls this function.
*
* If 'this_ls' correlates to the 'qidle'-loadsensor then
* also the XAUTHORITY environment variable will be set.
*
* INPUTS
* qualified_hostname - qualified host name
* this_ls - pointer to a CULL element of type LS_Type
*
* RESULT
* An additional loadsensor process will be started.
* [this_ls] - the CULL element will be modified
* LS_pid contains the pid of the ls process
* LS_in, LS_out, LS_err are the FILE-streams for the
* communication with the ls-process
* returns LS_OK
* If sge_peopen fails, returns LS_CANT_PEOPEN
******************************************************************************/
static int sge_ls_start_ls(const char *qualified_hostname, lListElem *this_ls)
{
pid_t pid = -1;
FILE *fp_in = NULL;
FILE *fp_out = NULL;
FILE *fp_err = NULL;
char buffer[1024];
char **envp = NULL;
DENTER(TOP_LAYER, "sge_ls_start_ls");
snprintf(buffer, sizeof(buffer), "%s=%s", "HOST", qualified_hostname);
if (has_to_use_qidle
&& !strcmp(lGetString(this_ls, LS_name), IDLE_LOADSENSOR_NAME)) {
envp = sge_malloc(sizeof(char *) * 3);
envp[0] = buffer;
envp[1] = "XAUTHORITY=/tmp/.xauthority";
envp[2] = NULL;
} else {
envp = sge_malloc(sizeof(char *) * 2);
envp[0] = buffer;
envp[1] = NULL;
}
/* we need fds for select() .. */
pid = sge_peopen("/bin/sh", 0, lGetString(this_ls, LS_command), NULL, envp,
&fp_in, &fp_out, &fp_err, true);
if (envp) {
sge_free(&envp);
}
if (pid == -1) {
return LS_CANT_PEOPEN;
}
/* we need load reports non blocking */
fcntl(fileno(fp_out), F_SETFL, O_NONBLOCK);
sge_ls_set_pid(this_ls, pid);
lSetRef(this_ls, LS_in, fp_in);
lSetRef(this_ls, LS_out, fp_out);
lSetRef(this_ls, LS_err, fp_err);
DPRINTF(("%s: successfully started load sensor \"%s\"\n",
SGE_FUNC, lGetString(this_ls, LS_command)));
/* request first load report after starting */
ls_send_command(this_ls, "\n");
return LS_OK;
}
/******* execd/loadsensor/sge_ls_create_ls() **********************************
* NAME
* sge_ls_create_ls -- creates a new CULL loadsensor element
*
* SYNOPSIS
* static lListElem* sge_ls_create_ls(const char *qualified_hostname,
* char *name, const char *scriptfile)
*
* FUNCTION
* The function creates a new CULL element of type LS_Type and
* returns a pointer to this object. The loadsensor will be
* started immediately.
* If it cannot be started then, LS_has_to_restart is set to
* true so that it will be attempted to be restarted in the next load interval
*
* INPUTS
* qualified_hostname - qualified host name
* name - pseudo name of the ls
* "extern" for user defined loadsensors
* "intern" for qidle and qloadsensor
* scriptfile - absolute path to the ls scriptfile
*
* RESULT
* new CULL element of type LS_Type will be returned
* and a new loadsensor process will be created by this function
******************************************************************************/
static lListElem *sge_ls_create_ls(const char *qualified_hostname, char *name, const char *scriptfile)
{
lListElem *new_ls = NULL; /* LS_Type */
SGE_STRUCT_STAT st;
DENTER(TOP_LAYER, "sge_ls_create_ls");
if (scriptfile != NULL) {
if (SGE_STAT(scriptfile, &st) != 0) {
if (strcmp(name, "extern") == 0) {
WARNING((SGE_EVENT, MSG_LS_NOMODTIME_SS, scriptfile,
strerror(errno)));
}
DRETURN(NULL);
}
new_ls = lCreateElem(LS_Type);
if (new_ls) {
/* initialize all attributes */
lSetString(new_ls, LS_name, name);
lSetString(new_ls, LS_command, scriptfile);
sge_ls_set_pid(new_ls, -1);
lSetRef(new_ls, LS_in, NULL);
lSetRef(new_ls, LS_out, NULL);
lSetRef(new_ls, LS_err, NULL);
lSetBool(new_ls, LS_has_to_restart, false);
lSetUlong(new_ls, LS_tag, 0);
lSetList(new_ls, LS_incomplete, lCreateList("", LR_Type));
lSetList(new_ls, LS_complete, lCreateList("", LR_Type));
lSetUlong(new_ls, LS_last_mod, st.st_mtime);
/* start loadsensor, if couldn't set the restart flag so that we
* restart it in the next load interval
*/
if (sge_ls_start_ls(qualified_hostname, new_ls) != LS_OK) {
lSetBool(new_ls, LS_has_to_restart, true);
}
}
}
DRETURN(new_ls);
}
/****** Eventmirror/job/job_update_master_list() *****************************
* NAME
* job_update_master_list() -- update the master list of jobs
*
* SYNOPSIS
* bool job_update_master_list(sge_object_type type,
* sge_event_action action,
* lListElem *event, void *clientdata)
*
* FUNCTION
* Update the global master list of jobs
* based on an event.
* The function is called from the event mirroring interface.
*
* A jobs array tasks are not updated by this function,
* as they are maintained by separate events.
* In addition, some scheduler specific attributes, that
* are only used in scheduler, are not updated.
*
* INPUTS
* sge_object_type type - event type
* sge_event_action action - action to perform
* lListElem *event - the raw event
* void *clientdata - client data
*
* RESULT
* bool - true, if update is successfull, else false
*
* NOTES
* The function should only be called from the event mirror interface.
*
* SEE ALSO
* Eventmirror/--Eventmirror
* Eventmirror/sge_mirror_update_master_list()
* Eventmirror/job/job_update_master_list_usage()
*******************************************************************************/
sge_callback_result
job_update_master_list(sge_evc_class_t *evc, object_description *object_base, sge_object_type type,
sge_event_action action, lListElem *event, void *clientdata)
{
lList **list;
const lDescr *list_descr;
u_long32 job_id;
lListElem *job = NULL;
lList *ja_tasks = NULL;
char id_buffer[MAX_STRING_SIZE];
dstring id_dstring;
DENTER(TOP_LAYER, "job_update_master_list");
sge_dstring_init(&id_dstring, id_buffer, MAX_STRING_SIZE);
list = sge_master_list(object_base, SGE_TYPE_JOB);
list_descr = lGetListDescr(lGetList(event, ET_new_version));
job_id = lGetUlong(event, ET_intkey);
job = job_list_locate(*list, job_id);
if (action == SGE_EMA_MOD) {
u_long32 event_type = lGetUlong(event, ET_type);
if (job == NULL) {
ERROR((SGE_EVENT, MSG_JOB_CANTFINDJOBFORUPDATEIN_SS,
job_get_id_string(job_id, 0, NULL, &id_dstring), "job_update_master_list"));
DRETURN(SGE_EMA_FAILURE);
}
if (event_type == sgeE_JOB_USAGE || event_type == sgeE_JOB_FINAL_USAGE ) {
/* special handling needed for JOB_USAGE and JOB_FINAL_USAGE events.
* they are sent for jobs, ja_tasks and pe_tasks and only contain
* the usage list.
* Preferable would probably be to send MOD events for the different
* object types.
*/
bool ret = job_update_master_list_usage(*list, event);
DRETURN(ret?SGE_EMA_OK:SGE_EMA_FAILURE);
} else {
/* this is the true modify event.
* we may not update several fields:
* - JB_ja_tasks is the task list - it is maintained by JATASK events
* - JB_host and JB_category are scheduler internal attributes
* they may not be overwritten.
* Better would be to move them from JB_Type to some scheduler specific
* object.
*/
lListElem *modified_job;
modified_job = lFirst(lGetList(event, ET_new_version));
if(job != NULL && modified_job != NULL) {
/* we want to preserve the old ja_tasks, since job update events to not contain them */
lXchgList(job, JB_ja_tasks, &ja_tasks);
lSetHost(modified_job, JB_host, lGetHost(job, JB_host));
lSetRef(modified_job, JB_category, lGetRef(job, JB_category));
}
}
}
if (sge_mirror_update_master_list(list, list_descr, job, job_get_id_string(job_id, 0, NULL, &id_dstring), action, event) != SGE_EM_OK) {
lFreeList(&ja_tasks);
DRETURN(SGE_EMA_FAILURE);
}
/* restore ja_task list after modify event */
if (action == SGE_EMA_MOD && ja_tasks != NULL) {
/* we have to search the replaced job */
job = job_list_locate(*list, job_id);
if(job == NULL) {
ERROR((SGE_EVENT, MSG_JOB_CANTFINDJOBFORUPDATEIN_SS,
job_get_id_string(job_id, 0, NULL, &id_dstring), "job_update_master_list"));
lFreeList(&ja_tasks);
DRETURN(SGE_EMA_FAILURE);
}
lXchgList(job, JB_ja_tasks, &ja_tasks);
lFreeList(&ja_tasks);
}
DRETURN(SGE_EMA_OK);
}
