NABoolean ComRtGetCpuStatus(char *nodeName, short cpuNum)
{
NABoolean retval = FALSE; // assume cpu is down
MS_Mon_Node_Info_Type nodeInfo;
memset(&nodeInfo, 0, sizeof(nodeInfo));
Int32 error = msg_mon_get_node_info_detail(cpuNum, &nodeInfo);
if ( XZFIL_ERR_OK == error ) {
if ( MS_Mon_State_Up == nodeInfo.node[0].state ) retval = TRUE;
}
return retval;
}
void node_cycle() {
int count;
int ferr;
MS_Mon_Node_Info_Type ninfo;
bool pp;
MS_Mon_Process_Info_Type *ppinfo1;
MS_Mon_Process_Info_Type pinfo1[2];
do {
pp = false;
do {
ferr = msg_mon_get_process_info_type(MS_ProcessType_TSE, // ptype
&count, // count
2, // max
pinfo1); // info
myassert(ferr == XZFIL_ERR_OK); // oops
if (count == 0) {
sleep(1);
printf("count is 0\n");
}
} while (count == 0);
myassert(count > 0);
if (count == 2)
pp = true;
else {
printf("server is not up yet\n");
usleep(10000); // 1/100 sec
}
} while (!pp);
// pick primary
if (pinfo1[0].backup)
ppinfo1 = &pinfo1[1];
else
ppinfo1 = &pinfo1[0];
assert(ferr == XZFIL_ERR_OK);
ferr = msg_mon_get_node_info_detail(ppinfo1->nid, &ninfo);
assert(ferr == XZFIL_ERR_OK);
if (gverbose)
printf("node-down(%d)\n", pinfo1->nid);
msg_mon_node_down(pinfo1->nid);
usleep(100000);
// wait_node_notice();
if (gverbose)
printf("node-up(%d)\n", pinfo1->nid);
msg_mon_node_up(pinfo1->nid);
usleep(100000);
// wait_node_notice();
}
bool pp_is_backup() {
bool backup;
int disable;
int ferr;
MS_Mon_Node_Info_Type node_info;
MS_Mon_Process_Info_Type proc_info;
ferr = msg_mon_get_process_info_detail(gname, &proc_info);
TEST_CHK_FEOK(ferr);
gnid = proc_info.nid;
gpid = proc_info.pid;
backup = proc_info.backup ? true : false;
if (backup) {
gpeer_nid = proc_info.parent_nid;
gpeer_pid = proc_info.parent_pid;
// set this up in case register fails
strcpy(gdisplay_name, gname);
sprintf(&gdisplay_name[strlen(gdisplay_name)],
"-%d-%d-%d-B", gnid, gpid, getpid());
if (gverbosepp)
pp_printf("we are the backup process.\n");
disable = msg_test_assert_disable();
ferr = msg_mon_register_death_notification(gpeer_nid, gpeer_pid);
if (ferr != XZFIL_ERR_OK) {
if (gverbosepp)
pp_printf("msg_mon_register_death_notification FAILED ferr=%d.\n", ferr);
ferr = msg_mon_get_node_info_detail(gpeer_nid, &node_info);
myassert(ferr == XZFIL_ERR_OK);
if (gverbosepp)
pp_printf("node-state=%d.\n", node_info.node[0].state);
if (node_info.node[0].state == MS_Mon_State_Shutdown)
gshutdown = true;
}
msg_test_assert_enable(disable);
}
return backup;
}
void kill_server() {
int count;
int err;
int ferr;
char killcmd[100];
MS_Mon_Node_Info_Type ninfo;
bool pp;
MS_Mon_Process_Info_Type *ppinfo1;
MS_Mon_Process_Info_Type pinfo1[2];
MS_Mon_Process_Info_Type pinfo2;
do {
pp = false;
ferr = msg_mon_get_process_info_type(MS_ProcessType_TSE, // ptype
&count, // count
2, // max
pinfo1); // info
myassert(ferr == XZFIL_ERR_OK); // oops
myassert(count > 0);
if (count == 2)
pp = true;
else {
printf("server is not up yet\n");
usleep(10000); // 1/100 sec
}
} while (!pp);
// pick primary
if (pinfo1[0].backup)
ppinfo1 = &pinfo1[1];
else
ppinfo1 = &pinfo1[0];
assert(ferr == XZFIL_ERR_OK);
ferr = msg_mon_get_node_info_detail(ppinfo1->nid, &ninfo);
assert(ferr == XZFIL_ERR_OK);
usleep(gksleep);
if (gvirtual || (ppinfo1->nid == gcnid)) {
do {
if (gverbose)
printf("kc=kill(%d, SIGURG)\n", pinfo1->pid);
kill(ppinfo1->pid, SIGURG);
usleep(1000);
err = kill(ppinfo1->pid, 0);
if (err == 0)
sleep(1);
} while (err == 0);
} else {
assert(ninfo.num_returned == 1);
sprintf(killcmd, "pdsh -w %s kill -s URG %d",
ninfo.node[0].node_name, ppinfo1->pid);
do {
if (gverbose)
printf("kc=%s\n", killcmd);
system(killcmd);
ferr = msg_mon_get_process_info_detail((char *) "$srv", &pinfo2);
if (ferr == XZFIL_ERR_OK) {
if ((pinfo2.nid != ppinfo1->nid) && (pinfo2.pid != ppinfo1->pid))
ferr = XZFIL_ERR_NOSUCHDEV;
else
sleep(1);
}
} while (ferr == XZFIL_ERR_OK);
}
}
