Add a new framework to ORTE for saving and recovering state information. Two components are included that use the db or dbm library for storing the data, with a distributed hash table component coming later.

Note that each of these components will only be selected if specifically requested - otherwise, a "NULL" component will be used.  The framework is only opened by the HNP and orteds, though neither is currently coded to save/restore state

This commit was SVN r22839.

configure.ac

@@ -525,7 +525,7 @@ AC_CHECK_HEADERS([alloca.h aio.h arpa/inet.h dirent.h \
     sys/types.h sys/uio.h net/uio.h sys/utsname.h sys/vfs.h sys/wait.h syslog.h \
     time.h termios.h ulimit.h unistd.h util.h utmp.h malloc.h \
     ifaddrs.h sys/sysctl.h crt_externs.h regex.h \
-    ioLib.h sockLib.h hostLib.h shlwapi.h sys/synch.h])
+    ioLib.h sockLib.h hostLib.h shlwapi.h sys/synch.h limits.h db.h ndbm.h])
 
 # Needed to work around Darwin requiring sys/socket.h for
 # net/if.h
@@ -656,7 +656,7 @@ OMPI_CHECK_FUNC_LIB([dirname], [gen])
 # Darwin doesn't need -lm, as it's a symlink to libSystem.dylib
 OMPI_CHECK_FUNC_LIB([ceil], [m])
 
-AC_CHECK_FUNCS([asprintf snprintf vasprintf vsnprintf openpty isatty getpwuid fork waitpid execve pipe ptsname setsid mmap tcgetpgrp posix_memalign strsignal sysconf syslog regcmp regexec regfree _NSGetEnviron socketpair strncpy_s _strdup usleep mkfifo])
+AC_CHECK_FUNCS([asprintf snprintf vasprintf vsnprintf openpty isatty getpwuid fork waitpid execve pipe ptsname setsid mmap tcgetpgrp posix_memalign strsignal sysconf syslog regcmp regexec regfree _NSGetEnviron socketpair strncpy_s _strdup usleep mkfifo dbopen dbm_open])
 
 # On some hosts, htonl is a define, so the AC_CHECK_FUNC will get
 # confused.  On others, it's in the standard library, but stubbed with

orte/mca/ess/base/ess_base_std_orted.c

@@ -59,6 +59,7 @@
 #include "orte/util/show_help.h"
 #include "orte/mca/notifier/base/base.h"
 #include "orte/mca/rmcast/base/base.h"
+#include "orte/mca/state/base/base.h"
 
 #include "orte/runtime/orte_cr.h"
 #include "orte/runtime/orte_wait.h"
@@ -388,6 +389,18 @@ int orte_ess_base_orted_setup(char **hosts)
         goto error;
     }
     
+    /* setup the state framework */
+    if (ORTE_SUCCESS != (ret = orte_state_base_open())) {
+        ORTE_ERROR_LOG(ret);
+        error = "orte_state_open";
+        goto error;
+    }
+    if (ORTE_SUCCESS != (ret = orte_state_base_select())) {
+        ORTE_ERROR_LOG(ret);
+        error = "orte_state_select";
+        goto error;
+    }
+    
     return ORTE_SUCCESS;
     
 error:
@@ -415,6 +428,7 @@ int orte_ess_base_orted_finalize(void)
         orte_grpcomm.onesided_barrier();
     }
     
+    orte_state_base_close();
     orte_notifier_base_close();
     
     orte_cr_finalize();

orte/mca/ess/hnp/ess_hnp_module.c

@@ -54,6 +54,7 @@
 #include "orte/mca/odls/base/base.h"
 #include "orte/mca/notifier/base/base.h"
 #include "orte/mca/rmcast/base/base.h"
+#include "orte/mca/state/base/base.h"
 
 #include "orte/mca/rmaps/base/base.h"
 #if OPAL_ENABLE_FT_CR == 1
@@ -547,6 +548,18 @@ static int rte_init(void)
         goto error;
     }
 
+    /* setup the state framework */
+    if (ORTE_SUCCESS != (ret = orte_state_base_open())) {
+        ORTE_ERROR_LOG(ret);
+        error = "orte_state_open";
+        goto error;
+    }
+    if (ORTE_SUCCESS != (ret = orte_state_base_select())) {
+        ORTE_ERROR_LOG(ret);
+        error = "orte_state_select";
+        goto error;
+    }
+    
     /* if a tool has launched us and is requesting event reports,
      * then set its contact info into the comm system
      */
@@ -607,6 +620,7 @@ static int rte_finalize(void)
     unlink(contact_path);
     free(contact_path);
     
+    orte_state_base_close();
     orte_notifier_base_close();
     
     orte_cr_finalize();

orte/mca/state/Makefile.am

@@ -0,0 +1,35 @@
+#
+# Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+# $COPYRIGHT$
+# 
+# Additional copyrights may follow
+# 
+# $HEADER$
+#
+
+AM_CPPFLAGS = $(LTDLINCL)
+
+# main library setup
+noinst_LTLIBRARIES = libmca_state.la
+libmca_state_la_SOURCES =
+
+# header setup
+nobase_orte_HEADERS =
+dist_pkgdata_DATA =
+
+# local files
+headers = state.h
+libmca_state_la_SOURCES += $(headers)
+
+# Conditionally install the header files
+if WANT_INSTALL_HEADERS
+nobase_orte_HEADERS += $(headers)
+ortedir = $(includedir)/openmpi/orte/mca/state
+else
+ortedir = $(includedir)
+endif
+
+include base/Makefile.am
+
+distclean-local:
+	rm -f base/static-components.h

orte/mca/state/base/Makefile.am

@@ -0,0 +1,16 @@
+#
+# Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+# $COPYRIGHT$
+# 
+# Additional copyrights may follow
+# 
+# $HEADER$
+#
+
+headers += \
+	base/base.h
+
+libmca_state_la_SOURCES += \
+	base/state_base_open.c \
+	base/state_base_close.c \
+	base/state_base_select.c

orte/mca/state/base/base.h

@@ -0,0 +1,54 @@
+/*
+ * Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+ * $COPYRIGHT$
+ * 
+ * Additional copyrights may follow
+ * 
+ * $HEADER$
+ */
+/** @file:
+ */
+
+#ifndef MCA_STATE_BASE_H
+#define MCA_STATE_BASE_H
+
+#include "orte_config.h"
+#include "orte/types.h"
+
+#include "opal/mca/mca.h"
+
+
+#include "orte/mca/state/state.h"
+
+BEGIN_C_DECLS
+
+/*
+ * Global functions for MCA overall collective open and close
+ */
+
+/**
+ * Open the state framework
+ */
+ORTE_DECLSPEC int orte_state_base_open(void);
+
+/**
+ * Select a state module
+ */
+ORTE_DECLSPEC int orte_state_base_select(void);
+
+/**
+ * Close the state framework
+ */
+ORTE_DECLSPEC int orte_state_base_close(void);
+
+/*
+ * The verbose channel for debug output
+ */
+ORTE_DECLSPEC extern int orte_state_base_output;
+
+ORTE_DECLSPEC extern opal_list_t orte_state_base_components_available;
+
+
+END_C_DECLS
+
+#endif

orte/mca/state/base/state_base_close.c

@@ -0,0 +1,50 @@
+/*
+ * Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+ * $COPYRIGHT$
+ * 
+ * Additional copyrights may follow
+ * 
+ * $HEADER$
+ */
+
+#include "orte_config.h"
+#include "orte/constants.h"
+
+#include <stdio.h>
+
+#include "opal/mca/mca.h"
+#include "opal/util/output.h"
+#include "opal/mca/base/base.h"
+#include "opal/mca/base/mca_base_component_repository.h"
+
+#include "orte/mca/state/base/base.h"
+
+extern opal_list_t orte_state_base_components_available;
+
+int
+orte_state_base_close(void)
+{
+    opal_list_item_t *item;
+    mca_base_component_list_item_t *cli;
+
+    if (NULL != orte_state.finalize) {
+        orte_state.finalize();
+    }
+    
+    /* unload all remaining components */
+    while (NULL != (item = opal_list_remove_first(&orte_state_base_components_available))) {
+        orte_state_base_component_t* component;
+        cli = (mca_base_component_list_item_t *) item;
+        component = (orte_state_base_component_t*) cli->cli_component;
+        opal_output_verbose(10, 0,
+                            "orte_state_base_close: module %s unloaded",
+                            component->base_version.mca_component_name);
+        mca_base_component_repository_release((mca_base_component_t *) component);
+        OBJ_RELEASE(item);
+    }
+    
+    OBJ_DESTRUCT(&orte_state_base_components_available);
+    opal_output_close(orte_state_base_output);
+    return ORTE_SUCCESS;
+}
+

orte/mca/state/base/state_base_open.c

@@ -0,0 +1,85 @@
+/*
+ * Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+ * $COPYRIGHT$
+ * 
+ * Additional copyrights may follow
+ * 
+ * $HEADER$
+ */
+
+
+#include "orte_config.h"
+#include "orte/constants.h"
+
+#include "opal/mca/mca.h"
+#include "opal/util/output.h"
+#include "opal/mca/base/base.h"
+#include "opal/dss/dss_types.h"
+
+#include "orte/mca/state/base/base.h"
+
+
+/*
+ * The following file was created by configure.  It contains extern
+ * statements and the definition of an array of pointers to each
+ * module's public mca_base_module_t struct.
+ */
+
+#include "orte/mca/state/base/static-components.h"
+
+opal_list_t orte_state_base_components_available;
+
+/* provide "NULL" functions */
+static int init(void);
+static int finalize(void);
+static int save(void *object, opal_data_type_t type);
+static int set_recover_source(orte_process_name_t *name);
+static int recover(void *object, opal_data_type_t type);
+
+orte_state_base_module_t orte_state = {
+    init,
+    finalize,
+    save,
+    set_recover_source,
+    recover
+};
+int orte_state_base_output;
+
+int
+orte_state_base_open(void)
+{
+    orte_state_base_output = opal_output_open(NULL);
+    
+    OBJ_CONSTRUCT(&orte_state_base_components_available, opal_list_t);
+
+    /* Open up all available components */
+    if (ORTE_SUCCESS != 
+        mca_base_components_open("state", orte_state_base_output, mca_state_base_static_components, 
+                                 &orte_state_base_components_available,
+                                 true)) {
+        return ORTE_ERROR;
+    }
+
+    return ORTE_SUCCESS;
+}
+
+static int init(void)
+{
+    return ORTE_SUCCESS;
+}
+static int finalize(void)
+{
+    return ORTE_SUCCESS;
+}
+static int save(void *object, opal_data_type_t type)
+{
+    return ORTE_SUCCESS;
+}
+static int set_recover_source(orte_process_name_t *name)
+{
+    return ORTE_SUCCESS;
+}
+static int recover(void *object, opal_data_type_t type)
+{
+    return ORTE_SUCCESS;
+}

orte/mca/state/base/state_base_select.c

@@ -0,0 +1,48 @@
+/*
+ * Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+ * $COPYRIGHT$
+ * 
+ * Additional copyrights may follow
+ * 
+ * $HEADER$
+ */
+
+#include "orte_config.h"
+#include "orte/constants.h"
+
+#include "opal/class/opal_list.h"
+#include "opal/mca/mca.h"
+#include "opal/mca/base/base.h"
+#include "opal/mca/base/mca_base_component_repository.h"
+
+#include "orte/mca/state/base/base.h"
+
+extern opal_list_t orte_state_base_components_available;
+
+int 
+orte_state_base_select(void)
+{
+    orte_state_base_component_t *best_component = NULL;
+    orte_state_base_module_t *best_module = NULL;
+
+    /*
+     * Select the best component
+     */
+    if( OPAL_SUCCESS != mca_base_select("state", orte_state_base_output,
+                                        &orte_state_base_components_available,
+                                        (mca_base_module_t **) &best_module,
+                                        (mca_base_component_t **) &best_component) ) {
+        /* It is okay to not select a component - default
+         * to using the base NULL component
+         */
+        return ORTE_SUCCESS;
+    }
+
+    /* Save and init the winner */
+    orte_state = *best_module;
+    if (NULL != orte_state.init) {
+        orte_state.init();
+    }
+
+    return ORTE_SUCCESS;
+}

orte/mca/state/db/Makefile.am

@@ -0,0 +1,34 @@
+#
+# Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+# $COPYRIGHT$
+# 
+# Additional copyrights may follow
+# 
+# $HEADER$
+#
+
+sources = \
+        state_db.h \
+        state_db_component.c \
+        state_db.c
+
+# Make the output library in this directory, and name it either
+# mca_<type>_<name>.la (for DSO builds) or libmca_<type>_<name>.la
+# (for static builds).
+
+if OMPI_BUILD_state_db_DSO
+component_noinst =
+component_install = mca_state_db.la
+else
+component_noinst = libmca_state_db.la
+component_install =
+endif
+
+mcacomponentdir = $(pkglibdir)
+mcacomponent_LTLIBRARIES = $(component_install)
+mca_state_db_la_SOURCES = $(sources)
+mca_state_db_la_LDFLAGS = -module -avoid-version
+
+noinst_LTLIBRARIES = $(component_noinst)
+libmca_state_db_la_SOURCES =$(sources)
+libmca_state_db_la_LDFLAGS = -module -avoid-version

orte/mca/state/db/configure.m4

@@ -0,0 +1,16 @@
+dnl -*- shell-script -*-
+dnl
+dnl Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+dnl $COPYRIGHT$
+dnl 
+dnl Additional copyrights may follow
+dnl 
+dnl $HEADER$
+dnl
+
+# MCA_state_db_CONFIG([action-if-found], [action-if-not-found])
+# -----------------------------------------------------------
+AC_DEFUN([MCA_state_db_CONFIG], [
+    # only build if db.h is found
+    AC_CHECK_HEADERS([db.h], [$1], [$2], [AC_INCLUDES_DEFAULT])
+])dnl

orte/mca/state/db/configure.params

@@ -0,0 +1,11 @@
+# -*- shell-script -*-
+#
+# Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+# $COPYRIGHT$
+# 
+# Additional copyrights may follow
+# 
+# $HEADER$
+#
+
+PARAM_CONFIG_FILES="Makefile"

orte/mca/state/db/state_db.c

@@ -0,0 +1,248 @@
+/*
+ * Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+ * $COPYRIGHT$
+ * 
+ * Additional copyrights may follow
+ * 
+ * $HEADER$
+ *
+ */
+
+#include "orte_config.h"
+#include "orte/constants.h"
+
+#include <string.h>
+#include <sys/types.h>
+#ifdef HAVE_LIMITS_H
+#include <limits.h>
+#endif
+#include <stdio.h>
+#ifdef HAVE_FCNTL_H
+#include <fcntl.h>
+#endif
+#ifdef HAVE_UNISTD_H
+#include <unistd.h>
+#endif
+#include <db.h>
+
+#include "opal/dss/dss_types.h"
+#include "opal/util/os_dirpath.h"
+#include "opal/util/os_path.h"
+#include "opal/util/output.h"
+#include "opal/util/malloc.h"
+#include "opal/util/basename.h"
+#include "opal/mca/pstat/base/base.h"
+#include "opal/mca/paffinity/base/base.h"
+#include "opal/mca/sysinfo/base/base.h"
+
+#include "orte/util/show_help.h"
+#include "orte/mca/errmgr/base/base.h"
+#include "orte/runtime/orte_globals.h"
+
+#include "state_db.h"
+
+static int init(void);
+static int finalize(void);
+static int save(void *object, opal_data_type_t type);
+static int set_recover_source(orte_process_name_t *name);
+static int recover(void *object, opal_data_type_t type);
+
+orte_state_base_module_t orte_state_db_module = {
+    init,
+    finalize,
+    save,
+    set_recover_source,
+    recover
+};
+
+/* local variables */
+static DB *save_db=NULL, *recover_db=NULL;
+
+static int init(void)
+{
+    char *path, *name;
+    
+    /* setup the database */
+    if (ORTE_SUCCESS != opal_os_dirpath_create(orte_state_db_directory, S_IRWXU)) {
+        orte_show_help("help-state-db.txt", "cannot-create-dir", true,
+                       orte_state_db_directory);
+        return ORTE_ERR_FILE_OPEN_FAILURE;
+    }
+    orte_util_convert_process_name_to_string(&name, ORTE_PROC_MY_NAME);
+    path = opal_os_path(false, orte_state_db_directory, name, NULL);
+    free(name);
+    if (NULL == (save_db = dbopen(path, O_CREAT | O_RDWR | O_TRUNC, S_IRWXU, DB_HASH, NULL))) {
+        orte_show_help("help-state-db.txt", "cannot-create-db", true, path);
+        free(path);
+        return ORTE_ERR_FILE_OPEN_FAILURE;
+    }
+    free(path);
+    
+    return ORTE_SUCCESS;
+}
+
+static int finalize(void)
+{
+    /* if we are normally terminating, remove the recovery file */
+    
+    return ORTE_SUCCESS;
+}
+
+static int save(void *object, opal_data_type_t type)
+{
+    DBT key, data;
+    opal_buffer_t buf;
+    orte_job_t *jdata;
+    orte_proc_t *proc;
+    char *name;
+    int rc=ORTE_SUCCESS, size;
+    
+    /* construct the buffer we will use for packing the data */
+    OBJ_CONSTRUCT(&buf, opal_buffer_t);
+    key.data = NULL;
+    data.data = NULL;
+    
+    switch (type) {
+        case ORTE_JOB:
+            jdata = (orte_job_t*)object;
+            opal_dss.pack(&buf, &jdata->state, 1, ORTE_JOB_STATE_T);
+            asprintf((char**)&key.data, "JOB:%s", ORTE_JOBID_PRINT(jdata->jobid));
+            key.size = strlen(key.data);
+            break;
+        case ORTE_PROC:
+            proc = (orte_proc_t*)object;
+            opal_dss.pack(&buf, &proc->state, 1, ORTE_PROC_STATE_T);
+            orte_util_convert_process_name_to_string(&name, &proc->name);
+            asprintf((char**)&key.data, "PROC:%s", name);
+            free(name);
+            key.size = strlen(key.data);
+            break;
+        default:
+            orte_show_help("help-state-db.txt", "unrecognized-type", true, type);
+            rc = ORTE_ERR_BAD_PARAM;
+            goto cleanup;
+            break;
+    }
+    
+    /* unload the data */
+    opal_dss.unload(&buf, (void**)&data.data, &size);
+    data.size = size;
+    OBJ_DESTRUCT(&buf);
+    
+    /* put the info into the db */
+    if (0 > save_db->put(save_db, &key, &data, 0)) {
+        orte_show_help("help-state-db.txt", "error-writing-db", true, (char*)key.data, strerror(errno));
+        rc = ORTE_ERR_FILE_WRITE_FAILURE;
+    }
+    /* sync it to force it to disk */
+    if (0 > save_db->sync(save_db, 0)) {
+        orte_show_help("help-state-db.txt", "error-syncing-db", true, (char*)key.data, strerror(errno));
+        rc = ORTE_ERR_FILE_WRITE_FAILURE;
+    }
+    
+cleanup:
+    /* cleanup */
+    if (NULL != key.data) {
+        free(key.data);
+    }
+    if (NULL != data.data) {
+        free(data.data);
+    }
+    return rc;
+}
+
+static int set_recover_source(orte_process_name_t *name)
+{
+    char *path, *pname;
+    int rc=ORTE_SUCCESS;
+    
+    /* setup the database */
+    orte_util_convert_process_name_to_string(&pname, name);
+    path = opal_os_path(false, orte_state_db_directory, pname, NULL);
+    free(pname);
+    if (NULL == (recover_db = dbopen(path, O_RDONLY, S_IRWXU, DB_HASH, NULL))) {
+        orte_show_help("help-state-db.txt", "cannot-open-db", true, path);
+        free(path);
+        return ORTE_ERR_FILE_OPEN_FAILURE;
+    }
+    free(path);
+    
+    return rc;
+}
+
+static int recover(void *object, opal_data_type_t type)
+{
+    DBT key, data;
+    opal_buffer_t buf;
+    orte_job_t *jdata;
+    orte_proc_t *proc;
+    char *name;
+    int rc=ORTE_SUCCESS;
+    int32_t n;
+    orte_job_state_t *jstate;
+    orte_proc_state_t *pstate;
+    
+    if (NULL == recover_db) {
+        orte_show_help("help-state-db.txt", "recover-source-undef", true);
+        rc = ORTE_ERR_NOT_FOUND;
+    }
+    
+    /* construct the buffer we will use for unpacking the data */
+    OBJ_CONSTRUCT(&buf, opal_buffer_t);
+    key.data = NULL;
+    data.data = NULL;
+    
+    switch (type) {
+        case ORTE_JOB:
+            jdata = (orte_job_t*)object;
+            asprintf((char**)&key.data, "JOB:%s", ORTE_JOBID_PRINT(jdata->jobid));
+            key.size = strlen(key.data);
+            break;
+        case ORTE_PROC:
+            proc = (orte_proc_t*)object;
+            orte_util_convert_process_name_to_string(&name, &proc->name);
+            asprintf((char**)&key.data, "PROC:%s", name);
+            free(name);
+            key.size = strlen(key.data);
+            break;
+        default:
+            orte_show_help("help-state-db.txt", "unrecognized-type", true, type);
+            rc = ORTE_ERR_BAD_PARAM;
+            goto cleanup;
+            break;
+    }
+    
+    /* get the specified data */
+    if (0 > recover_db->get(recover_db, &key, &data, 0)) {
+        orte_show_help("help-state-db.txt", "error-reading-db", true, (char*)key.data, strerror(errno));
+        rc = ORTE_ERR_FILE_READ_FAILURE;
+        goto cleanup;
+    }
+    
+    /* populate the recovered info */
+    opal_dss.load(&buf, data.data, data.size);
+    switch (type) {
+        case ORTE_JOB:
+            n=1;
+            opal_dss.unpack(&buf, &jstate, &n, ORTE_JOB_STATE_T);
+            jdata->state = *jstate;
+            break;
+        case ORTE_PROC:
+            n=1;
+            opal_dss.unpack(&buf, &pstate, &n, ORTE_PROC_STATE_T);
+            proc->state = *pstate;
+            break;
+        default:
+            break;
+    }
+    
+cleanup:
+    if (NULL != key.data) {
+        free(key.data);
+    }
+    if (NULL != data.data) {
+        free(data.data);
+    }
+    OBJ_DESTRUCT(&buf);
+    return rc;
+}

orte/mca/state/db/state_db.h

@@ -0,0 +1,31 @@
+/*
+ * Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+ * $COPYRIGHT$
+ * 
+ * Additional copyrights may follow
+ * 
+ * $HEADER$
+ */
+
+#ifndef ORTE_STATE_DB_H
+#define ORTE_STATE_DB_H
+
+#include "orte/mca/state/state.h"
+
+BEGIN_C_DECLS
+
+/*
+ * Module open / close
+ */
+int orte_state_db_component_open(void);
+int orte_state_db_component_close(void);
+int orte_state_db_component_query(mca_base_module_t **module, int *priority);
+
+
+ORTE_MODULE_DECLSPEC extern orte_state_base_component_t mca_state_db_component;
+ORTE_DECLSPEC extern orte_state_base_module_t orte_state_db_module;
+extern char *orte_state_db_directory;
+
+END_C_DECLS
+
+#endif /* ORTE_STATE_DB_H */

orte/mca/state/db/state_db_component.c

@@ -0,0 +1,101 @@
+/*
+ * Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+ * $COPYRIGHT$
+ * 
+ * Additional copyrights may follow
+ * 
+ * $HEADER$
+ *
+ * These symbols are in a file by themselves to provide nice linker
+ * semantics.  Since linkers generally pull in symbols by object
+ * files, keeping these symbols as the only symbols in this file
+ * prevents utility programs such as "ompi_info" from having to import
+ * entire components just to query their version and parameters.
+ */
+
+#include "orte_config.h"
+#include "orte/constants.h"
+
+#include "opal/mca/base/base.h"
+#include "opal/mca/base/mca_base_param.h"
+
+#include "orte/util/proc_info.h"
+
+#include "orte/mca/state/state.h"
+#include "orte/mca/state/base/base.h"
+#include "state_db.h"
+
+extern orte_state_base_module_t orte_state_db_module;
+char *orte_state_db_filename;
+
+/*
+ * Instantiate the public struct with all of our public information
+ * and pointers to our public functions in it
+ */
+orte_state_base_component_t mca_state_db_component = {
+    {
+        ORTE_STATE_BASE_VERSION_1_0_0,
+
+        /* Component name and version */
+        "db",
+        ORTE_MAJOR_VERSION,
+        ORTE_MINOR_VERSION,
+        ORTE_RELEASE_VERSION,
+
+        /* Component open and close functions */
+        orte_state_db_component_open,
+        orte_state_db_component_close,
+        orte_state_db_component_query
+    },
+    {
+        /* The component is checkpoint ready */
+        MCA_BASE_METADATA_PARAM_CHECKPOINT
+    }
+};
+
+
+int
+orte_state_db_component_open(void)
+{
+    return ORTE_SUCCESS;
+}
+
+
+int orte_state_db_component_query(mca_base_module_t **module, int *priority)
+{
+
+    /* we are the file module - we need to be selected
+     * IFF we are requested
+     */
+    bool is_required = false;
+    mca_base_component_t *c = &mca_state_db_component.base_version;
+
+    /* retrieve the name of the file to be used */
+    mca_base_param_reg_string(c, "dir",
+                              "Name of directory to be used for storing and recovering state information",
+                              false, false, NULL, &orte_state_db_directory);
+    
+    mca_base_is_component_required(&orte_state_base_components_available,
+                                   &mca_state_db_component.base_version,
+                                   true,
+                                   &is_required);
+    
+    if( is_required || NULL != orte_state_db_directory) {
+        *priority = 1000;
+        *module = (mca_base_module_t*)&orte_state_db_module;
+        return ORTE_SUCCESS;
+    }
+    
+    
+    *priority = 0;
+    *module = NULL;
+    return ORTE_ERROR;
+}
+
+
+int
+orte_state_db_component_close(void)
+{
+    return ORTE_SUCCESS;
+}
+

orte/mca/state/dbm/Makefile.am

@@ -0,0 +1,34 @@
+#
+# Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+# $COPYRIGHT$
+# 
+# Additional copyrights may follow
+# 
+# $HEADER$
+#
+
+sources = \
+        state_dbm.h \
+        state_dbm_component.c \
+        state_dbm.c
+
+# Make the output library in this directory, and name it either
+# mca_<type>_<name>.la (for DSO builds) or libmca_<type>_<name>.la
+# (for static builds).
+
+if OMPI_BUILD_state_dbm_DSO
+component_noinst =
+component_install = mca_state_dbm.la
+else
+component_noinst = libmca_state_dbm.la
+component_install =
+endif
+
+mcacomponentdir = $(pkglibdir)
+mcacomponent_LTLIBRARIES = $(component_install)
+mca_state_dbm_la_SOURCES = $(sources)
+mca_state_dbm_la_LDFLAGS = -module -avoid-version -ldbm
+
+noinst_LTLIBRARIES = $(component_noinst)
+libmca_state_dbm_la_SOURCES =$(sources)
+libmca_state_dbm_la_LDFLAGS = -module -avoid-version -ldbm

orte/mca/state/dbm/configure.m4

@@ -0,0 +1,16 @@
+dnl -*- shell-script -*-
+dnl
+dnl Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+dnl $COPYRIGHT$
+dnl 
+dnl Additional copyrights may follow
+dnl 
+dnl $HEADER$
+dnl
+
+# MCA_state_dbm_CONFIG([action-if-found], [action-if-not-found])
+# -----------------------------------------------------------
+AC_DEFUN([MCA_state_dbm_CONFIG], [
+    # only build if ndbm.h is found
+    AC_CHECK_HEADERS([ndbm.h], [$1], [$2], [AC_INCLUDES_DEFAULT])
+])dnl

orte/mca/state/dbm/configure.params

@@ -0,0 +1,11 @@
+# -*- shell-script -*-
+#
+# Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+# $COPYRIGHT$
+# 
+# Additional copyrights may follow
+# 
+# $HEADER$
+#
+
+PARAM_CONFIG_FILES="Makefile"

orte/mca/state/dbm/state_dbm.c

@@ -0,0 +1,244 @@
+/*
+ * Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+ * $COPYRIGHT$
+ * 
+ * Additional copyrights may follow
+ * 
+ * $HEADER$
+ *
+ */
+
+#include "orte_config.h"
+#include "orte/constants.h"
+
+#include <string.h>
+#include <sys/types.h>
+#ifdef HAVE_LIMITS_H
+#include <limits.h>
+#endif
+#include <stdio.h>
+#ifdef HAVE_FCNTL_H
+#include <fcntl.h>
+#endif
+#ifdef HAVE_UNISTD_H
+#include <unistd.h>
+#endif
+#include <ndbm.h>
+
+#include "opal/dss/dss_types.h"
+#include "opal/util/os_dirpath.h"
+#include "opal/util/os_path.h"
+#include "opal/util/output.h"
+#include "opal/util/malloc.h"
+#include "opal/util/basename.h"
+#include "opal/mca/pstat/base/base.h"
+#include "opal/mca/paffinity/base/base.h"
+#include "opal/mca/sysinfo/base/base.h"
+
+#include "orte/util/show_help.h"
+#include "orte/mca/errmgr/base/base.h"
+#include "orte/runtime/orte_globals.h"
+
+#include "state_dbm.h"
+
+static int init(void);
+static int finalize(void);
+static int save(void *object, opal_data_type_t type);
+static int set_recover_source(orte_process_name_t *name);
+static int recover(void *object, opal_data_type_t type);
+
+orte_state_base_module_t orte_state_dbm_module = {
+    init,
+    finalize,
+    save,
+    set_recover_source,
+    recover
+};
+
+/* local variables */
+static DBM *save_dbm=NULL, *recover_dbm=NULL;
+
+static int init(void)
+{
+    char *path, *name;
+    
+    /* setup the database */
+    if (ORTE_SUCCESS != opal_os_dirpath_create(orte_state_dbm_directory, S_IRWXU)) {
+        orte_show_help("help-state-dbm.txt", "cannot-create-dir", true,
+                       orte_state_dbm_directory);
+        return ORTE_ERR_FILE_OPEN_FAILURE;
+    }
+    orte_util_convert_process_name_to_string(&name, ORTE_PROC_MY_NAME);
+    path = opal_os_path(false, orte_state_dbm_directory, name, NULL);
+    free(name);
+    if (NULL == (save_dbm = dbm_open(path, O_CREAT | O_RDWR | O_TRUNC, S_IRWXU))) {
+        orte_show_help("help-state-dbm.txt", "cannot-create-dbm", true, path);
+        free(path);
+        return ORTE_ERR_FILE_OPEN_FAILURE;
+    }
+    free(path);
+    
+    return ORTE_SUCCESS;
+}
+
+static int finalize(void)
+{
+    /* if we are normally terminating, remove the recovery file */
+    
+    return ORTE_SUCCESS;
+}
+
+static int save(void *object, opal_data_type_t type)
+{
+    datum key, data;
+    opal_buffer_t buf;
+    orte_job_t *jdata;
+    orte_proc_t *proc;
+    char *name;
+    int rc=ORTE_SUCCESS, size;
+    
+    /* construct the buffer we will use for packing the data */
+    OBJ_CONSTRUCT(&buf, opal_buffer_t);
+    key.dptr = NULL;
+    data.dptr = NULL;
+    
+    switch (type) {
+        case ORTE_JOB:
+            jdata = (orte_job_t*)object;
+            opal_dss.pack(&buf, &jdata->state, 1, ORTE_JOB_STATE_T);
+            asprintf((char**)&key.dptr, "JOB:%s", ORTE_JOBID_PRINT(jdata->jobid));
+            key.dsize = strlen(key.dptr);
+            break;
+        case ORTE_PROC:
+            proc = (orte_proc_t*)object;
+            opal_dss.pack(&buf, &proc->state, 1, ORTE_PROC_STATE_T);
+            orte_util_convert_process_name_to_string(&name, &proc->name);
+            asprintf((char**)&key.dptr, "PROC:%s", name);
+            free(name);
+            key.dsize = strlen(key.dptr);
+            break;
+        default:
+            orte_show_help("help-state-dbm.txt", "unrecognized-type", true, type);
+            rc = ORTE_ERR_BAD_PARAM;
+            goto cleanup;
+            break;
+    }
+    
+    /* unload the data */
+    opal_dss.unload(&buf, (void**)&data.dptr, &size);
+    data.dsize = size;
+    OBJ_DESTRUCT(&buf);
+    
+    /* put the info into the dbm */
+    if (0 > dbm_store(save_dbm, key, data, DBM_REPLACE)) {
+        orte_show_help("help-state-dbm.txt", "error-writing-dbm", true, (char*)key.dptr, strerror(errno));
+        rc = ORTE_ERR_FILE_WRITE_FAILURE;
+    }
+    
+cleanup:
+    /* cleanup */
+    if (NULL != key.dptr) {
+        free(key.dptr);
+    }
+    if (NULL != data.dptr) {
+        free(data.dptr);
+    }
+    return rc;
+}
+
+static int set_recover_source(orte_process_name_t *name)
+{
+    char *path, *pname;
+    int rc=ORTE_SUCCESS;
+    
+    /* setup the database */
+    orte_util_convert_process_name_to_string(&pname, name);
+    path = opal_os_path(false, orte_state_dbm_directory, pname, NULL);
+    free(pname);
+    if (NULL == (recover_dbm = dbm_open(path, O_RDONLY, S_IRWXU))) {
+        orte_show_help("help-state-dbm.txt", "cannot-open-dbm", true, path);
+        free(path);
+        return ORTE_ERR_FILE_OPEN_FAILURE;
+    }
+    free(path);
+    
+    return rc;
+}
+
+static int recover(void *object, opal_data_type_t type)
+{
+    datum key, data;
+    opal_buffer_t buf;
+    orte_job_t *jdata;
+    orte_proc_t *proc;
+    char *name;
+    int rc=ORTE_SUCCESS;
+    int32_t n;
+    orte_job_state_t *jstate;
+    orte_proc_state_t *pstate;
+    
+    if (NULL == recover_dbm) {
+        orte_show_help("help-state-dbm.txt", "recover-source-undef", true);
+        rc = ORTE_ERR_NOT_FOUND;
+    }
+    
+    /* construct the buffer we will use for unpacking the data */
+    OBJ_CONSTRUCT(&buf, opal_buffer_t);
+    key.dptr = NULL;
+    data.dptr = NULL;
+    
+    switch (type) {
+        case ORTE_JOB:
+            jdata = (orte_job_t*)object;
+            asprintf((char**)&key.dptr, "JOB:%s", ORTE_JOBID_PRINT(jdata->jobid));
+            key.dsize = strlen(key.dptr);
+            break;
+        case ORTE_PROC:
+            proc = (orte_proc_t*)object;
+            orte_util_convert_process_name_to_string(&name, &proc->name);
+            asprintf((char**)&key.dptr, "PROC:%s", name);
+            free(name);
+            key.dsize = strlen(key.dptr);
+            break;
+        default:
+            orte_show_help("help-state-dbm.txt", "unrecognized-type", true, type);
+            rc = ORTE_ERR_BAD_PARAM;
+            goto cleanup;
+            break;
+    }
+    
+    /* get the specified data */
+    data = dbm_fetch(recover_dbm, key);
+    if (NULL == data.dptr) {
+        orte_show_help("help-state-dbm.txt", "error-reading-dbm", true, (char*)key.dptr, strerror(errno));
+        rc = ORTE_ERR_FILE_READ_FAILURE;
+        goto cleanup;
+    }
+    
+    /* populate the recovered info */
+    opal_dss.load(&buf, data.dptr, data.dsize);
+    switch (type) {
+        case ORTE_JOB:
+            n=1;
+            opal_dss.unpack(&buf, &jstate, &n, ORTE_JOB_STATE_T);
+            jdata->state = *jstate;
+            break;
+        case ORTE_PROC:
+            n=1;
+            opal_dss.unpack(&buf, &pstate, &n, ORTE_PROC_STATE_T);
+            proc->state = *pstate;
+            break;
+        default:
+            break;
+    }
+    
+cleanup:
+    if (NULL != key.dptr) {
+        free(key.dptr);
+    }
+    if (NULL != data.dptr) {
+        free(data.dptr);
+    }
+    OBJ_DESTRUCT(&buf);
+    return rc;
+}

orte/mca/state/dbm/state_dbm.h

@@ -0,0 +1,31 @@
+/*
+ * Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+ * $COPYRIGHT$
+ * 
+ * Additional copyrights may follow
+ * 
+ * $HEADER$
+ */
+
+#ifndef ORTE_STATE_DBM_H
+#define ORTE_STATE_DBM_H
+
+#include "orte/mca/state/state.h"
+
+BEGIN_C_DECLS
+
+/*
+ * Module open / close
+ */
+int orte_state_dbm_component_open(void);
+int orte_state_dbm_component_close(void);
+int orte_state_dbm_component_query(mca_base_module_t **module, int *priority);
+
+
+ORTE_MODULE_DECLSPEC extern orte_state_base_component_t mca_state_dbm_component;
+ORTE_DECLSPEC extern orte_state_base_module_t orte_state_dbm_module;
+extern char *orte_state_dbm_directory;
+
+END_C_DECLS
+
+#endif /* ORTE_STATE_DBM_H */

orte/mca/state/dbm/state_dbm_component.c

@@ -0,0 +1,101 @@
+/*
+ * Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+ * $COPYRIGHT$
+ * 
+ * Additional copyrights may follow
+ * 
+ * $HEADER$
+ *
+ * These symbols are in a file by themselves to provide nice linker
+ * semantics.  Since linkers generally pull in symbols by object
+ * files, keeping these symbols as the only symbols in this file
+ * prevents utility programs such as "ompi_info" from having to import
+ * entire components just to query their version and parameters.
+ */
+
+#include "orte_config.h"
+#include "orte/constants.h"
+
+#include "opal/mca/base/base.h"
+#include "opal/mca/base/mca_base_param.h"
+
+#include "orte/util/proc_info.h"
+
+#include "orte/mca/state/state.h"
+#include "orte/mca/state/base/base.h"
+#include "state_dbm.h"
+
+extern orte_state_base_module_t orte_state_dbm_module;
+char *orte_state_dbm_filename;
+
+/*
+ * Instantiate the public struct with all of our public information
+ * and pointers to our public functions in it
+ */
+orte_state_base_component_t mca_state_dbm_component = {
+    {
+        ORTE_STATE_BASE_VERSION_1_0_0,
+
+        /* Component name and version */
+        "dbm",
+        ORTE_MAJOR_VERSION,
+        ORTE_MINOR_VERSION,
+        ORTE_RELEASE_VERSION,
+
+        /* Component open and close functions */
+        orte_state_dbm_component_open,
+        orte_state_dbm_component_close,
+        orte_state_dbm_component_query
+    },
+    {
+        /* The component is checkpoint ready */
+        MCA_BASE_METADATA_PARAM_CHECKPOINT
+    }
+};
+
+
+int
+orte_state_dbm_component_open(void)
+{
+    return ORTE_SUCCESS;
+}
+
+
+int orte_state_dbm_component_query(mca_base_module_t **module, int *priority)
+{
+
+    /* we are the file module - we need to be selected
+     * IFF we are requested
+     */
+    bool is_required = false;
+    mca_base_component_t *c = &mca_state_dbm_component.base_version;
+
+    /* retrieve the name of the file to be used */
+    mca_base_param_reg_string(c, "dir",
+                              "Name of directory to be used for storing and recovering state information",
+                              false, false, NULL, &orte_state_dbm_directory);
+    
+    mca_base_is_component_required(&orte_state_base_components_available,
+                                   &mca_state_dbm_component.base_version,
+                                   true,
+                                   &is_required);
+    
+    if( is_required || NULL != orte_state_dbm_directory) {
+        *priority = 1000;
+        *module = (mca_base_module_t*)&orte_state_dbm_module;
+        return ORTE_SUCCESS;
+    }
+    
+    
+    *priority = 0;
+    *module = NULL;
+    return ORTE_ERROR;
+}
+
+
+int
+orte_state_dbm_component_close(void)
+{
+    return ORTE_SUCCESS;
+}
+

orte/mca/state/state.h

@@ -0,0 +1,91 @@
+/*
+ * Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved. 
+ * $COPYRIGHT$
+ * 
+ * Additional copyrights may follow
+ * 
+ * $HEADER$
+ */
+/** @file:
+ *
+ * The OpenRTE State Save/Recovery Service
+ *
+ */
+
+#ifndef ORTE_STATE_H
+#define ORTE_STATE_H
+
+#include "orte_config.h"
+#include "orte/types.h"
+
+#include "opal/mca/mca.h"
+#include "opal/dss/dss_types.h"
+
+BEGIN_C_DECLS
+
+/*
+ * API functions
+ */
+
+/*
+ * Initialize the module
+ */
+typedef int (*orte_state_base_module_init_fn_t)(void);
+
+/*
+ * Finalize the module
+ */
+typedef int (*orte_state_base_module_finalize_fn_t)(void);
+
+/*
+ * Save the state of the provided object
+ */
+typedef int (*orte_state_base_module_save_fn_t)(void *object, opal_data_type_t type);
+
+/*
+ * Set the source for recovering state info
+ */
+typedef int (*orte_state_base_module_set_recover_source_fn_t)(orte_process_name_t *name);
+
+/*
+ * Recover the state of an object
+ */
+typedef int (*orte_state_base_module_recover_fn_t)(void *object, opal_data_type_t type);
+
+
+/*
+ * the standard module data structure
+ */
+struct orte_state_base_module_1_0_0_t {
+    orte_state_base_module_init_fn_t                init;
+    orte_state_base_module_finalize_fn_t            finalize;
+    orte_state_base_module_save_fn_t                save;
+    orte_state_base_module_set_recover_source_fn_t  set_recover_source;
+    orte_state_base_module_recover_fn_t             recover;
+};
+typedef struct orte_state_base_module_1_0_0_t orte_state_base_module_1_0_0_t;
+typedef struct orte_state_base_module_1_0_0_t orte_state_base_module_t;
+
+/*
+ * the standard component data structure
+ */
+struct orte_state_base_component_1_0_0_t {
+    mca_base_component_t base_version;
+    mca_base_component_data_t base_data;
+};
+typedef struct orte_state_base_component_1_0_0_t orte_state_base_component_1_0_0_t;
+typedef struct orte_state_base_component_1_0_0_t orte_state_base_component_t;
+
+/*
+ * Macro for use in components that are of type state
+ */
+#define ORTE_STATE_BASE_VERSION_1_0_0 \
+  MCA_BASE_VERSION_2_0_0, \
+  "state", 1, 0, 0
+
+/* Global structure for accessing STATE functions */
+ORTE_DECLSPEC extern orte_state_base_module_t orte_state;  /* holds selected module's function pointers */
+
+END_C_DECLS
+
+#endif