NOTE: These changes may break any code that uses the ompi_process_name_t type. Please recompile and check for the changes detailed below.

Major change to the name server system to introduce the OOB interface. Changed the ompi_process_name_t to no longer be an OBJ, but instead just a regular C structure. Include the packing-based oob interface. This commit was SVN r2114.
2004-08-13 15:09:24 +00:00 · 2004-08-13 15:09:24 +00:00 · 4f7febc161
--- a/src/mca/ns/base/base.h
+++ b/src/mca/ns/base/base.h
@ -35,6 +35,8 @@ extern "C" {
 						     mca_ns_base_jobid_t job,
 						     mca_ns_base_vpid_t vpid);

+    ompi_process_name_t* ns_base_copy_process_name(ompi_process_name_t* name);
+
    ompi_process_name_t* ns_base_convert_string_to_process_name(const char* name);

    char* ns_base_get_proc_name_string(const ompi_process_name_t* name);
@ -74,7 +76,7 @@ extern "C" {

 extern int mca_ns_base_output;
 extern mca_ns_base_module_t ompi_name_server;  /* holds selected module's function pointers */
-extern ompi_process_name_t mca_ns_my_replica;
+extern ompi_process_name_t *mca_ns_my_replica;  /* the name of the replica for this process */
 extern bool mca_ns_base_selected;
 extern ompi_list_t mca_ns_base_components_available;
 extern mca_ns_base_component_t mca_ns_base_selected_component;
--- a/src/mca/ns/base/ns_base_local_fns.c
+++ b/src/mca/ns/base/ns_base_local_fns.c
@ -34,7 +34,7 @@ ompi_process_name_t* ns_base_create_process_name(mca_ns_base_cellid_t cell,
 	return(NULL);
    }

-    newname = OBJ_NEW(ompi_process_name_t);
+    newname = (ompi_process_name_t*)malloc(sizeof(ompi_process_name_t));
    if (NULL == newname) { /* got an error */
 	return(NULL);
    }
@ -45,6 +45,24 @@ ompi_process_name_t* ns_base_create_process_name(mca_ns_base_cellid_t cell,
    return(newname);
 }

+ompi_process_name_t* ns_base_copy_process_name(ompi_process_name_t* name)
+{
+    mca_ns_base_cellid_t cell;
+    mca_ns_base_jobid_t job;
+    mca_ns_base_vpid_t vpid;
+    ompi_process_name_t *newname;
+
+    if (NULL == name) {
+	return NULL;
+    }
+
+    cell = ns_base_get_cellid(name);
+    job = ns_base_get_jobid(name);
+    vpid = ns_base_get_vpid(name);
+
+    newname = ns_base_create_process_name(cell, job, vpid);
+    return newname;
+}

 char* ns_base_get_proc_name_string(const ompi_process_name_t* name)
 {
@ -254,3 +272,12 @@ int ns_base_compare(ompi_ns_cmp_bitmask_t fields,
    return(0);
 }

+
+int ns_base_free_name(ompi_process_name_t* name)
+{
+    if (NULL != name) {
+	free(name);
+    }
+
+    return OMPI_SUCCESS;
+}
--- a/src/mca/ns/base/ns_base_open.c
+++ b/src/mca/ns/base/ns_base_open.c
@ -22,32 +22,12 @@
 * globals
 */

-/* constructor - used to initialize state of name instance */
-static void ompi_name_construct(ompi_process_name_t* name)
-{
-    name->cellid = 0;
-    name->jobid = 0;
-    name->vpid = 0;
-}
-
-/* destructor - used to free any resources held by instance */
-static void ompi_name_destructor(ompi_process_name_t* name)
-{
-}
-
-OBJ_CLASS_INSTANCE(
-		   ompi_process_name_t,   /* type name */
-		   ompi_object_t,         /* parent "class" name */
-		   ompi_name_construct,   /* constructor */
-		   ompi_name_destructor);  /* destructor */
-
-
 /*
 * Global variables
 */
 int mca_ns_base_output = -1;
 mca_ns_base_module_t ompi_name_server;
-ompi_process_name_t mca_ns_my_replica;
+ompi_process_name_t *mca_ns_my_replica;
 bool mca_ns_base_selected = false;
 ompi_list_t mca_ns_base_components_available;
 mca_ns_base_component_t mca_ns_base_selected_component;
--- a/src/mca/ns/base/ns_base_remote_fns.c
+++ b/src/mca/ns/base/ns_base_remote_fns.c
@ -7,6 +7,7 @@

 #include "ompi_config.h"
 #include "mca/mca.h"
+#include "util/pack.h"
 #include "mca/oob/base/base.h"
 #include "mca/ns/base/base.h"

@ -20,45 +21,118 @@

 mca_ns_base_cellid_t ns_base_create_cellid(void)
 {
-    ompi_ns_msg_buffer_t cmd, *answer;
-    struct iovec msg;
+    ompi_buffer_t cmd;
    mca_ns_base_cellid_t cell;
+    ompi_buffer_t *answer;
+    mca_ns_cmd_flag_t command;
+    int recv_tag;

-    cmd.command = OMPI_NS_CREATE_CELLID;
-    cmd.buflen = 0;
-    cmd.buf = NULL;
+    command = MCA_NS_CREATE_CELLID_CMD;
+    recv_tag = MCA_OOB_TAG_NS;

-    msg.iov_base = (char*)&cmd;
-    msg.iov_len = sizeof(cmd);
+    if (OMPI_SUCCESS != ompi_buffer_init(&cmd, 0)) { /* got a problem */
+	return OMPI_ERROR;
+    }

-    if (0 > mca_oob_send(&mca_ns_my_replica, &msg, 1, MCA_OOB_TAG_ANY, 0)) { /* error on send */
-	    return 0;
-	}
+    if (OMPI_SUCCESS != ompi_pack(cmd, (void*)&command, 1, MCA_NS_OOB_PACK_CMD)) {
+	return OMPI_ERROR;
+    }

-    if (0 > mca_oob_recv(&mca_ns_my_replica, &msg, 1, MCA_OOB_TAG_ANY, 0)) { /* error on recv */
-	    return 0;
-	}
+    if (0 > mca_oob_send_packed(mca_ns_my_replica, cmd, MCA_OOB_TAG_NS, 0)) {
+	return MCA_NS_BASE_CELLID_MAX;
+    }

-    answer = (ompi_ns_msg_buffer_t*)msg.iov_base;
-    cell = (mca_ns_base_cellid_t)answer->buf;
-    return cell;
+    if (0 > mca_oob_recv_packed(mca_ns_my_replica, answer, &recv_tag)) {
+	return MCA_NS_BASE_CELLID_MAX;
+    }
+
+    if (OMPI_SUCCESS != ompi_unpack(*answer, &cell, 1, MCA_NS_OOB_PACK_CELLID)) {
+	ompi_buffer_free(*answer);
+	return MCA_NS_BASE_CELLID_MAX;
+    } else {
+	ompi_buffer_free(*answer);
+	return cell;
+    }
 }

+
 mca_ns_base_jobid_t ns_base_create_jobid(void)
 {
-  /* JMS fill in here */
-  return 0;
+    ompi_buffer_t cmd;
+    mca_ns_base_jobid_t job;
+    ompi_buffer_t *answer;
+    mca_ns_cmd_flag_t command;
+    int recv_tag;
+
+    command = MCA_NS_CREATE_JOBID_CMD;
+    recv_tag = MCA_OOB_TAG_NS;
+
+    if (OMPI_SUCCESS != ompi_buffer_init(&cmd, 0)) { /* got a problem */
+	return OMPI_ERROR;
+    }
+
+    if (OMPI_SUCCESS != ompi_pack(cmd, (void*)&command, 1, MCA_NS_OOB_PACK_CMD)) { /* got a problem */
+	return OMPI_ERROR;
+    }
+
+    if (0 > mca_oob_send_packed(mca_ns_my_replica, cmd, MCA_OOB_TAG_NS, 0)) {
+	return MCA_NS_BASE_JOBID_MAX;
+    }
+
+    if (0 > mca_oob_recv_packed(mca_ns_my_replica, answer, &recv_tag)) {
+	return MCA_NS_BASE_JOBID_MAX;
+    }
+
+    if (OMPI_SUCCESS != ompi_unpack(*answer, &job, 1, MCA_NS_OOB_PACK_JOBID)) {
+	ompi_buffer_free(*answer);
+	return MCA_NS_BASE_JOBID_MAX;
+    } else {
+	ompi_buffer_free(*answer);
+	return job;
+    }
 }


 mca_ns_base_vpid_t ns_base_reserve_range(mca_ns_base_jobid_t job, mca_ns_base_vpid_t range)
 {
-  /* JMS fill in here */
-  return 0;
-}
+    ompi_buffer_t cmd;
+    mca_ns_base_vpid_t starting_vpid;
+    ompi_buffer_t *answer;
+    mca_ns_cmd_flag_t command;
+    int recv_tag;

+    command = MCA_NS_RESERVE_RANGE_CMD;
+    recv_tag = MCA_OOB_TAG_NS;

-int ns_base_free_name(ompi_process_name_t* name)
-{
-    return OMPI_SUCCESS;
+    if (OMPI_SUCCESS != ompi_buffer_init(&cmd, 0)) { /* got a problem */
+	return OMPI_ERROR;
+    }
+
+    if (OMPI_SUCCESS != ompi_pack(cmd, (void*)&command, 1, MCA_NS_OOB_PACK_CMD)) { /* got a problem */
+	return OMPI_ERROR;
+    }
+
+    if (OMPI_SUCCESS != ompi_pack(cmd, (void*)&job, 1, MCA_NS_OOB_PACK_JOBID)) { /* got a problem */
+	return OMPI_ERROR;
+    }
+
+    if (OMPI_SUCCESS != ompi_pack(cmd, (void*)&range, 1, MCA_NS_OOB_PACK_VPID)) { /* got a problem */
+	return OMPI_ERROR;
+    }
+
+    if (0 > mca_oob_send_packed(mca_ns_my_replica, cmd, MCA_OOB_TAG_NS, 0)) {
+	return MCA_NS_BASE_VPID_MAX;
+    }
+
+    if (0 > mca_oob_recv_packed(mca_ns_my_replica, answer, &recv_tag)) {
+	return MCA_NS_BASE_VPID_MAX;
+    }
+
+    if (OMPI_SUCCESS != ompi_unpack(*answer, &starting_vpid, 1, MCA_NS_OOB_PACK_VPID)) {
+	ompi_buffer_free(*answer);
+	return MCA_NS_BASE_VPID_MAX;
+    } else {
+	ompi_buffer_free(*answer);
+	return starting_vpid;
+    }
 }
--- a/src/mca/ns/ns.h
+++ b/src/mca/ns/ns.h
@ -26,6 +26,7 @@
 #include "class/ompi_list.h"

 #include "mca/mca.h"
+#include "util/pack.h"


 /*
@ -43,14 +44,39 @@
 #define MCA_NS_BASE_JOBID_MAX  UINT32_MAX
 #define MCA_NS_BASE_VPID_MAX   UINT32_MAX

+/*
+ * define flag values for remote commands - only used internally
+ */
+#define MCA_NS_CREATE_CELLID_CMD   0x01
+#define MCA_NS_CREATE_JOBID_CMD    0x02
+#define MCA_NS_RESERVE_RANGE_CMD   0x04
+#define MCA_NS_FREE_NAME_CMD       0x08
+
+
 /*
 * general typedefs & structures
 */
-/**< Set the allowed range for ids in each space */
+/** Set the allowed range for ids in each space */
+/* CAUTION - any changes here must also change corresponding
+ * OOB_PACK definitions below
+ */
 typedef uint32_t mca_ns_base_jobid_t;
 typedef uint32_t mca_ns_base_cellid_t;
 typedef uint32_t mca_ns_base_vpid_t;
 typedef uint8_t ompi_ns_cmp_bitmask_t;  /**< Bit mask for comparing process names */
+typedef uint16_t mca_ns_cmd_flag_t;
+
+/*
+ * packing type definitions
+ */
+/* CAUTION - any changes here must also change corresponding
+ * typedefs above
+ */
+#define MCA_NS_OOB_PACK_JOBID   OMPI_INT32
+#define MCA_NS_OOB_PACK_CELLID  OMPI_INT32
+#define MCA_NS_OOB_PACK_VPID    OMPI_INT32
+#define MCA_NS_OOB_PACK_CMD     OMPI_INT16
+

 struct ompi_process_name_t {
    mca_ns_base_cellid_t cellid;  /**< Cell number */
@ -59,8 +85,6 @@ struct ompi_process_name_t {
 };
 typedef struct ompi_process_name_t ompi_process_name_t;

-/* declare the class */
-OBJ_CLASS_DECLARATION(ompi_process_name_t);

 /*
 * define the command names/ids for use in OOB buffers.
@ -161,6 +185,18 @@ typedef mca_ns_base_jobid_t (*mca_ns_base_module_create_jobid_fn_t)(void);
 */
 typedef ompi_process_name_t* (*mca_ns_base_module_create_proc_name_fn_t)(mca_ns_base_cellid_t cell, mca_ns_base_jobid_t job, mca_ns_base_vpid_t vpid);

+/**
+ * Make a copy of a process name.
+ * Given a process name, this function creates a copy of it and returns a pointer
+ * to the duplicate structure.
+ *
+ * @param *name Pointer to an existing process name structure.
+ *
+ * @retval *newname Pointer to the duplicate structure, with all fields transferred.
+ * @retval NULL Indicates an error - most likely due to a NULL process name
+ * pointer being supplied as input.
+ */
+typedef ompi_process_name_t* (*mca_ns_base_module_copy_proc_name_fn_t)(ompi_process_name_t* name);

 /**
 * Convert a string representation to a process name.
@ -403,6 +439,7 @@ struct mca_ns_base_module_1_0_0_t {
    mca_ns_base_module_create_cellid_fn_t create_cellid;
    mca_ns_base_module_create_jobid_fn_t create_jobid;
    mca_ns_base_module_create_proc_name_fn_t create_process_name;
+    mca_ns_base_module_copy_proc_name_fn_t copy_process_name;
    mca_ns_base_module_convert_string_to_process_name_fn_t convert_string_to_process_name;
    mca_ns_base_module_reserve_range_fn_t reserve_range;
    mca_ns_base_module_free_name_fn_t free_name;
--- a/src/mca/ns/proxy/src/ns_proxy.h
+++ b/src/mca/ns/proxy/src/ns_proxy.h
@ -30,7 +30,4 @@ int mca_ns_proxy_finalize(void);
 * globals used within proxy component
 */

-extern ompi_process_name_t mca_ns_my_replica;
-
-
 #endif
--- a/src/mca/ns/proxy/src/ns_proxy_component.c
+++ b/src/mca/ns/proxy/src/ns_proxy_component.c
@ -21,6 +21,7 @@
 #include "util/proc_info.h"
 #include "util/output.h"
 #include "mca/mca.h"
+#include "mca/base/mca_base_param.h"
 #include "mca/ns/base/base.h"
 #include "ns_proxy.h"

@ -53,6 +54,7 @@ static mca_ns_base_module_t mca_ns_proxy = {
    ns_base_create_cellid,
    ns_base_create_jobid,
    ns_base_create_process_name,
+    ns_base_copy_process_name,
    ns_base_convert_string_to_process_name,
    ns_base_reserve_range,
    ns_base_free_name,
@ -75,10 +77,10 @@ static bool initialized = false;
 * globals needed within proxy component
 */

+ompi_process_name_t *mca_ns_my_replica;

 /*
- * don't really need this function - could just put NULL in the above structure
- * Just holding the place in case we decide there is something we need to do
+ * Open the proxy component and obtain the name of my replica.
 */
 int mca_ns_proxy_open(void)
 {
@ -95,34 +97,50 @@ int mca_ns_proxy_close(void)

 mca_ns_base_module_t* mca_ns_proxy_init(bool *allow_multi_user_threads, bool *have_hidden_threads, int *priority)
 {
+    int index, error_code;
+    char *my_replica_string;
+
    /* If we're NOT the seed, then we want to be selected, so do all
       the setup and return the module */
-    ompi_output(mca_ns_base_output, "ns_proxy: entered init\n");
+    /*    ompi_output(mca_ns_base_output, "ns_proxy: entered init\n"); */
    if (!ompi_process_info.seed) {

-      /* Return a module (choose an arbitrary, positive priority --
-         it's only relevant compared to other ns components).  If
-         we're not the seed, then we don't want to be selected, so
-         return NULL. */
+	/* Return a module (choose an arbitrary, positive priority --
+	   it's only relevant compared to other ns components).  If
+	   we're not the seed, then we don't want to be selected, so
+	   return NULL. */

-      *priority = 10;
+	*priority = 10;

-      /* We allow multi user threads but don't have any hidden threads */
+	/* We allow multi user threads but don't have any hidden threads */

-      *allow_multi_user_threads = true;
-      *have_hidden_threads = false;
+	*allow_multi_user_threads = true;
+	*have_hidden_threads = false;

-      /* define the replica for us to use - for now, use only the seed */
-      mca_ns_my_replica.cellid = 0;
-      mca_ns_my_replica.jobid = 0;
-      mca_ns_my_replica.vpid = 0;
+	/* define the replica for us to use */
+	/* register the mca parameter containing my replica's contact info */
+	index = mca_base_param_register_string("NS", "PROXY", "REPLICA", NULL, NULL);

-      /* Return the module */
+	error_code = mca_base_param_lookup_string(index, &my_replica_string);
+	if (OMPI_ERROR == error_code) {  /* can't lookup mca parameter for some reason - can't operate */
+	    return NULL;
+	}

-      initialized = true;
-      return &mca_ns_proxy;
+	mca_ns_my_replica = mca_ns_proxy.convert_string_to_process_name(my_replica_string);
+	if(NULL == mca_ns_my_replica) {
+	    /* couldn't convert string to name - probably parameter not set - use seed as default */
+	    mca_ns_my_replica = mca_ns_proxy.create_process_name(0,0,0);
+	    if (NULL == mca_ns_my_replica) {  /* couldn't create process name - can't operate */
+		return NULL;
+	    }
+	}
+
+	/* Return the module */
+
+	initialized = true;
+	return &mca_ns_proxy;
    } else {
-      return NULL;
+	return NULL;
    }
 }

--- a/src/mca/ns/replica/src/ns_replica.c
+++ b/src/mca/ns/replica/src/ns_replica.c
@ -69,5 +69,9 @@ mca_ns_base_vpid_t ns_replica_reserve_range(mca_ns_base_jobid_t job, mca_ns_base

 int ns_replica_free_name(ompi_process_name_t* name)
 {
+    if (NULL != name) {
+	free(name);
+    }
+
    return OMPI_SUCCESS;
 }
--- a/src/mca/ns/replica/src/ns_replica_component.c
+++ b/src/mca/ns/replica/src/ns_replica_component.c
@ -54,6 +54,7 @@ static mca_ns_base_module_t mca_ns_replica = {
    ns_replica_create_cellid,
    ns_replica_create_jobid,
    ns_base_create_process_name,
+    ns_base_copy_process_name,
    ns_base_convert_string_to_process_name,
    ns_replica_reserve_range,
    ns_replica_free_name,
@ -118,6 +119,7 @@ int mca_ns_replica_close(void)

 mca_ns_base_module_t* mca_ns_replica_init(bool *allow_multi_user_threads, bool *have_hidden_threads, int *priority)
 {
+
    /* If we're the seed, then we want to be selected, so do all the
       setup and return the module */

@ -142,6 +144,10 @@ mca_ns_base_module_t* mca_ns_replica_init(bool *allow_multi_user_threads, bool *

      OBJ_CONSTRUCT(&mca_ns_replica_name_tracker, ompi_list_t);

+      /* set my_replica to point to myself */
+
+      mca_ns_my_replica = mca_ns_replica.copy_process_name(&ompi_process_info.name);
+
      /* Return the module */

      initialized = true;
--- a/src/mca/oob/base/base.h
+++ b/src/mca/oob/base/base.h
@ -39,9 +39,15 @@ extern ompi_process_name_t mca_oob_name_self;
 * Other constants
 */
 /**
- * Recieve from any tag.
+ * Receive from any tag.
 */
 #define MCA_OOB_TAG_ANY 0
+/**
+ * Service tags
+ */
+#define MCA_OOB_TAG_NS  1
+#define MCA_OOB_TAG_GPR 2
+

 /*
 * OOB API 
--- a/src/mca/oob/base/oob_base_init.c
+++ b/src/mca/oob/base/oob_base_init.c
@ -10,6 +10,7 @@
 #include "util/output.h"
 #include "mca/mca.h"
 #include "mca/base/base.h"
+#include "mca/ns/base/base.h"
 #include "mca/pcm/pcm.h"
 #include "mca/oob/oob.h"
 #include "mca/oob/base/base.h"
@ -49,7 +50,6 @@ int mca_oob_base_init(bool *user_threads, bool *hidden_threads)
    ompi_process_name_t *self;

    /* setup local name */
-    OBJ_CONSTRUCT(&mca_oob_name_self, ompi_process_name_t);
    self = mca_pcm.pcm_self();
    if(NULL == self) {
        ompi_output(0, "mca_oob_base_init: could not get PCM self pointer");
@ -58,13 +58,12 @@ int mca_oob_base_init(bool *user_threads, bool *hidden_threads)
    mca_oob_name_self = *self;

    /* setup wildcard name */
-    OBJ_CONSTRUCT(&mca_oob_name_any, ompi_process_name_t);
-    mca_oob_name_any.cellid = -1;
-    mca_oob_name_any.jobid = -1;
-    mca_oob_name_any.vpid = -1;
+    mca_oob_name_any = *ompi_name_server.create_process_name(MCA_NS_BASE_CELLID_MAX,
+							    MCA_NS_BASE_JOBID_MAX,
+							    MCA_NS_BASE_VPID_MAX);

    /* setup seed daemons name */
-    OBJ_CONSTRUCT(&mca_oob_name_seed, ompi_process_name_t);
+    mca_oob_name_seed = *ompi_name_server.create_process_name(0,0,0);

    /* Traverse the list of available modules; call their init functions. */
    for (item = ompi_list_get_first(&mca_oob_base_components);