static void
test_rmessage(struct pbc_env *env, struct pbc_slice *slice) {
struct pbc_rmessage * m = pbc_rmessage_new(env, "tutorial.Person", slice);
printf("name = %s\n", pbc_rmessage_string(m , "name" , 0 , NULL));
printf("id = %d\n", pbc_rmessage_integer(m , "id" , 0 , NULL));
printf("email = %s\n", pbc_rmessage_string(m , "email" , 0 , NULL));
int phone_n = pbc_rmessage_size(m, "phone");
int i;
for (i=0;i<phone_n;i++) {
struct pbc_rmessage * p = pbc_rmessage_message(m , "phone", i);
printf("\tnumber[%d] = %s\n",i,pbc_rmessage_string(p , "number", i ,NULL));
printf("\ttype[%d] = %s\n",i,pbc_rmessage_string(p, "type", i, NULL));
}
int n = pbc_rmessage_size(m , "test");
for (i=0;i<n;i++) {
printf("test[%d] = %d\n",i, pbc_rmessage_integer(m , "test" , i , NULL));
}
printf("tutorial.Ext.test = %d\n", pbc_rmessage_integer(m,"tutorial.Ext.test",0,NULL));
pbc_rmessage_delete(m);
}
static void
_register(struct pbc_env *p, struct pbc_rmessage * file, struct _stringpool *pool) {
int package_sz;
const char *package = pbc_rmessage_string(file, "package", 0, &package_sz);
pbc_array queue;
_pbcA_open(queue);
int enum_count = pbc_rmessage_size(file, "enum_type");
int i;
for (i=0;i<enum_count;i++) {
struct pbc_rmessage * enum_type = pbc_rmessage_message(file, "enum_type", i);
_register_enum(p, pool , enum_type, package, package_sz);
}
int message_count = pbc_rmessage_size(file, "message_type");
for (i=0;i<message_count;i++) {
struct pbc_rmessage * message_type = pbc_rmessage_message(file, "message_type", i);
_register_message(p, pool, message_type, package, package_sz, queue);
}
_register_extension(p, pool, package, package_sz, file , queue);
_pbcB_register_fields(p, queue);
_pbcA_close(queue);
}
static void
test_rmessage(struct pbc_env *env, struct pbc_slice *slice) {
struct pbc_rmessage * m = pbc_rmessage_new(env, "tutorial.Person", slice);
if (m==NULL) {
printf("Error : %s",pbc_error(env));
return;
}
printf("name = %s\n", pbc_rmessage_string(m , "name" , 0 , NULL));
printf("id = %d\n", pbc_rmessage_integer(m , "id" , 0 , NULL));
printf("email = %s\n", pbc_rmessage_string(m , "email" , 0 , NULL));
int phone_n = pbc_rmessage_size(m, "phone");
int i;
const char * field_name;
pbc_type(env, "tutorial.Person", "phone", &field_name);
printf("phone type [%s]\n",field_name);
for (i=0;i<phone_n;i++) {
struct pbc_rmessage * p = pbc_rmessage_message(m , "phone", i);
printf("\tnumber[%d] = %s\n",i,pbc_rmessage_string(p , "number", i ,NULL));
printf("\ttype[%d] = %s\n",i,pbc_rmessage_string(p, "type", i, NULL));
}
int n = pbc_rmessage_size(m , "test");
for (i=0;i<n;i++) {
printf("test[%d] = %d\n",i, pbc_rmessage_integer(m , "test" , i , NULL));
}
printf("tutorial.Ext.test = %d\n", pbc_rmessage_integer(m,"tutorial.Ext.test",0,NULL));
pbc_rmessage_delete(m);
}
static void
_register_extension(struct pbc_env *p, struct _stringpool *pool , const char * prefix, int prefix_sz, struct pbc_rmessage * msg, pbc_array queue) {
int extension_count = pbc_rmessage_size(msg , "extension");
if (extension_count <= 0)
return;
int i;
const char * last = NULL;
for (i=0;i<extension_count;i++) {
struct pbc_rmessage * extension = pbc_rmessage_message(msg, "extension", i);
int field_name_sz = 0;
struct _field f;
const char * field_name = pbc_rmessage_string(extension , "name" , 0, &field_name_sz);
f.name = _concat_name(pool, prefix, prefix_sz, field_name, field_name_sz, NULL);
_register_field(extension, &f , pool);
const char * extendee = pbc_rmessage_string(extension , "extendee" , 0, NULL);
_pbcP_push_message(p, extendee + 1 , &f , queue);
if (last == NULL) {
last = extendee;
} else if (strcmp(extendee,last) != 0) {
_pbcP_init_message(p, last+1);
last = extendee;
}
}
_pbcP_init_message(p, last+1);
}
static int
_rmessage_size(lua_State *L) {
struct pbc_rmessage * m = (struct pbc_rmessage *)checkuserdata(L,1);
const char * key = luaL_checkstring(L,2);
int sz = pbc_rmessage_size(m, key);
lua_pushinteger(L, sz);
return 1;
}
static void
_register_message(struct pbc_env *p, struct _stringpool *pool, struct pbc_rmessage * message_type, const char *prefix, int prefix_sz, pbc_array queue) {
int name_sz;
const char * name = pbc_rmessage_string(message_type, "name", 0 , &name_sz);
int sz = 0;
const char *temp = _concat_name(pool, prefix , prefix_sz , name , name_sz, &sz);
int field_count = pbc_rmessage_size(message_type, "field");
int i;
for (i=0;i<field_count;i++) {
struct pbc_rmessage * field = pbc_rmessage_message(message_type, "field" , i);
struct _field f;
int field_name_sz;
const char * field_name = pbc_rmessage_string(field, "name", 0 , &field_name_sz);
f.name = _pbcS_build(pool,field_name,field_name_sz);
_register_field(field, &f , pool);
_pbcP_push_message(p, temp , &f , queue);
}
_pbcP_init_message(p, temp);
_register_extension(p, pool, temp, sz,message_type, queue);
// nested enum
int enum_count = pbc_rmessage_size(message_type, "enum_type");
for (i=0;i<enum_count;i++) {
struct pbc_rmessage * enum_type = pbc_rmessage_message(message_type, "enum_type", i);
_register_enum(p, pool, enum_type, temp, sz);
}
// nested type
int message_count = pbc_rmessage_size(message_type, "nested_type");
for (i=0;i<message_count;i++) {
struct pbc_rmessage * nested_type = pbc_rmessage_message(message_type, "nested_type", i);
_register_message(p, pool, nested_type, temp, sz, queue);
}
}
static void
_register_field(struct pbc_rmessage * field, struct _field * f, struct _stringpool *pool) {
f->id = pbc_rmessage_integer(field, "number", 0 , 0);
f->type = pbc_rmessage_integer(field, "type", 0 , 0); // enum
f->label = pbc_rmessage_integer(field, "label", 0, 0) - 1; // LABEL_OPTIONAL = 0
if (pbc_rmessage_size(field , "options") > 0) {
struct pbc_rmessage * options = pbc_rmessage_message(field, "options" , 0);
int packed = pbc_rmessage_integer(options , "packed" , 0 , NULL);
if (packed) {
f->label = LABEL_PACKED;
}
}
f->type_name.n = pbc_rmessage_string(field, "type_name", 0 , NULL) +1; // abandon prefix '.'
int vsz;
const char * default_value = pbc_rmessage_string(field, "default_value", 0 , &vsz);
_set_default(pool , f , f->type, default_value , vsz);
}
static void
_register_enum(struct pbc_env *p, struct _stringpool *pool, struct pbc_rmessage * enum_type, const char *prefix, int prefix_sz) {
int field_count = pbc_rmessage_size(enum_type, "value");
struct map_kv *table = malloc(field_count * sizeof(struct map_kv));
int i;
for (i=0;i<field_count;i++) {
struct pbc_rmessage * value = pbc_rmessage_message(enum_type, "value", i);
int enum_name_sz;
const char *enum_name = pbc_rmessage_string(value , "name" , 0 , &enum_name_sz);
table[i].pointer = (void *)_pbcS_build(pool, enum_name , enum_name_sz);
table[i].id = pbc_rmessage_integer(value , "number", 0 , 0);
}
int name_sz;
const char * name = pbc_rmessage_string(enum_type, "name", 0 , &name_sz);
const char *temp = _concat_name(pool, prefix , prefix_sz , name , name_sz, NULL);
_pbcP_push_enum(p,temp,table,field_count);
free(table);
}
int
pbc_register(struct pbc_env * p, struct pbc_slice *slice) {
//dump((uint8_t *)slice->buffer, slice->len);
struct pbc_rmessage * message = pbc_rmessage_new(p, "google.protobuf.FileDescriptorSet", slice);
if (message == NULL) {
p->lasterror = "register open google.protobuf.FileDescriptorSet fail";
return 1;
}
int n = pbc_rmessage_size(message, "file");
struct pbc_rmessage ** files = (struct pbc_rmessage **)_pbcM_malloc(n*(sizeof(struct pbc_rmessage *)));
int i;
int r = n;
if (n == 0) {
p->lasterror = "register empty";
goto _error;
}
for (i=0;i<n;i++) {
files[i] = pbc_rmessage_message(message, "file", i);
if (files[i] == NULL) {
p->lasterror = "register open fail";
goto _error;
}
}
do {
int rr = _register_no_dependency(p,files , n);
if (rr == r) {
p->lasterror = "register dependency error";
goto _error;
}
r = rr;
} while (r>0);
pbc_rmessage_delete(message);
free(files);
return 0;
_error:
pbc_rmessage_delete(message);
free(files);
return 1;
}
static void
dump_message(struct pbc_rmessage *m, int level) {
int t = 0;
const char *key = NULL;
for (;;) {
t = pbc_rmessage_next(m, &key);
if (key == NULL)
break;
if (t & PBC_REPEATED) {
int n = pbc_rmessage_size(m, key);
int i;
for (i=0;i<n;i++) {
dump_value(m, key , t , i , level);
}
} else {
dump_value(m, key , t , 0 , level);
}
}
}
static int
_check_file_name(struct pbc_env * p , struct pbc_rmessage * file, const char ** fname) {
const char * filename = pbc_rmessage_string(file, "name", 0, NULL);
// printf("reg :%s\n",filename);
if (_pbcM_sp_query(p->files, filename)) {
return CHECK_FILE_EXIST;
}
int sz = pbc_rmessage_size(file, "dependency");
int i;
for (i=0;i<sz;i++) {
const char *dname = pbc_rmessage_string(file,"dependency",i,NULL);
// printf("dependency :%s\n",dname);
if (_pbcM_sp_query(p->files, dname) == NULL) {
return CHECK_FILE_DEPENDENCY;
}
}
*fname = filename;
return CHECK_FILE_OK;
}
static void
test_rmessage(struct pbc_env *env, struct pbc_slice *slice) {
struct pbc_rmessage * m = pbc_rmessage_new(env, "test", slice);
if (m==NULL) {
printf("Error : %s",pbc_error(env));
return;
}
int phone_n = pbc_rmessage_size(m, "el");
int i;
for (i=0;i<phone_n;i++) {
struct pbc_rmessage * p = pbc_rmessage_message(m , "el", i);
printf("\tint16_min[%d] = %d\n",i,pbc_rmessage_integer(p , "int16_min", i ,NULL));
printf("\tdouble_max[%d] = %f\n",i,pbc_rmessage_real(p, "double_max", i));
printf("\tstring[%d] = %s\n",i,pbc_rmessage_string(p, "str", i, NULL));
}
pbc_rmessage_delete(m);
}
int
pbc_register(struct pbc_env * p, struct pbc_slice *slice) {
struct pbc_rmessage * message = pbc_rmessage_new(p, "google.protobuf.FileDescriptorSet", slice);
if (message == NULL) {
p->lasterror = "register open google.protobuf.FileDescriptorSet fail";
return 1;
}
int n = pbc_rmessage_size(message, "file");
struct pbc_rmessage * files[n];
int i;
if (n == 0) {
p->lasterror = "register empty";
goto _error;
}
for (i=0;i<n;i++) {
files[i] = pbc_rmessage_message(message, "file", i);
if (files[i] == NULL) {
p->lasterror = "register open fail";
goto _error;
}
}
int r = n;
do {
int rr = _register_no_dependency(p,files , n);
if (rr == r) {
p->lasterror = "register dependency error";
goto _error;
}
r = rr;
} while (r>0);
pbc_rmessage_delete(message);
return 0;
_error:
pbc_rmessage_delete(message);
return 1;
}
static int orte_ras_yarn_allocate_internal(int np, opal_list_t* nodes) {
int rc, i;
// create and send allocate message
struct pbc_wmessage* msg = pbc_wmessage_new(orte_hdclient_pb_env, "AllocateRequestProto");
if (!msg) {
opal_output(0, "%s ras:yarn failed to create AllocateRequestProto",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
return ORTE_ERROR;
}
pbc_wmessage_integer(msg, "resource_count", np, 0);
rc = orte_hdclient_send_message_and_delete(msg, HAMSTER_MSG_ALLOCATE);
if (rc != 0) {
opal_output(0, "%s ras:yarn error happened when send allocate msg to AM",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
return ORTE_ERROR;
}
// read rmessage out
struct pbc_rmessage* response = orte_hdclient_recv_message("AllocateResponseProto");
if (!response) {
opal_output(0, "%s ras:yarn error happened when recv allocate response from AM",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
return ORTE_ERROR;
}
int n = pbc_rmessage_size(response, "node_resources");
if (n <= 0) {
opal_output(0, "%s ras:yarn got n(=%d) <= 0, please check",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), n);
return ORTE_ERROR;
}
// read node resources
for (i = 0; i < n; i++) {
struct pbc_rmessage* node_res = pbc_rmessage_message(response, "node_resources", i);
if (!node_res) {
opal_output(0, "%s ras:yarn error when parse returned resource from AM",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
return ORTE_ERROR;
}
// parse host, slot
const char* host = pbc_rmessage_string(node_res, "host_name", 0, NULL);
if (!host) {
opal_output(0, "%s ras:yarn error when parse host from returned resource from AM",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
return ORTE_ERROR;
}
int slot = pbc_rmessage_integer(node_res, "slot", 0, NULL);
// make node_t and add it to nodes
orte_node_t* node = OBJ_NEW(orte_node_t);
if (!node) {
opal_output(0, "%s ras:yarn failed to create orte_node_t obj",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
return ORTE_ERROR;
}
node->name = strdup(host);
node->state = ORTE_NODE_STATE_UP;
node->slots_inuse = 0;
node->slots_max = 0;
node->slots = slot;
opal_list_append(nodes, &node->super);
OPAL_OUTPUT_VERBOSE((5, orte_ras_base.ras_output,
"%s ras:yarn: adding node %s with %d slot",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
host, slot));
}
// All done
OPAL_OUTPUT_VERBOSE((1, orte_ras_base.ras_output,
"%s ras:yarn:allocate: success",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
return ORTE_SUCCESS;
}
static int common_launch_process(orte_job_t *jdata, bool launch_daemon, int *launched_proc_num)
{
int i, rc;
orte_proc_t* proc = NULL;
char **argv;
int argc;
char **env;
bool error_flag = false;
int launched_num = 0;
/* 1. create launch message */
/*
message LaunchRequestProto {
repeated LaunchContextProto launch_contexts = 1;
}
message LaunchContextProto {
repeated string envars = 1;
optional string args = 2;
optional string host_name = 3;
optional ProcessNameProto name = 4;
}
message ProcessNameProto {
optional int32 jobid = 1;
optional int32 vpid = 2;
}
*/
struct pbc_wmessage* request_msg = pbc_wmessage_new(orte_hdclient_pb_env, "LaunchRequestProto");
if (!request_msg) {
opal_output(0, "%s plm:yarn:common_process_launch: failed to create AllocateRequestProto",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
return ORTE_ERROR;
}
/* when launch_daemon, start from 1 because we don't need launch HNP process */
i = launch_daemon ? 1 : 0;
for (; i < jdata->num_procs; i++) {
argv = NULL;
argc = 0;
env = NULL;
/* setup env/argv */
proc = opal_pointer_array_get_item(jdata->procs, i);
if (!proc) {
opal_output(0, "%s plm:yarn:common_launch_process: proc[%d] is NULL",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), i);
ORTE_ERROR_LOG(ORTE_ERROR_DEFAULT_EXIT_CODE);
}
if (launch_daemon) {
rc = setup_daemon_proc_env_and_argv(proc, &argv, &argc, &env);
} else {
orte_app_context_t* app = (orte_app_context_t*) opal_pointer_array_get_item(jdata->apps, proc->app_idx);
rc = setup_proc_env_and_argv(jdata, app, proc, &argv, &env);
}
if (0 != rc) {
opal_output(0,
"%s plm:yarn:common_launch_process: failed to setup env/argv of proc[%d]",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), i);
ORTE_ERROR_LOG(ORTE_ERROR_DEFAULT_EXIT_CODE);
error_flag = true;
goto cleanup;
}
/* print launch commandline and env when this env is specified */
if (getenv("HAMSTER_VERBOSE")) {
char* join_argv = opal_argv_join(argv, ' ');
OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output, "%s plm:yarn:common_launch_process: launch argv=%s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), join_argv));
if (join_argv) {
free(join_argv);
}
}
OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output,
"%s plm:yarn:common_launch_process: after setup env and argv for proc=%d.",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), i));
/* now start packing request_msg */
struct pbc_wmessage *launch_contexts_msg = pbc_wmessage_message(request_msg, "launch_contexts");
if (!launch_contexts_msg) {
opal_output(0,
"%s plm:yarn:common_process_launch: create launch_contexts_msg failed",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
error_flag = true;
goto cleanup;
}
char **tmp_env = env;
while (*tmp_env) {
pbc_wmessage_string(launch_contexts_msg, "envars", *tmp_env, strlen(*tmp_env));
tmp_env++;
}
char* join_argv = opal_argv_join(argv, ' ');
pbc_wmessage_string(launch_contexts_msg, "args", join_argv, strlen(join_argv));
pbc_wmessage_string(launch_contexts_msg, "host_name", proc->node->name, strlen(proc->node->name));
struct pbc_wmessage *proccess_name_msg = pbc_wmessage_message(launch_contexts_msg, "name");
if (!proccess_name_msg) {
opal_output(0,
"%s plm:yarn:common_process_launch: create proccess_name_msg failed",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
error_flag = true;
goto cleanup;
}
rc = pbc_wmessage_integer(proccess_name_msg, "jobid", ORTE_LOCAL_JOBID(proc->name.jobid), 0);
if (0 != rc) {
opal_output(0,
"%s plm:yarn:common_process_launch: pack jobid in proccess_name_msg failed",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
error_flag = true;
goto cleanup;
}
rc = pbc_wmessage_integer(proccess_name_msg, "vpid", proc->name.vpid,
0);
if (0 != rc) {
opal_output(0,
"%s plm:yarn:common_process_launch: pack vpid in proccess_name_msg failed",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
error_flag = true;
goto cleanup;
}
cleanup:
/* free argv and env for this proc */
if (argv) {
opal_argv_free(argv);
}
if (env) {
opal_argv_free(env);
}
if (join_argv) {
free(join_argv);
}
if (error_flag) {
pbc_wmessage_delete(request_msg);
ORTE_UPDATE_EXIT_STATUS(ORTE_ERROR_DEFAULT_EXIT_CODE);
return ORTE_ERROR;
}
}
/* 2. send launch deamon procs request msg */
rc = orte_hdclient_send_message_and_delete(request_msg, HAMSTER_MSG_LAUNCH);
if (rc != 0) {
opal_output(0,
"%s plm:yarn:common_process_launch: error happened when send launch proc request to AM",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
if (request_msg) {
pbc_wmessage_delete(request_msg);
}
return ORTE_ERROR;
}
/* 3. recv response and parse the msg*/
/*
message LaunchResponseProto {
repeated LaunchResultProto results = 1;
}
message LaunchResultProto {
optional ProcessNameProto name = 1;
optional bool success = 2;
}
message ProcessNameProto {
optional int32 jobid = 1;
optional int32 vpid = 2;
}
*/
struct pbc_rmessage* response_msg = NULL;
response_msg = orte_hdclient_recv_message("LaunchResponseProto");
if (!response_msg) {
opal_output(0,
"%s plm:yarn:common_process_launch: error happened when recv launch response msg from AM",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
goto launch_failed;
}
int n = pbc_rmessage_size(response_msg, "results");
if (n < 0) {
opal_output(0,
"%s plm:yarn:common_process_launch: got n(=%d) < 0, please check",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), n);
goto launch_failed;
}
for (i = 0; i < n; i++) {
struct pbc_rmessage* results_msg = pbc_rmessage_message(response_msg, "results", i);
if (!results_msg) {
opal_output(0,
"%s plm:yarn:launch_daemons: error when parse returned launch results from AM",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
goto launch_failed;
}
struct pbc_rmessage* proc_name_msg = pbc_rmessage_message(results_msg, "name", 0);
if (!proc_name_msg) {
opal_output(0,
"%s plm:yarn:common_process_launch: error when parse returned proc_name_msg from AM",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
goto launch_failed;
}
orte_jobid_t local_jobid = pbc_rmessage_integer(proc_name_msg, "jobid", 0, NULL);
orte_vpid_t vpid = pbc_rmessage_integer(proc_name_msg, "vpid", 0, NULL);
bool success = pbc_rmessage_integer(results_msg, "success", 0, NULL);
orte_proc_t* proc = (orte_proc_t*) opal_pointer_array_get_item(jdata->procs, vpid);
if (success) {
proc->state = ORTE_PROC_STATE_RUNNING;
launched_num++;
} else {
opal_output(0,
"%s plm:yarn:common_process_launch: launch proc failed when jobid = %u, vpid = %u",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), local_jobid, vpid);
proc->state = ORTE_PROC_STATE_FAILED_TO_START;
jdata->state = ORTE_JOB_STATE_FAILED_TO_START;
goto launch_failed;
}
}
/* to return back */
*launched_proc_num = launched_num;
return ORTE_SUCCESS;
launch_failed:
if (response_msg) {
pbc_rmessage_delete(response_msg);
}
ORTE_UPDATE_EXIT_STATUS(ORTE_ERROR_DEFAULT_EXIT_CODE);
return ORTE_ERROR;
}
static void heartbeat_with_AM_cb(int fd, short event, void *data)
{
int i, rc;
orte_job_t *jdata = (orte_job_t*)data;
orte_job_t* daemons = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
/* 1. create heartbeat request msg */
/*
message HeartbeatRequestProto {
}
*/
struct pbc_wmessage* request_msg = pbc_wmessage_new(orte_hdclient_pb_env, "HeartbeatRequestProto");
if (!request_msg) {
opal_output(0, "%s plm:yarn:heartbeat_with_AM_cb: failed to create request_msg",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
ORTE_ERROR_LOG(ORTE_ERROR_DEFAULT_EXIT_CODE);
return;
}
/* 2. send heartbeat request msg */
rc = orte_hdclient_send_message_and_delete(request_msg, HAMSTER_MSG_HEARTBEAT);
if (rc != 0) {
opal_output(0,
"%s plm:yarn:heartbeat_with_AM_cb: error happened when send request_msg to AM",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
ORTE_ERROR_LOG(ORTE_ERROR_DEFAULT_EXIT_CODE);
return;
}
/* 3. recv response and parse the msg*/
/*
message HeartbeatResponseProto {
repeated ProcessStatusProto completed_processes = 1;
}
message ProcessStatusProto {
optional ProcessNameProto name = 1;
optional ProcessStateProto state = 2;
optional int32 exit_value = 3;
}
enum ProcessStateProto {
RUNNING = 1;
COMPLETED = 2;
}
message ProcessNameProto {
optional int32 jobid = 1;
optional int32 vpid = 2;
}
*/
struct pbc_rmessage* response_msg = orte_hdclient_recv_message("HeartbeatResponseProto");
if (!response_msg) {
opal_output(0,
"%s plm:yarn:heartbeat_with_AM_cb: error happened when recv HeartbeatResponseProto msg from AM",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
goto cleanup;
}
int n = pbc_rmessage_size(response_msg, "completed_processes");
if (n < 0) {
opal_output(0,
"%s plm:yarn:heartbeat_with_AM_cb: got n(=%d) < 0, please check",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), n);
goto cleanup;
}
for (i = 0; i < n; i++) {
struct pbc_rmessage* completed_procs_msg = pbc_rmessage_message(response_msg, "completed_processes", i);
if (!completed_procs_msg) {
opal_output(0,
"%s plm:yarn:heartbeat_with_AM_cb: error when parse returned completed_procs_msg from AM",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
goto cleanup;
}
struct pbc_rmessage* proc_name_msg = pbc_rmessage_message(completed_procs_msg, "name", 0);
if (!proc_name_msg) {
opal_output(0,
"%s plm:yarn:heartbeat_with_AM_cb: error when parse proc_name_msg",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
goto cleanup;
}
uint32_t local_jobid = pbc_rmessage_integer(proc_name_msg, "jobid", 0, NULL);
uint32_t vpid = pbc_rmessage_integer(proc_name_msg, "vpid", 0, NULL);
uint32_t exit_value = pbc_rmessage_integer(completed_procs_msg, "exit_value", 0, NULL);
/* next, we will modify proc's state */
orte_job_t* tmp_jdata = (orte_job_t*) opal_pointer_array_get_item(orte_job_data, local_jobid);
orte_proc_t* proc = (orte_proc_t*) opal_pointer_array_get_item(tmp_jdata->procs, vpid);
if (tmp_jdata->jobid == jdata->jobid) {
num_completed_jdata_procs++;
}
if (exit_value == 0) {
proc->state = ORTE_PROC_STATE_TERMINATED;
}
/* if this process is already terminated, just skip over */
if (proc->state >= ORTE_PROC_STATE_TERMINATED) {
continue;
}
if (exit_value == -1000 || exit_value == -100 || exit_value == -101) {
opal_output(0, "%s plm:yarn:heartbeat_with_AM_cb proc failed to start", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
ORTE_ERROR_LOG(ORTE_ERROR);
proc->state = ORTE_PROC_STATE_FAILED_TO_START;
ORTE_ACTIVATE_PROC_STATE(&proc->name, ORTE_PROC_STATE_FAILED_TO_LAUNCH);
} else {
/* here, means currently the proc's state < ORTE_PROC_STATE_TERMINATED,
* however, from AM's heartbeat response, we got the proc's container is terminated,
* to solve this dilemma , we set a timer event to reconfirm this proc's state,
*/
opal_event_t *ev = NULL;
ev = (opal_event_t*) malloc(sizeof(opal_event_t));
struct timeval delay;
delay.tv_sec = 15;
delay.tv_usec = 0;
opal_event_evtimer_set(orte_event_base, ev, process_state_monitor_cb, proc);
opal_event_evtimer_add(ev, &delay);
}
}
cleanup:
if (response_msg) {
pbc_rmessage_delete(response_msg);
}
if (num_completed_jdata_procs == jdata->num_procs) {
/*
* all procs are completed, send finish request to AM,
* modify job state to ORTE_JOB_STATE_TERMINATED
*/
jdata->state = ORTE_JOB_STATE_TERMINATED;
finish_app_master(0 == orte_exit_status);
return;
} else {
/* next heartbeat */
opal_event_t *ev = NULL;
ev = (opal_event_t*) malloc(sizeof(opal_event_t));
struct timeval delay;
delay.tv_sec = 1;
delay.tv_usec = 0;
opal_event_evtimer_set(orte_event_base, ev, heartbeat_with_AM_cb, jdata);
opal_event_evtimer_add(ev, &delay);
}
}
