Add keepalive support to the TCP OOB component

orte/mca/oob/tcp/oob_tcp_common.c

@@ -77,6 +77,75 @@
  * Set socket buffering
  */
 
+static void set_keepalive(int sd)
+{
+    int option;
+    socklen_t optlen;
+
+    /* see if the keepalive option is available */
+    optlen = sizeof(option);
+    if (getsockopt(sd, SOL_SOCKET, SO_KEEPALIVE, &option, &optlen) < 0) {
+        /* not available, so just return */
+        return;
+    }
+
+    /* Set the option active */
+    option = 1;
+    if (setsockopt(sd, SOL_SOCKET, SO_KEEPALIVE, &option, optlen) < 0) {
+        opal_output(0, "[%s:%d] setsockopt(SO_KEEPALIVE) failed: %s (%d)", 
+                    __FILE__, __LINE__, 
+                    strerror(opal_socket_errno),
+                    opal_socket_errno);
+        return;
+    }
+#if defined(TCP_KEEPALIVE)
+    /* set the idle time */
+    if (setsockopt(sd, IPPROTO_TCP, TCP_KEEPALIVE,
+                   &mca_oob_tcp_component.keepalive_time,
+                   sizeof(mca_oob_tcp_component.keepalive_time)) < 0) {
+        opal_output(0, "[%s:%d] setsockopt(TCP_KEEPALIVE) failed: %s (%d)", 
+                    __FILE__, __LINE__, 
+                    strerror(opal_socket_errno),
+                    opal_socket_errno);
+        return;
+    }
+#elif defined(TCP_KEEPIDLE)
+    /* set the idle time */
+    if (setsockopt(sd, IPPROTO_TCP, TCP_KEEPIDLE,
+                   &mca_oob_tcp_component.keepalive_time,
+                   sizeof(mca_oob_tcp_component.keepalive_time)) < 0) {
+        opal_output(0, "[%s:%d] setsockopt(TCP_KEEPIDLE) failed: %s (%d)", 
+                    __FILE__, __LINE__, 
+                    strerror(opal_socket_errno),
+                    opal_socket_errno);
+        return;
+    }
+#endif  // TCP_KEEPIDLE
+#if defined(TCP_KEEPINTVL)
+    /* set the keepalive interval */
+    if (setsockopt(sd, IPPROTO_TCP, TCP_KEEPINTVL,
+                   &mca_oob_tcp_component.keepalive_intvl,
+                   sizeof(mca_oob_tcp_component.keepalive_intvl)) < 0) {
+        opal_output(0, "[%s:%d] setsockopt(TCP_KEEPINTVL) failed: %s (%d)", 
+                    __FILE__, __LINE__, 
+                    strerror(opal_socket_errno),
+                    opal_socket_errno);
+        return;
+    }
+#endif  // TCP_KEEPINTVL
+#if defined(TCP_KEEPCNT)
+    /* set the miss rate */
+    if (setsockopt(sd, IPPROTO_TCP, TCP_KEEPCNT,
+                   &mca_oob_tcp_component.keepalive_probes,
+                   sizeof(mca_oob_tcp_component.keepalive_probes)) < 0) {
+        opal_output(0, "[%s:%d] setsockopt(TCP_KEEPCNT) failed: %s (%d)", 
+                    __FILE__, __LINE__, 
+                    strerror(opal_socket_errno),
+                    opal_socket_errno);
+    }
+#endif  // TCP_KEEPCNT
+}
+
 void orte_oob_tcp_set_socket_options(int sd)
 {
 #if defined(TCP_NODELAY)
@@ -108,6 +177,11 @@ void orte_oob_tcp_set_socket_options(int sd)
                     opal_socket_errno);
     }
 #endif
+#if defined(SO_KEEPALIVE)
+    if (0 < mca_oob_tcp_component.keepalive_time) {
+        set_keepalive(sd);
+    }
+#endif  // SO_KEEPALIVE
 }
 
 mca_oob_tcp_peer_t* mca_oob_tcp_peer_lookup(const orte_process_name_t *name)

orte/mca/oob/tcp/oob_tcp_component.c

@@ -404,6 +404,28 @@ static int tcp_component_register(void)
                                           &mca_oob_tcp_component.disable_ipv6_family);
 #endif
 
+    mca_oob_tcp_component.keepalive_time = 10;
+    (void)mca_base_component_var_register(component, "keepalive_time",
+                                          "Idle time in seconds before starting to send keepalives (num <= 0 ----> disable keepalive)",
+                                          MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
+                                          OPAL_INFO_LVL_9,
+                                          MCA_BASE_VAR_SCOPE_READONLY,
+                                          &mca_oob_tcp_component.keepalive_time);
+
+    mca_oob_tcp_component.keepalive_intvl = 5;
+    (void)mca_base_component_var_register(component, "keepalive_intvl",
+                                          "Time between keepalives, in seconds",
+                                          MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
+                                          OPAL_INFO_LVL_9,
+                                          MCA_BASE_VAR_SCOPE_READONLY,
+                                          &mca_oob_tcp_component.keepalive_intvl);
+    mca_oob_tcp_component.keepalive_probes = 3;
+    (void)mca_base_component_var_register(component, "keepalive_probes",
+                                          "Number of keepalives that can be missed before declaring error",
+                                          MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
+                                          OPAL_INFO_LVL_9,
+                                          MCA_BASE_VAR_SCOPE_READONLY,
+                                          &mca_oob_tcp_component.keepalive_probes);
     return ORTE_SUCCESS;
 }
 

orte/mca/oob/tcp/oob_tcp_component.h

@@ -77,6 +77,9 @@ typedef struct {
     bool               listen_thread_active;
     struct timeval     listen_thread_tv;       /**< Timeout when using listen thread */
     int                stop_thread[2];         /**< pipe used to exit the listen thread */
+    int                keepalive_probes;       /**< number of keepalives that can be missed before declaring error */
+    int                keepalive_time;         /**< idle time in seconds before starting to send keepalives */
+    int                keepalive_intvl;        /**< time between keepalives, in seconds */
 } mca_oob_tcp_component_t;
 
 ORTE_MODULE_DECLSPEC extern mca_oob_tcp_component_t mca_oob_tcp_component;