int perform_communications_with_retry(
char *hostname,
int MOMPort,
int *fail_count)
{
int retries = 0;
int local_errno;
int rc = PBSE_NONE;
while (retries < 5)
{
rc = do_mom(hostname, MOMPort, CmdIndex);
if (rc >= 0)
break;
*fail_count = *fail_count + 1;
local_errno = -1 * rc;
switch (local_errno)
{
case EACCES:
case EPERM:
case ECONNREFUSED:
case ENETUNREACH:
case EFAULT:
case EAFNOSUPPORT:
return(rc);
break;
}
fprintf(stdout, "attempting command again\n");
retries++;
sleep(1);
continue;
}
return(rc);
} /* END perform_communications_with_retry() */
int main(
int ArgC, /* I */
char **ArgV) /* I */
{
const char *OptString = "c:Cd:f:h:p:q:r:sv";
char HostList[65536];
char *HPtr;
int c;
int HostCount;
int FailCount;
/* initialize */
HostList[0] = '\0';
ConfigBuf[0] = '\0';
if (getuid() != 0)
{
fprintf(stderr, "ERROR: must be root to run this command\n");
exit(EXIT_FAILURE);
}
while ((c = getopt(ArgC, ArgV, OptString)) != EOF)
{
switch (c)
{
case 'c':
/* clear stale job */
JPtr = optarg;
CmdIndex = momClear;
break;
case 'C':
/* force cycle */
CmdIndex = momQuery;
Query[QueryI] = strdup("cycle");
QueryI++;
break;
case 'd':
/* diagnose */
/* FORMAT: momctl -d<X> */
CmdIndex = momQuery;
if ((Query[QueryI] = calloc(strlen(DiagPtr) + 3, sizeof(char))) == NULL)
{
fprintf(stderr,"ERROR: could not calloc %d bytes!\n",
(int)strlen(DiagPtr) + 3);
exit(EXIT_FAILURE);
}
if (optarg == NULL)
{
strncpy(Query[QueryI],DiagPtr,strlen(DiagPtr));
}
else
{
snprintf(Query[QueryI],strlen(DiagPtr) + 2,"%s%s",
DiagPtr,
optarg);
}
QueryI++;
break;
case 'f':
{
int rc;
FILE *fp;
long size;
if ((fp = fopen(optarg, "r")) == NULL)
{
fprintf(stderr, "ERROR: cannot open file '%s', errno: %d (%s)\n",
optarg,
errno,
strerror(errno));
exit(EXIT_FAILURE);
}
rc = fread(HostList, sizeof(HostList), 1, fp);
if ((rc == 0) && (!feof(fp)))
{
fprintf(stderr, "ERROR: cannot read file '%s', errno: %d (%s)\n",
optarg,
errno,
strerror(errno));
exit(EXIT_FAILURE);
}
size = ftell(fp);
HostList[MIN(size,(long)sizeof(HostList) - 1)] = '\0';
fclose(fp);
} /* END BLOCK */
break;
case 'h':
/* connect to specified host */
strncpy(HostList,optarg,sizeof(HostList));
break;
case 'p':
/* port */
if (optarg == NULL)
MCShowUsage("port not specified");
MOMPort = (int)strtol(optarg, NULL, 10);
if (MOMPort == 0)
MCShowUsage("invalid port specified");
break;
case 'q':
/* query resources */
if (optarg == NULL)
{
MCShowUsage("query not specified");
Query[QueryI] = strdup(DiagPtr);
}
else
{
Query[QueryI] = strdup(optarg);
}
QueryI++;
CmdIndex = momQuery;
break;
case 'r':
/* reconfigure */
{
CmdIndex = momReconfig;
/* NOTE: specify remote file to load -> 'fname' */
/* specify local file to stage -> 'LOCAL:fname' */
if (optarg == NULL)
MCShowUsage("file not specified");
if (!strncmp(optarg, "LOCAL:", strlen("LOCAL:")))
{
FILE *fp;
int size;
int rc;
char *ptr;
char *cptr;
strcpy(ConfigBuf, "CONFIG:");
cptr = ConfigBuf + strlen(ConfigBuf);
ptr = optarg + strlen("LOCAL:");
if ((fp = fopen(ptr, "r")) == NULL)
{
fprintf(stderr, "ERROR: cannot open file '%s', errno: %d (%s)\n",
optarg,
errno,
strerror(errno));
exit(EXIT_FAILURE);
}
rc = fread(cptr, sizeof(ConfigBuf) - strlen(ConfigBuf), 1, fp);
if ((rc == 0) && (!feof(fp)))
{
fprintf(stderr, "ERROR: cannot read file '%s', errno: %d (%s)\n",
optarg,
errno,
strerror(errno));
exit(EXIT_FAILURE);
}
size = ftell(fp);
ConfigBuf[MIN(size + strlen("CONFIG:"),sizeof(ConfigBuf) - 1)] = '\0';
fclose(fp);
}
else
{
strncpy(ConfigBuf, optarg, sizeof(ConfigBuf));
}
} /* END (case 'r') */
break;
case 's':
/* shutdown */
CmdIndex = momShutdown;
break;
case 'v':
/* report verbose logging */
IsVerbose = TRUE;
break;
} /* END switch (c) */
} /* END while (c = getopt()) */
if (CmdIndex == momNONE)
{
MCShowUsage("no command specified");
}
if (HostList[0] == '\0')
strcpy(HostList, LocalHost);
HPtr = strtok(HostList, ", \t\n");
HostCount = 0;
FailCount = 0;
/* at this point, all args processing and setup is completed ...
* ... now we run through each comma-delimited word in HPtr */
while (HPtr != NULL)
{
if ((*HPtr == ':') && (*(HPtr + 1) != '\0'))
{
/* finds nodes with this property */
int con;
char *def_server, *pserver, *servername;
struct batch_status *bstatus, *pbstat;
struct attrl *nodeattrs;
def_server = pbs_default();
if ((pserver = strchr(HPtr,'@')) != NULL)
{
*pserver = '\0';
servername = pserver + 1;
}
else
{
servername = def_server;
}
con = pbs_connect(servername);
if (con < 0)
{
fprintf(stderr,"failed to connect to pbs_server:%s\n",
servername);
exit(EXIT_FAILURE);
}
/* get a batch_status entry for each node in ":property" */
bstatus = pbs_statnode(con,HPtr,NULL,NULL);
if (bstatus != NULL)
{
for (pbstat = bstatus; pbstat != NULL; pbstat = pbstat->next)
{
/* check state first, only do_mom() if not down */
for (nodeattrs = pbstat->attribs; nodeattrs != NULL; nodeattrs = nodeattrs->next)
{
if (!strcmp(nodeattrs->name, ATTR_NODE_state))
{
if (!strstr(nodeattrs->value, ND_down))
{
do_mom(pbstat->name, MOMPort, CmdIndex) >= 0 ? HostCount++ : FailCount++;
}
else
{
fprintf(stderr,"%12s: skipping down node\n",
pbstat->name);
}
break;
} /* END if (attrib name eq state) */
} /* END for (nodeattrs) */
} /* END for (pbstat) */
pbs_statfree(bstatus);
} /* END if (bstatus != NULL) */
else
{
fprintf(stderr,"no nodes found in %s on %s\n",
HPtr,
servername);
}
pbs_disconnect(con);
if (pserver != NULL)
*pserver = '@';
}
else
{
do_mom(HPtr, MOMPort, CmdIndex) >= 0 ? HostCount++ : FailCount++;
} /* END if (*HPtr == ':') */
HPtr = strtok(NULL, ", \t\n");
} /* END while (HPtr != NULL) */
if (IsVerbose == TRUE)
{
fprintf(stdout, "Node Summary: %d Successful %d Failed\n",
HostCount,
FailCount);
}
/* SUCCESS */
exit(EXIT_SUCCESS);
} /* END main() */
