node_info.c

openPBS的开放源代码

  char errbuf[256];
  int i;

  if( ninfo != NULL && !ninfo -> is_down )
  {
    if( ( mom_sd = openrm(ninfo -> name, pbs_rm_port) ) < 0 )
    {
      log(PBSEVENT_SYSTEM, PBS_EVENTCLASS_REQUEST, ninfo -> name, "Can not open connection to mom");
      return 1;
    }

    for(i = 0; i < num_resget; i++)
      addreq(mom_sd, (char *) res_to_get[i]);
    
    for( i = 0; i < num_resget && (mom_ans = getreq(mom_sd)) != NULL; i++ )
    {
      if( !strcmp(res_to_get[i], "max_load") )
      {
	testd = strtod(mom_ans, &endp);
	if( *endp == '\0' )
	  ninfo -> max_load = testd;
	else 
	  ninfo -> max_load = ninfo -> ncpus;
	free(mom_ans);
      }
      else if( !strcmp(res_to_get[i], "ideal_load") )
      {
	testd = strtod(mom_ans, &endp);
	if( *endp == '\0' )
	  ninfo -> ideal_load = testd;
	else
	  ninfo -> ideal_load = ninfo -> ncpus;
	free(mom_ans);
      }
      else if( !strcmp(res_to_get[i], "arch") )
	ninfo -> arch = mom_ans;
      else if( !strcmp(res_to_get[i], "ncpus") )
      {
	testi = strtol(mom_ans, &endp, 10);
	if( *endp == '\0' )
	  ninfo -> ncpus = testi;
	else
	  ninfo -> ncpus = 1;
	free(mom_ans);
      }
      else if( !strcmp(res_to_get[i], "physmem") )
      {
	ninfo -> physmem = res_to_num( mom_ans );
	free(mom_ans);
      }
      else if( !strcmp( res_to_get[i], "loadave") )
      {
	testd = strtod(mom_ans, &endp);
	if( *endp == '\0' )
	  ninfo -> loadave = testd;
	else 
	  ninfo -> loadave = -1.0;
	free(mom_ans);
      }
      else
      {
	sprintf(errbuf, "Unknown resource value[%d]: %s", i, mom_ans);
	log(PBSEVENT_SCHED, PBS_EVENTCLASS_NODE, ninfo -> name, errbuf);
      }
    }
    closerm(mom_sd);
  }
  return 0;
}

/*
 *
 *	node_filter - filter a node array and return a new filterd array
 *
 *	  nodes - the array to filter
 *	  size  - size of nodes
 * 	  filter_func - pointer to a function that will filter the nodes
 *		- returns 1: job will be added to filtered array
 *		- returns 0: job will NOT be added to filtered array
 *	  arg - an optional arg passed to filter_func
 *
 *	returns pointer to filtered array
 *
 *	filter_func prototype: int func( node_info *, void * )
 *
 */
node_info **node_filter( node_info **nodes, int size, 
			 int (*filter_func) (node_info*, void*), void *arg )
{
  node_info **new_nodes = NULL;			/* the new node array */
  int i, j;

  if( ( new_nodes = (node_info **) malloc( ( size + 1) * sizeof( node_info *) ) ) == NULL )
  {
    perror("Memory Allocation Error");
    return NULL;
  }

  for(i = 0, j = 0; i < size; i++)
  {
    if( filter_func( nodes[i], arg) )
    {
      new_nodes[j] = nodes[i];
      j++;
    }
  }
  new_nodes[j] = NULL;

  if( j == 0 )
  {
    free(new_nodes);
    new_nodes = NULL;
  }
  else if( (new_nodes = realloc(new_nodes, (j+1) * sizeof( node_info * ) ) ) == NULL)
  {
    perror( "Memory Allocation Error" );
    free(new_nodes);
  }

  return new_nodes;
}

/*
 *
 *      is_node_timeshared - check if a node is timeshared
 *
 *        node - node to check
 *        arg  - unused argument
 *
 *      returns 
 *        1: is a timeshared node
 *        0: is not a timeshared node
 *
 *      NOTE: this function used for node_filter
 *
 */
int is_node_timeshared( node_info *node, void *arg )
{
  if( node != NULL )
    return node -> is_timeshare;

  return 0;
} 

/*
 *
 *	find_best_node - find the best node to run a job
 *
 *	  jinfo - the job to run
 *	  ninfo_arr - array of nodes to find the best of
 *
 *	returns the node to run the job on
 *
 */
node_info *find_best_node( job_info *jinfo, node_info **ninfo_arr )
{
  node_info *possible_node = NULL;	/* node which under max node not ideal*/
  node_info *good_node = NULL;		/* node which is under ideal load */
  resource_req *req;			/* used to find requested resources */
  sch_resource_t ncpus;			/* used for number of CPUS on nodes */
  sch_resource_t mem;			/* used for memory on the nodes */
  char *arch;				/* used for the architecture of nodes */
  char *host;				/* used for the node name */
  int i, c;				/* index & loop vars */
  int ln_i;				/* index of the last node in ninfo_arr*/
  float good_node_la = 1.0e10;		/* the best load ave found yet */
  float proj_la;			/* the projected load average */
  char logbuf[256];			/* buffer for log messages */

  /* name of the last node a job ran on - used in load balance round robin */
  static char last_node_name[PBS_MAXSVRJOBID];

  if( ninfo_arr == NULL || jinfo == NULL )
    return NULL;
  
  /* if the job is requesting nodes, then don't try and load balance it */
  if( find_resource_req( jinfo -> resreq, "nodes" ) != NULL )
    return NULL;
  
  if( ( req = find_resource_req( jinfo -> resreq, "ncpus" ) ) == NULL )
    ncpus = 1;
  else 
    ncpus = req -> amount;

  if( ( req = find_resource_req( jinfo -> resreq, "mem" ) ) == NULL )
    mem = 0;
  else
    mem = req -> amount;
  
  if( ( req = find_resource_req( jinfo -> resreq, "arch" ) ) == NULL )
    arch = NULL;
  else
    arch = req -> res_str;
  if( ( req = find_resource_req( jinfo -> resreq, "host" ) ) == NULL )
    host = NULL;
  else
    host = req -> res_str;

  if( last_node_name[0] == '\0' && ninfo_arr[0] != NULL )
    strcpy(last_node_name, ninfo_arr[0] -> name);

  if( cstat.load_balancing_rr )
  {
    log(PBSEVENT_DEBUG2, PBS_EVENTCLASS_NODE, last_node_name, "Last node a job was run on");
    /* find the node we last ran a job on */
    for(ln_i = 0; ninfo_arr[ln_i] != NULL && 
    	strcmp(ninfo_arr[ln_i] -> name, last_node_name); ln_i++)
      ;

    /* if the last node has been deleted since the last scheduling cycle */
    if( ninfo_arr[ln_i] == NULL )
    {
      strcpy(last_node_name, ninfo_arr[0] -> name);
      ln_i = 0;
    }
    /* if the last node is at the end of the list, cycle back to the front */
    else if( ninfo_arr[ln_i+1] == NULL)
      i = 0;
    /* start one after the last node */
    else
      i = ln_i + 1;
  } 
  else
  {
    i = 0;
    ln_i = jinfo -> queue -> server -> num_nodes;
  }


  for( c = 0; c < jinfo -> queue -> server -> num_nodes; c++ )
  {
    /* if the job didn't specify memory, it will default to 0, and if 
     * the mom didn't return physmem, it defaults to 0.
     */
    if( ninfo_arr[i] -> is_free  &&
	(arch == NULL || !strcmp(arch, ninfo_arr[i] -> arch)) &&
	(host == NULL || !strcmp(host, ninfo_arr[i] -> name)) &&
	mem <= ninfo_arr[i] -> physmem )
    {
      proj_la = ninfo_arr[i] -> loadave + ncpus;

      if(cstat.load_balancing)
      {
	if(proj_la <= ninfo_arr[i] -> ideal_load) 
	{
	  if(ninfo_arr[i] -> loadave < good_node_la) 
	  {
	    sprintf(logbuf, "Possible low-loaded node load: %6.2f proj load: %6.2f ideal_load: %6.2f last good ideal: %6.2f", ninfo_arr[i] -> loadave, proj_la, ninfo_arr[i] -> ideal_load, good_node_la);

	    good_node = ninfo_arr[i];
	    good_node_la = ninfo_arr[i] -> loadave;
	  }
	  else
	    sprintf(logbuf, "Node Rejected, Load higher then last possible node: load: %6.2f last good ideal: %6.2f", ninfo_arr[i] -> loadave, good_node_la);
	} 
	else if( possible_node == NULL && proj_la <= ninfo_arr[i] -> max_load)
	{
	  if(ninfo_arr[i] -> loadave < good_node_la) 
	  {
	    sprintf(logbuf, "Possible medium-loaded node load: %6.2f max_load: %6.2f proj load: %6.2f last good ideal: %6.2f", ninfo_arr[i] -> loadave, ninfo_arr[i] -> max_load, proj_la, good_node_la);
	    possible_node = ninfo_arr[i];
	    good_node_la = ninfo_arr[i] -> loadave;
	  }
	  else
	    sprintf(logbuf, "Node Rejected, Load higher then last possible node: load: %6.2f last good ideal: %6.2f", ninfo_arr[i] -> loadave, good_node_la);
	}
	else
	  sprintf(logbuf, "Node Rejected, Load too high: load: %6.2f proj load: %6.2f max_load: %6.2f", ninfo_arr[i] -> loadave, proj_la, ninfo_arr[i] -> max_load);
      }
      else 
      {
	if( ninfo_arr[i] -> loadave + proj_la  < ninfo_arr[i] -> max_load )
	{
	  good_node = ninfo_arr[i];
	  break;
	}
	else
	  sprintf(logbuf, "Node Rejected, Load too high: load: %6.2f proj load: %6.2f max_load: %6.2f", ninfo_arr[i] -> loadave, proj_la, ninfo_arr[i] -> max_load);
      }
    }
    else
      sprintf(logbuf, "Node Rejected, node does not fit job requirements.");


    log(PBSEVENT_DEBUG2, PBS_EVENTCLASS_NODE, ninfo_arr[i] -> name, logbuf);
    logbuf[0] = '\0';
    i++;
    if( ninfo_arr[i] == NULL )
      i = 0;
  }

  if( good_node == NULL )
  {
    if( cstat.load_balancing_rr )
      strcpy(last_node_name, possible_node -> name);
    log(PBSEVENT_DEBUG2, PBS_EVENTCLASS_NODE, possible_node -> name, "Node Chosen to run job on");
    return possible_node;
  }
  else
  {
    if( cstat.load_balancing_rr )
      strcpy(last_node_name, good_node -> name);
    log(PBSEVENT_DEBUG2, PBS_EVENTCLASS_NODE, good_node -> name, "Node Chosen to run job on");
    return good_node;
  }
  return NULL;		/* should never get here */
}

/*
 *
 *	find_node_info - find a node in a node array
 *
 *	  nodename - the node to find
 *	  ninfo_arr - the array of nodes to look in
 *
 *	returns the node or NULL of not found
 *
 */
node_info *find_node_info( char *nodename, node_info **ninfo_arr )
{
  int i;

  if( nodename != NULL && ninfo_arr != NULL )
  {
    for(i = 0; ninfo_arr[i] != NULL && 
    				strcmp(ninfo_arr[i] -> name, nodename); i++)
      ;

    return ninfo_arr[i];
  }
  return NULL;
}

/*
 *
 *	print_node - print all the information in a node.  Mainly used for 
 * 		     debugging purposes 
 *
 *	  ninfo - the node to print
 *	  brief - boolean: only print the name ?
 *
 *	returns nothing
 *
 */
void print_node( node_info *ninfo, int brief )
{
  int i;
  if( ninfo != NULL )
  {
    if( ninfo -> name != NULL )
      printf("Node: %s\n", ninfo -> name);
    
    if( !brief )
    {
      printf("is_down: %s\n", ninfo -> is_down ? "TRUE" : "FALSE");
      printf("is_free: %s\n", ninfo -> is_free ? "TRUE" : "FALSE"); 
      printf("is_offline: %s\n", ninfo -> is_offline ? "TRUE" : "FALSE"); 
      printf("is_unknown: %s\n", ninfo -> is_unknown ? "TRUE" : "FALSE"); 
      printf("is_reserved: %s\n", ninfo -> is_reserved ? "TRUE" : "FALSE"); 
      printf("is_exclusive: %s\n", ninfo -> is_exclusive ? "TRUE" : "FALSE"); 
      printf("is_sharing: %s\n", ninfo -> is_sharing ? "TRUE" : "FALSE"); 
      printf("is_timeshare: %s\n", ninfo -> is_timeshare ? "TRUE" : "FALSE"); 
      printf("is_cluster: %s\n", ninfo -> is_cluster ? "TRUE" : "FALSE"); 

      if( ninfo -> properties != NULL )
      {
	printf("Properties: ");
	for( i = 0; ninfo -> properties[i] != NULL; i++ )
	{
	  if( i )
	    printf(", ");
	  printf("%s", ninfo -> properties[i]);
	}
	printf("\n");
      }

      if( ninfo -> jobs != NULL )
      {
	printf("Running Jobs: ");
	for( i = 0; ninfo -> jobs[i] != NULL; i++ )
	{
	  if( i )
	    printf(", ");
	  printf("%s", ninfo -> jobs[i]);
	}
      }
    }
  }
}