int main(int argc, char **argv)
{
pmix_status_t rc;
pmix_value_t *val;
pmix_proc_t proc;
pmix_info_t *info;
size_t ninfo;
volatile int active;
pmix_query_t *query;
size_t nq, n;
myquery_data_t myquery_data;
fprintf(stderr, "I AM HERE\n");
fflush(stderr);
sleep(10);
exit(0);
/* init us - since we were launched by the RM, our connection info
* will have been provided at startup. */
if (PMIX_SUCCESS != (rc = PMIx_tool_init(&myproc, NULL, 0))) {
fprintf(stderr, "Debugger daemon ns %s rank %d: PMIx_tool_init failed: %d\n", myproc.nspace, myproc.rank, rc);
exit(0);
}
fprintf(stderr, "Debugger daemon ns %s rank %d: Running\n", myproc.nspace, myproc.rank);
/* register our default event handler */
active = -1;
PMIx_Register_event_handler(NULL, 0, NULL, 0,
notification_fn, evhandler_reg_callbk, (void*)&active);
while (-1 == active) {
usleep(10);
}
if (0 != active) {
exit(active);
}
/* get the nspace of the job we are to debug */
(void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_NSLEN);
proc.rank = PMIX_RANK_WILDCARD;
if (PMIX_SUCCESS != (rc = PMIx_Get(&proc, PMIX_DEBUG_JOB, NULL, 0, &val))) {
fprintf(stderr, "[%s:%d] Failed to get job being debugged - error %d\n", myproc.nspace, myproc.rank, rc);
goto done;
}
if (NULL == val) {
fprintf(stderr, "Got NULL return\n");
goto done;
}
fprintf(stderr, "[%s:%d] Debugging %s\n", myproc.nspace, myproc.rank, val->data.string);
/* get our local proctable - for scalability reasons, we don't want to
* have our "root" debugger process get the proctable for everybody and
* send it out to us. So ask the local PMIx server for the pid's of
* our local target processes */
nq = 1;
PMIX_QUERY_CREATE(query, nq);
PMIX_ARGV_APPEND(rc, query[0].keys, PMIX_QUERY_LOCAL_PROC_TABLE);
query[0].nqual = 1;
PMIX_INFO_CREATE(query[0].qualifiers, 1);
PMIX_INFO_LOAD(&query[0].qualifiers[0], PMIX_NSPACE, val->data.string, PMIX_STRING); // the nspace we are enquiring about
/* setup the caddy to retrieve the data */
myquery_data.info = NULL;
myquery_data.ninfo = 0;
myquery_data.active = true;
/* execute the query */
if (PMIX_SUCCESS != (rc = PMIx_Query_info_nb(query, nq, cbfunc, (void*)&myquery_data))) {
fprintf(stderr, "PMIx_Query_info failed: %d\n", rc);
goto done;
}
while (myquery_data.active) {
usleep(10);
}
fprintf(stderr, "[%s:%d] Local proctable received\n", myproc.nspace, myproc.rank);
/* now that we have the proctable for our local processes, we can do our
* magic debugger stuff and attach to them. We then send a "release" event
* to them - i.e., it's the equivalent to setting the MPIR breakpoint. We
* do this with the event notification system */
(void)strncpy(proc.nspace, val->data.string, PMIX_MAX_NSLEN);
proc.rank = PMIX_RANK_WILDCARD;
/* we send the notification to just the local procs of the job being debugged */
ninfo = 1;
PMIX_INFO_CREATE(info, ninfo);
PMIX_INFO_LOAD(&info[0], PMIX_EVENT_CUSTOM_RANGE, &proc, PMIX_PROC); // deliver to the target nspace
fprintf(stderr, "[%s:%u] Sending release\n", myproc.nspace, myproc.rank);
PMIx_Notify_event(PMIX_ERR_DEBUGGER_RELEASE,
NULL, PMIX_RANGE_LOCAL,
info, ninfo, NULL, NULL);
/* do some debugger magic */
n = 0;
fprintf(stderr, "[%s:%u] Hanging around awhile, doing debugger magic\n", myproc.nspace, myproc.rank);
while (n < 5) {
usleep(1000);
++n;
}
done:
/* finalize us */
fprintf(stderr, "Debugger daemon ns %s rank %d: Finalizing\n", myproc.nspace, myproc.rank);
if (PMIX_SUCCESS != (rc = PMIx_Finalize(NULL, 0))) {
fprintf(stderr, "Debugger daemon ns %s rank %d:PMIx_Finalize failed: %d\n", myproc.nspace, myproc.rank, rc);
} else {
fprintf(stderr, "Debugger daemon ns %s rank %d:PMIx_Finalize successfully completed\n", myproc.nspace, myproc.rank);
}
fflush(stderr);
return(0);
}
static int attach_to_running_job(char *nspace)
{
pmix_status_t rc;
pmix_proc_t myproc;
pmix_query_t *query;
size_t nq;
myquery_data_t *q;
/* query the active nspaces so we can verify that the
* specified one exists */
nq = 1;
PMIX_QUERY_CREATE(query, nq);
PMIX_ARGV_APPEND(rc, query[0].keys, PMIX_QUERY_NAMESPACES);
q = (myquery_data_t*)malloc(sizeof(myquery_data_t));
DEBUG_CONSTRUCT_LOCK(&q->lock);
if (PMIX_SUCCESS != (rc = PMIx_Query_info_nb(query, nq, cbfunc, (void*)q))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Query_info failed: %d\n", myproc.nspace, myproc.rank, rc);
return -1;
}
DEBUG_WAIT_THREAD(&q->lock);
DEBUG_DESTRUCT_LOCK(&q->lock);
if (NULL == q->info) {
fprintf(stderr, "Query returned no info\n");
return -1;
}
/* the query should have returned a comma-delimited list of nspaces */
if (PMIX_STRING != q->info[0].value.type) {
fprintf(stderr, "Query returned incorrect data type: %d\n", q->info[0].value.type);
return -1;
}
if (NULL == q->info[0].value.data.string) {
fprintf(stderr, "Query returned no active nspaces\n");
return -1;
}
fprintf(stderr, "Query returned %s\n", q->info[0].value.data.string);
return 0;
#if 0
/* split the returned string and look for the given nspace */
/* if not found, then we have an error */
PMIX_INFO_FREE(info, ninfo);
/* get the proctable for this nspace */
ninfo = 1;
PMIX_INFO_CREATE(info, ninfo);
(void)strncpy(info[0].key, PMIX_QUERY_PROC_TABLE, PMIX_MAX_KEYLEN);
(void)strncpy(info[0].qualifier, nspace, PMIX_MAX_KEYLEN);
if (PMIX_SUCCESS != (rc = PMIx_Query_info_nb(info, ninfo, infocbfunc, (void*)&active))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Query_info_nb failed: %d\n", myproc.nspace, myproc.rank, rc);
return -1;
}
/* wait to get a response */
/* the query should have returned a data_array */
if (PMIX_DATA_ARRAY != info[0].type) {
fprintf(stderr, "Query returned incorrect data type: %d\n", info[0].type);
return -1;
}
if (NULL == info[0].data.darray.array) {
fprintf(stderr, "Query returned no proctable info\n");
return -1;
}
/* the data array consists of a struct:
* size_t size;
* void* array;
*
* In this case, the array is composed of pmix_proc_info_t structs:
* pmix_proc_t proc; // contains the nspace,rank of this proc
* char* hostname;
* char* executable_name;
* pid_t pid;
* int exit_code;
* pmix_proc_state_t state;
*/
/* this is where a debugger tool would process the proctable to
* create whatever blob it needs to provide to its daemons */
PMIX_INFO_FREE(info, ninfo);
/* setup the debugger daemon spawn request */
napps = 1;
PMIX_APP_CREATE(app, napps);
/* setup the name of the daemon executable to launch */
app[0].cmd = strdup("debuggerdaemon");
app[0].argc = 1;
app[0].argv = (char**)malloc(2*sizeof(char*));
app[0].argv[0] = strdup("debuggerdaemon");
app[0].argv[1] = NULL;
/* provide directives so the daemons go where we want, and
* let the RM know these are debugger daemons */
ninfo = 3;
PMIX_INFO_CREATE(app[0].info, ninfo);
PMIX_INFO_LOAD(&app[0].info[0], PMIX_MAPBY, "ppr:1:node", PMIX_STRING); // instruct the RM to launch one copy of the executable on each node
PMIX_INFO_LOAD(&app[0].info[1], PMIX_DEBUGGER_DAEMONS, true, PMIX_BOOL); // these are debugger daemons
PMIX_INFO_LOAD(&app[0].info[2], PMIX_DEBUG_TARGET, nspace, PMIX_STRING); // the "jobid" of the application to be debugged
/* spawn the daemons */
PMIx_Spawn(NULL, 0, app, napps, dspace);
/* cleanup */
PMIX_APP_FREE(app, napps);
/* this is where a debugger tool would wait until the debug operation is complete */
return 0;
#endif
}
/* Xfer FUNCTIONS FOR GENERIC PMIX TYPES */
pmix_status_t pmix_bfrops_base_value_xfer(pmix_value_t *p,
pmix_value_t *src)
{
size_t n, m;
pmix_status_t rc;
char **prarray, **strarray;
pmix_value_t *pv, *sv;
pmix_info_t *p1, *s1;
pmix_app_t *pa, *sa;
pmix_pdata_t *pd, *sd;
pmix_buffer_t *pb, *sb;
pmix_byte_object_t *pbo, *sbo;
pmix_kval_t *pk, *sk;
pmix_modex_data_t *pm, *sm;
pmix_proc_info_t *pi, *si;
pmix_query_t *pq, *sq;
pmix_envar_t *pe, *se;
/* copy the right field */
p->type = src->type;
switch (src->type) {
case PMIX_UNDEF:
break;
case PMIX_BOOL:
p->data.flag = src->data.flag;
break;
case PMIX_BYTE:
p->data.byte = src->data.byte;
break;
case PMIX_STRING:
if (NULL != src->data.string) {
p->data.string = strdup(src->data.string);
} else {
p->data.string = NULL;
}
break;
case PMIX_SIZE:
p->data.size = src->data.size;
break;
case PMIX_PID:
p->data.pid = src->data.pid;
break;
case PMIX_INT:
/* to avoid alignment issues */
memcpy(&p->data.integer, &src->data.integer, sizeof(int));
break;
case PMIX_INT8:
p->data.int8 = src->data.int8;
break;
case PMIX_INT16:
/* to avoid alignment issues */
memcpy(&p->data.int16, &src->data.int16, 2);
break;
case PMIX_INT32:
/* to avoid alignment issues */
memcpy(&p->data.int32, &src->data.int32, 4);
break;
case PMIX_INT64:
/* to avoid alignment issues */
memcpy(&p->data.int64, &src->data.int64, 8);
break;
case PMIX_UINT:
/* to avoid alignment issues */
memcpy(&p->data.uint, &src->data.uint, sizeof(unsigned int));
break;
case PMIX_UINT8:
p->data.uint8 = src->data.uint8;
break;
case PMIX_UINT16:
/* to avoid alignment issues */
memcpy(&p->data.uint16, &src->data.uint16, 2);
break;
case PMIX_UINT32:
/* to avoid alignment issues */
memcpy(&p->data.uint32, &src->data.uint32, 4);
break;
case PMIX_UINT64:
/* to avoid alignment issues */
memcpy(&p->data.uint64, &src->data.uint64, 8);
break;
case PMIX_FLOAT:
p->data.fval = src->data.fval;
break;
case PMIX_DOUBLE:
p->data.dval = src->data.dval;
break;
case PMIX_TIMEVAL:
memcpy(&p->data.tv, &src->data.tv, sizeof(struct timeval));
break;
case PMIX_TIME:
memcpy(&p->data.time, &src->data.time, sizeof(time_t));
break;
case PMIX_STATUS:
memcpy(&p->data.status, &src->data.status, sizeof(pmix_status_t));
break;
case PMIX_PROC:
PMIX_PROC_CREATE(p->data.proc, 1);
if (NULL == p->data.proc) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.proc, src->data.proc, sizeof(pmix_proc_t));
break;
case PMIX_PROC_RANK:
memcpy(&p->data.rank, &src->data.rank, sizeof(pmix_rank_t));
break;
case PMIX_BYTE_OBJECT:
case PMIX_COMPRESSED_STRING:
memset(&p->data.bo, 0, sizeof(pmix_byte_object_t));
if (NULL != src->data.bo.bytes && 0 < src->data.bo.size) {
p->data.bo.bytes = malloc(src->data.bo.size);
memcpy(p->data.bo.bytes, src->data.bo.bytes, src->data.bo.size);
p->data.bo.size = src->data.bo.size;
} else {
p->data.bo.bytes = NULL;
p->data.bo.size = 0;
}
break;
case PMIX_PERSIST:
memcpy(&p->data.persist, &src->data.persist, sizeof(pmix_persistence_t));
break;
case PMIX_SCOPE:
memcpy(&p->data.scope, &src->data.scope, sizeof(pmix_scope_t));
break;
case PMIX_DATA_RANGE:
memcpy(&p->data.range, &src->data.range, sizeof(pmix_data_range_t));
break;
case PMIX_PROC_STATE:
memcpy(&p->data.state, &src->data.state, sizeof(pmix_proc_state_t));
break;
case PMIX_PROC_INFO:
PMIX_PROC_INFO_CREATE(p->data.pinfo, 1);
if (NULL != src->data.pinfo->hostname) {
p->data.pinfo->hostname = strdup(src->data.pinfo->hostname);
}
if (NULL != src->data.pinfo->executable_name) {
p->data.pinfo->executable_name = strdup(src->data.pinfo->executable_name);
}
memcpy(&p->data.pinfo->pid, &src->data.pinfo->pid, sizeof(pid_t));
memcpy(&p->data.pinfo->exit_code, &src->data.pinfo->exit_code, sizeof(int));
memcpy(&p->data.pinfo->state, &src->data.pinfo->state, sizeof(pmix_proc_state_t));
break;
case PMIX_DATA_ARRAY:
p->data.darray = (pmix_data_array_t*)calloc(1, sizeof(pmix_data_array_t));
p->data.darray->type = src->data.darray->type;
p->data.darray->size = src->data.darray->size;
if (0 == p->data.darray->size || NULL == src->data.darray->array) {
p->data.darray->array = NULL;
p->data.darray->size = 0;
break;
}
/* allocate space and do the copy */
switch (src->data.darray->type) {
case PMIX_UINT8:
case PMIX_INT8:
case PMIX_BYTE:
p->data.darray->array = (char*)malloc(src->data.darray->size);
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size);
break;
case PMIX_UINT16:
case PMIX_INT16:
p->data.darray->array = (char*)malloc(src->data.darray->size * sizeof(uint16_t));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(uint16_t));
break;
case PMIX_UINT32:
case PMIX_INT32:
p->data.darray->array = (char*)malloc(src->data.darray->size * sizeof(uint32_t));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(uint32_t));
break;
case PMIX_UINT64:
case PMIX_INT64:
p->data.darray->array = (char*)malloc(src->data.darray->size * sizeof(uint64_t));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(uint64_t));
break;
case PMIX_BOOL:
p->data.darray->array = (char*)malloc(src->data.darray->size * sizeof(bool));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(bool));
break;
case PMIX_SIZE:
p->data.darray->array = (char*)malloc(src->data.darray->size * sizeof(size_t));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(size_t));
break;
case PMIX_PID:
p->data.darray->array = (char*)malloc(src->data.darray->size * sizeof(pid_t));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(pid_t));
break;
case PMIX_STRING:
p->data.darray->array = (char**)malloc(src->data.darray->size * sizeof(char*));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
prarray = (char**)p->data.darray->array;
strarray = (char**)src->data.darray->array;
for (n=0; n < src->data.darray->size; n++) {
if (NULL != strarray[n]) {
prarray[n] = strdup(strarray[n]);
}
}
break;
case PMIX_INT:
case PMIX_UINT:
p->data.darray->array = (char*)malloc(src->data.darray->size * sizeof(int));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(int));
break;
case PMIX_FLOAT:
p->data.darray->array = (char*)malloc(src->data.darray->size * sizeof(float));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(float));
break;
case PMIX_DOUBLE:
p->data.darray->array = (char*)malloc(src->data.darray->size * sizeof(double));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(double));
break;
case PMIX_TIMEVAL:
p->data.darray->array = (struct timeval*)malloc(src->data.darray->size * sizeof(struct timeval));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(struct timeval));
break;
case PMIX_TIME:
p->data.darray->array = (time_t*)malloc(src->data.darray->size * sizeof(time_t));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(time_t));
break;
case PMIX_STATUS:
p->data.darray->array = (pmix_status_t*)malloc(src->data.darray->size * sizeof(pmix_status_t));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(pmix_status_t));
break;
case PMIX_VALUE:
PMIX_VALUE_CREATE(p->data.darray->array, src->data.darray->size);
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
pv = (pmix_value_t*)p->data.darray->array;
sv = (pmix_value_t*)src->data.darray->array;
for (n=0; n < src->data.darray->size; n++) {
if (PMIX_SUCCESS != (rc = pmix_value_xfer(&pv[n], &sv[n]))) {
PMIX_VALUE_FREE(pv, src->data.darray->size);
return rc;
}
}
break;
case PMIX_PROC:
PMIX_PROC_CREATE(p->data.darray->array, src->data.darray->size);
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(pmix_proc_t));
break;
case PMIX_APP:
PMIX_APP_CREATE(p->data.darray->array, src->data.darray->size);
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
pa = (pmix_app_t*)p->data.darray->array;
sa = (pmix_app_t*)src->data.darray->array;
for (n=0; n < src->data.darray->size; n++) {
if (NULL != sa[n].cmd) {
pa[n].cmd = strdup(sa[n].cmd);
}
if (NULL != sa[n].argv) {
pa[n].argv = pmix_argv_copy(sa[n].argv);
}
if (NULL != sa[n].env) {
pa[n].env = pmix_argv_copy(sa[n].env);
}
if (NULL != sa[n].cwd) {
pa[n].cwd = strdup(sa[n].cwd);
}
pa[n].maxprocs = sa[n].maxprocs;
if (0 < sa[n].ninfo && NULL != sa[n].info) {
PMIX_INFO_CREATE(pa[n].info, sa[n].ninfo);
if (NULL == pa[n].info) {
PMIX_APP_FREE(pa, src->data.darray->size);
return PMIX_ERR_NOMEM;
}
pa[n].ninfo = sa[n].ninfo;
for (m=0; m < pa[n].ninfo; m++) {
PMIX_INFO_XFER(&pa[n].info[m], &sa[n].info[m]);
}
}
}
break;
case PMIX_INFO:
PMIX_INFO_CREATE(p->data.darray->array, src->data.darray->size);
p1 = (pmix_info_t*)p->data.darray->array;
s1 = (pmix_info_t*)src->data.darray->array;
for (n=0; n < src->data.darray->size; n++) {
PMIX_INFO_XFER(&p1[n], &s1[n]);
}
break;
case PMIX_PDATA:
PMIX_PDATA_CREATE(p->data.darray->array, src->data.darray->size);
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
pd = (pmix_pdata_t*)p->data.darray->array;
sd = (pmix_pdata_t*)src->data.darray->array;
for (n=0; n < src->data.darray->size; n++) {
PMIX_PDATA_XFER(&pd[n], &sd[n]);
}
break;
case PMIX_BUFFER:
p->data.darray->array = (pmix_buffer_t*)malloc(src->data.darray->size * sizeof(pmix_buffer_t));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
pb = (pmix_buffer_t*)p->data.darray->array;
sb = (pmix_buffer_t*)src->data.darray->array;
for (n=0; n < src->data.darray->size; n++) {
PMIX_CONSTRUCT(&pb[n], pmix_buffer_t);
pmix_bfrops_base_copy_payload(&pb[n], &sb[n]);
}
break;
case PMIX_BYTE_OBJECT:
case PMIX_COMPRESSED_STRING:
p->data.darray->array = (pmix_byte_object_t*)malloc(src->data.darray->size * sizeof(pmix_byte_object_t));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
pbo = (pmix_byte_object_t*)p->data.darray->array;
sbo = (pmix_byte_object_t*)src->data.darray->array;
for (n=0; n < src->data.darray->size; n++) {
if (NULL != sbo[n].bytes && 0 < sbo[n].size) {
pbo[n].size = sbo[n].size;
pbo[n].bytes = (char*)malloc(pbo[n].size);
memcpy(pbo[n].bytes, sbo[n].bytes, pbo[n].size);
} else {
pbo[n].bytes = NULL;
pbo[n].size = 0;
}
}
break;
case PMIX_KVAL:
p->data.darray->array = (pmix_kval_t*)calloc(src->data.darray->size , sizeof(pmix_kval_t));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
pk = (pmix_kval_t*)p->data.darray->array;
sk = (pmix_kval_t*)src->data.darray->array;
for (n=0; n < src->data.darray->size; n++) {
if (NULL != sk[n].key) {
pk[n].key = strdup(sk[n].key);
}
if (NULL != sk[n].value) {
PMIX_VALUE_CREATE(pk[n].value, 1);
if (NULL == pk[n].value) {
free(p->data.darray->array);
return PMIX_ERR_NOMEM;
}
if (PMIX_SUCCESS != (rc = pmix_value_xfer(pk[n].value, sk[n].value))) {
return rc;
}
}
}
break;
case PMIX_MODEX:
PMIX_MODEX_CREATE(p->data.darray->array, src->data.darray->size);
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
pm = (pmix_modex_data_t*)p->data.darray->array;
sm = (pmix_modex_data_t*)src->data.darray->array;
for (n=0; n < src->data.darray->size; n++) {
memcpy(&pm[n], &sm[n], sizeof(pmix_modex_data_t));
if (NULL != sm[n].blob && 0 < sm[n].size) {
pm[n].blob = (uint8_t*)malloc(sm[n].size);
if (NULL == pm[n].blob) {
return PMIX_ERR_NOMEM;
}
memcpy(pm[n].blob, sm[n].blob, sm[n].size);
pm[n].size = sm[n].size;
} else {
pm[n].blob = NULL;
pm[n].size = 0;
}
}
break;
case PMIX_PERSIST:
p->data.darray->array = (pmix_persistence_t*)malloc(src->data.darray->size * sizeof(pmix_persistence_t));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(pmix_persistence_t));
break;
case PMIX_POINTER:
p->data.darray->array = (char**)malloc(src->data.darray->size * sizeof(char*));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
prarray = (char**)p->data.darray->array;
strarray = (char**)src->data.darray->array;
for (n=0; n < src->data.darray->size; n++) {
prarray[n] = strarray[n];
}
break;
case PMIX_SCOPE:
p->data.darray->array = (pmix_scope_t*)malloc(src->data.darray->size * sizeof(pmix_scope_t));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(pmix_scope_t));
break;
case PMIX_DATA_RANGE:
p->data.darray->array = (pmix_data_range_t*)malloc(src->data.darray->size * sizeof(pmix_data_range_t));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(pmix_data_range_t));
break;
case PMIX_COMMAND:
p->data.darray->array = (pmix_cmd_t*)malloc(src->data.darray->size * sizeof(pmix_cmd_t));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(pmix_cmd_t));
break;
case PMIX_INFO_DIRECTIVES:
p->data.darray->array = (pmix_info_directives_t*)malloc(src->data.darray->size * sizeof(pmix_info_directives_t));
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
memcpy(p->data.darray->array, src->data.darray->array, src->data.darray->size * sizeof(pmix_info_directives_t));
break;
case PMIX_PROC_INFO:
PMIX_PROC_INFO_CREATE(p->data.darray->array, src->data.darray->size);
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
pi = (pmix_proc_info_t*)p->data.darray->array;
si = (pmix_proc_info_t*)src->data.darray->array;
for (n=0; n < src->data.darray->size; n++) {
memcpy(&pi[n].proc, &si[n].proc, sizeof(pmix_proc_t));
if (NULL != si[n].hostname) {
pi[n].hostname = strdup(si[n].hostname);
} else {
pi[n].hostname = NULL;
}
if (NULL != si[n].executable_name) {
pi[n].executable_name = strdup(si[n].executable_name);
} else {
pi[n].executable_name = NULL;
}
pi[n].pid = si[n].pid;
pi[n].exit_code = si[n].exit_code;
pi[n].state = si[n].state;
}
break;
case PMIX_DATA_ARRAY:
PMIX_ERROR_LOG(PMIX_ERR_NOT_SUPPORTED);
return PMIX_ERR_NOT_SUPPORTED; // don't support iterative arrays
case PMIX_QUERY:
PMIX_QUERY_CREATE(p->data.darray->array, src->data.darray->size);
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
pq = (pmix_query_t*)p->data.darray->array;
sq = (pmix_query_t*)src->data.darray->array;
for (n=0; n < src->data.darray->size; n++) {
if (NULL != sq[n].keys) {
pq[n].keys = pmix_argv_copy(sq[n].keys);
}
if (NULL != sq[n].qualifiers && 0 < sq[n].nqual) {
PMIX_INFO_CREATE(pq[n].qualifiers, sq[n].nqual);
if (NULL == pq[n].qualifiers) {
PMIX_QUERY_FREE(pq, src->data.darray->size);
return PMIX_ERR_NOMEM;
}
for (m=0; m < sq[n].nqual; m++) {
PMIX_INFO_XFER(&pq[n].qualifiers[m], &sq[n].qualifiers[m]);
}
pq[n].nqual = sq[n].nqual;
} else {
pq[n].qualifiers = NULL;
pq[n].nqual = 0;
}
}
break;
case PMIX_ENVAR:
PMIX_ENVAR_CREATE(p->data.darray->array, src->data.darray->size);
if (NULL == p->data.darray->array) {
return PMIX_ERR_NOMEM;
}
pe = (pmix_envar_t*)p->data.darray->array;
se = (pmix_envar_t*)src->data.darray->array;
for (n=0; n < src->data.darray->size; n++) {
if (NULL != se[n].envar) {
pe[n].envar = strdup(se[n].envar);
}
if (NULL != se[n].value) {
pe[n].value = strdup(se[n].value);
}
pe[n].separator = se[n].separator;
}
break;
default:
return PMIX_ERR_UNKNOWN_DATA_TYPE;
}
break;
case PMIX_POINTER:
memcpy(&p->data.ptr, &src->data.ptr, sizeof(void*));
break;
case PMIX_ENVAR:
PMIX_ENVAR_CONSTRUCT(&p->data.envar);
if (NULL != src->data.envar.envar) {
p->data.envar.envar = strdup(src->data.envar.envar);
}
if (NULL != src->data.envar.value) {
p->data.envar.value = strdup(src->data.envar.value);
}
p->data.envar.separator = src->data.envar.separator;
break;
/**** DEPRECATED ****/
case PMIX_INFO_ARRAY:
p->data.array->size = src->data.array->size;
if (0 < src->data.array->size) {
p->data.array->array = (pmix_info_t*)malloc(src->data.array->size * sizeof(pmix_info_t));
if (NULL == p->data.array->array) {
return PMIX_ERR_NOMEM;
}
p1 = (pmix_info_t*)p->data.array->array;
s1 = (pmix_info_t*)src->data.array->array;
for (n=0; n < src->data.darray->size; n++) {
PMIX_INFO_XFER(&p1[n], &s1[n]);
}
}
break;
/********************/
default:
pmix_output(0, "XFER-PMIX-VALUE: UNSUPPORTED TYPE %d", (int)src->type);
assert(0);
return PMIX_ERROR;
}
return PMIX_SUCCESS;
}
int main(int argc, char **argv)
{
pmix_status_t rc;
pmix_value_t *val;
pmix_proc_t proc;
pmix_info_t *info;
size_t ninfo;
pmix_query_t *query;
size_t nq, n;
myquery_data_t myquery_data;
pid_t pid;
pmix_status_t code = PMIX_ERR_JOB_TERMINATED;
mylock_t mylock;
myrel_t myrel;
uint16_t localrank;
char *target = NULL;
pid = getpid();
/* init us - since we were launched by the RM, our connection info
* will have been provided at startup. */
if (PMIX_SUCCESS != (rc = PMIx_tool_init(&myproc, NULL, 0))) {
fprintf(stderr, "Debugger daemon: PMIx_tool_init failed: %d\n", rc);
exit(0);
}
fprintf(stderr, "Debugger daemon ns %s rank %d pid %lu: Running\n", myproc.nspace, myproc.rank, (unsigned long)pid);
/* register our default event handler */
DEBUG_CONSTRUCT_LOCK(&mylock);
PMIx_Register_event_handler(NULL, 0, NULL, 0,
notification_fn, evhandler_reg_callbk, (void*)&mylock);
DEBUG_WAIT_THREAD(&mylock);
if (PMIX_SUCCESS != mylock.status) {
rc = mylock.status;
DEBUG_DESTRUCT_LOCK(&mylock);
goto done;
}
DEBUG_DESTRUCT_LOCK(&mylock);
/* get the nspace of the job we are to debug - it will be in our JOB info */
#ifdef PMIX_LOAD_PROCID
PMIX_LOAD_PROCID(&proc, myproc.nspace, PMIX_RANK_WILDCARD);
#else
PMIX_PROC_CONSTRUCT(&proc);
(void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_KEYLEN);
proc.rank = PMIX_RANK_WILDCARD;
#endif
if (PMIX_SUCCESS != (rc = PMIx_Get(&proc, PMIX_DEBUG_JOB, NULL, 0, &val))) {
fprintf(stderr, "[%s:%d:%lu] Failed to get job being debugged - error %s\n",
myproc.nspace, myproc.rank,
(unsigned long)pid, PMIx_Error_string(rc));
goto done;
}
if (NULL == val || PMIX_STRING != val->type || NULL == val->data.string) {
fprintf(stderr, "[%s:%d:%lu] Failed to get job being debugged - NULL data returned\n",
myproc.nspace, myproc.rank, (unsigned long)pid);
goto done;
}
/* save it for later */
target = strdup(val->data.string);
PMIX_VALUE_RELEASE(val);
fprintf(stderr, "[%s:%d:%lu] Debugging %s\n", myproc.nspace, myproc.rank,
(unsigned long)pid, target);
/* get my local rank so I can determine which local proc is "mine"
* to debug */
val = NULL;
if (PMIX_SUCCESS != (rc = PMIx_Get(&myproc, PMIX_LOCAL_RANK, NULL, 0, &val))) {
fprintf(stderr, "[%s:%d:%lu] Failed to get my local rank - error %s\n",
myproc.nspace, myproc.rank,
(unsigned long)pid, PMIx_Error_string(rc));
goto done;
}
if (NULL == val) {
fprintf(stderr, "[%s:%d:%lu] Failed to get my local rank - NULL data returned\n",
myproc.nspace, myproc.rank, (unsigned long)pid);
goto done;
}
if (PMIX_UINT16 != val->type) {
fprintf(stderr, "[%s:%d:%lu] Failed to get my local rank - returned wrong type %s\n",
myproc.nspace, myproc.rank, (unsigned long)pid, PMIx_Data_type_string(val->type));
goto done;
}
/* save the data */
localrank = val->data.uint16;
PMIX_VALUE_RELEASE(val);
fprintf(stderr, "[%s:%d:%lu] my local rank %d\n", myproc.nspace, myproc.rank,
(unsigned long)pid, (int)localrank);
/* register another handler specifically for when the target
* job completes */
DEBUG_CONSTRUCT_LOCK(&myrel.lock);
myrel.nspace = strdup(proc.nspace);
PMIX_INFO_CREATE(info, 2);
PMIX_INFO_LOAD(&info[0], PMIX_EVENT_RETURN_OBJECT, &myrel, PMIX_POINTER);
/* only call me back when this specific job terminates */
PMIX_LOAD_PROCID(&proc, target, PMIX_RANK_WILDCARD);
PMIX_INFO_LOAD(&info[1], PMIX_EVENT_AFFECTED_PROC, &proc, PMIX_PROC);
fprintf(stderr, "[%s:%d:%lu] registering for termination of %s\n", myproc.nspace, myproc.rank,
(unsigned long)pid, proc.nspace);
DEBUG_CONSTRUCT_LOCK(&mylock);
PMIx_Register_event_handler(&code, 1, info, 2,
release_fn, evhandler_reg_callbk, (void*)&mylock);
DEBUG_WAIT_THREAD(&mylock);
if (PMIX_SUCCESS != mylock.status) {
rc = mylock.status;
DEBUG_DESTRUCT_LOCK(&mylock);
PMIX_INFO_FREE(info, 2);
goto done;
}
DEBUG_DESTRUCT_LOCK(&mylock);
PMIX_INFO_FREE(info, 2);
/* get our local proctable - for scalability reasons, we don't want to
* have our "root" debugger process get the proctable for everybody and
* send it out to us. So ask the local PMIx server for the pid's of
* our local target processes */
nq = 1;
PMIX_QUERY_CREATE(query, nq);
PMIX_ARGV_APPEND(rc, query[0].keys, PMIX_QUERY_LOCAL_PROC_TABLE);
query[0].nqual = 1;
PMIX_INFO_CREATE(query[0].qualifiers, 1);
PMIX_INFO_LOAD(&query[0].qualifiers[0], PMIX_NSPACE, target, PMIX_STRING); // the nspace we are enquiring about
/* setup the caddy to retrieve the data */
DEBUG_CONSTRUCT_LOCK(&myquery_data.lock);
myquery_data.info = NULL;
myquery_data.ninfo = 0;
/* execute the query */
if (PMIX_SUCCESS != (rc = PMIx_Query_info_nb(query, nq, cbfunc, (void*)&myquery_data))) {
fprintf(stderr, "PMIx_Query_info failed: %d\n", rc);
goto done;
}
DEBUG_WAIT_THREAD(&myquery_data.lock);
DEBUG_DESTRUCT_LOCK(&myquery_data.lock);
PMIX_QUERY_FREE(query, nq);
if (PMIX_SUCCESS != myquery_data.status) {
rc = myquery_data.status;
goto done;
}
fprintf(stderr, "[%s:%d:%lu] Local proctable received\n", myproc.nspace, myproc.rank, (unsigned long)pid);
/* now that we have the proctable for our local processes, we can do our
* magic debugger stuff and attach to them. We then send a "release" event
* to them - i.e., it's the equivalent to setting the MPIR breakpoint. We
* do this with the event notification system. For this example, we just
* send it to all local procs of the job being debugged */
(void)strncpy(proc.nspace, target, PMIX_MAX_NSLEN);
proc.rank = PMIX_RANK_WILDCARD;
ninfo = 2;
PMIX_INFO_CREATE(info, ninfo);
PMIX_INFO_LOAD(&info[0], PMIX_EVENT_CUSTOM_RANGE, &proc, PMIX_PROC); // deliver to the target nspace
PMIX_INFO_LOAD(&info[1], PMIX_EVENT_NON_DEFAULT, NULL, PMIX_BOOL); // deliver to the target nspace
fprintf(stderr, "[%s:%u:%lu] Sending release\n", myproc.nspace, myproc.rank, (unsigned long)pid);
rc = PMIx_Notify_event(PMIX_ERR_DEBUGGER_RELEASE,
NULL, PMIX_RANGE_CUSTOM,
info, ninfo, NULL, NULL);
if (PMIX_SUCCESS != rc) {
fprintf(stderr, "%s[%s:%u:%lu] Sending release failed with error %s(%d)\n",
myproc.nspace, myproc.rank, (unsigned long)pid, PMIx_Error_string(rc), rc);
goto done;
}
/* do some debugger magic while waiting for the job to terminate */
DEBUG_WAIT_THREAD(&myrel.lock);
done:
if (NULL != target) {
free(target);
}
/* finalize us */
fprintf(stderr, "Debugger daemon ns %s rank %d pid %lu: Finalizing\n", myproc.nspace, myproc.rank, (unsigned long)pid);
if (PMIX_SUCCESS != (rc = PMIx_Finalize(NULL, 0))) {
fprintf(stderr, "Debugger daemon ns %s rank %d:PMIx_Finalize failed: %d\n", myproc.nspace, myproc.rank, rc);
} else {
fprintf(stderr, "Debugger daemon ns %s rank %d pid %lu:PMIx_Finalize successfully completed\n", myproc.nspace, myproc.rank, (unsigned long)pid);
}
fflush(stderr);
return(0);
}
