Merge branch 'master' of github.com:ClusterLabs/pacemaker

cib/notify.c

@@ -116,28 +116,6 @@ cib_notify_send_one(gpointer key, gpointer value, gpointer user_data)
     if (do_send) {
         switch (client->kind) {
             case CRM_CLIENT_IPC:
-#ifdef HAVE_IPCS_GET_BUFFER_SIZE
-            {
-                int max_send_size = qb_ipcs_connection_get_buffer_size(client->ipcs);
-                if (max_send_size < update->iov_size) {
-                    struct iovec *iov;
-                    int new_size = crm_ipc_prepare(0, update->msg, &iov, max_send_size);
-
-                    if (new_size > 0) {
-                        free(update->iov[0].iov_base);
-                        free(update->iov[1].iov_base);
-                        free(update->iov);
-
-                        update->iov = iov;
-                        update->iov_size = new_size;
-                    } else {
-                        crm_warn("Preparing notification for client %s/%s failed", client->name, client->id);
-                        return FALSE;
-                    }
-                }
-            }
-#endif
-
                 if (crm_ipcs_sendv(client, update->iov, crm_ipc_server_event) < 0) {
                     crm_warn("Notification of client %s/%s failed", client->name, client->id);
                 }

configure.ac

@@ -1017,7 +1017,7 @@ AC_CHECK_LIB(qb, qb_ipcs_connection_auth_set)
 LIBQB_LOG=1
 PCMK_FEATURES="$PCMK_FEATURES libqb-logging libqb-ipc"
 
-AC_CHECK_FUNCS(qb_ipcs_connection_get_buffer_size, AC_DEFINE(HAVE_IPCS_GET_BUFFER_SIZE,  1, [Have qb_ipcs_connection_get_buffer_size function]))
+AC_CHECK_FUNCS(qb_ipcs_connection_get_buffer_size, AC_DEFINE(HAVE_IPCS_GET_BUFFER_SIZE,  1, [Have qb_ipcc_get_buffer_size function]))
 
 if
    ! pkg-config --atleast-version 0.13 libqb

include/crm/common/ipc.h

@@ -73,6 +73,7 @@ int crm_ipc_ready(crm_ipc_t * client);
 long crm_ipc_read(crm_ipc_t * client);
 const char *crm_ipc_buffer(crm_ipc_t * client);
 const char *crm_ipc_name(crm_ipc_t * client);
+int crm_ipc_default_buffer_size(void);
 
 /* Utils */
 xmlNode *create_hello_message(const char *uuid, const char *client_name,

include/crm/pengine/internal.h

@@ -264,5 +264,5 @@ gboolean xml_contains_remote_node(xmlNode *xml);
 gboolean is_baremetal_remote_node(node_t *node);
 gboolean is_container_remote_node(node_t *node);
 gboolean is_remote_node(node_t *node);
-gboolean rsc_contains_remote_node(pe_working_set_t * data_set, resource_t *rsc);
+resource_t * rsc_contains_remote_node(pe_working_set_t * data_set, resource_t *rsc);
 #endif

include/crm/pengine/status.h

@@ -72,7 +72,6 @@ enum pe_find {
 #  define pe_flag_startup_probes	0x00010000ULL
 #  define pe_flag_have_status		0x00020000ULL
 #  define pe_flag_have_remote_nodes	0x00040000ULL
-#  define pe_flag_container_probes	0x00080000ULL
 
 #  define pe_flag_quick_location  	0x00100000ULL
 
@@ -186,6 +185,7 @@ struct node_s {
 #  define pe_rsc_migrating	0x00400000ULL
 
 #  define pe_rsc_failure_ignored 0x01000000ULL
+#  define pe_rsc_unexpectedly_running 0x02000000ULL
 
 #  define pe_rsc_needs_quorum	 0x10000000ULL
 #  define pe_rsc_needs_fencing	 0x20000000ULL

lib/common/ipc.c

@@ -60,6 +60,12 @@ crm_ipc_init(void)
     }
 }
 
+int
+crm_ipc_default_buffer_size(void)
+{
+    return pick_ipc_buffer(0);
+}
+
 static char *
 generateReference(const char *custom1, const char *custom2)
 {
@@ -668,21 +674,13 @@ crm_ipcs_send(crm_client_t * c, uint32_t request, xmlNode * message,
 {
     struct iovec *iov = NULL;
     ssize_t rc = 0;
-    int32_t max_msg_size = 0;
 
     if(c == NULL) {
         return -EDESTADDRREQ;
     }
     crm_ipc_init();
 
-    /* when sending from server to client, we need to use the client's
-     * max buffer size if possible */
-    max_msg_size = ipc_buffer_max;
-#ifdef HAVE_IPCS_GET_BUFFER_SIZE
-    max_msg_size = qb_ipcs_connection_get_buffer_size(c->ipcs);
-#endif
-
-    rc = crm_ipc_prepare(request, message, &iov, max_msg_size);
+    rc = crm_ipc_prepare(request, message, &iov, ipc_buffer_max);
     if (rc > 0) {
         rc = crm_ipcs_sendv(c, iov, flags | crm_ipc_server_free);
 
@@ -790,6 +788,15 @@ crm_ipc_connect(crm_ipc_t * client)
 
     qb_ipcc_context_set(client->ipc, client);
 
+#ifdef HAVE_IPCS_GET_BUFFER_SIZE
+    client->max_buf_size = qb_ipcc_get_buffer_size(client->ipc);
+    if (client->max_buf_size < client->buf_size) {
+        free(client->buffer);
+        client->buffer = calloc(1, client->max_buf_size);
+        client->buf_size = client->max_buf_size;
+    }
+#endif
+
     return TRUE;
 }
 

lib/common/mainloop.c

@@ -573,6 +573,12 @@ mainloop_add_ipc_server(const char *name, enum qb_ipc_type type,
 
     crm_client_init();
     server = qb_ipcs_create(name, 0, pick_ipc_type(type), callbacks);
+
+#ifdef HAVE_IPCS_GET_BUFFER_SIZE
+    /* All clients should use at least ipc_buffer_max as their buffer size */
+    qb_ipcs_enforce_buffer_size(server, crm_ipc_default_buffer_size());
+#endif
+
     qb_ipcs_poll_handlers_set(server, &gio_poll_funcs);
 
     rc = qb_ipcs_run(server);

lib/pengine/common.c

@@ -116,8 +116,6 @@ pe_cluster_option pe_opts[] = {
 	  "This was the old default.  However when set to FALSE, the cluster will instead use the resource's failcount and value for resource-failure-stickiness" },
 	{ "enable-startup-probes", NULL, "boolean", NULL, "true", &check_boolean,
 	  "Should the cluster check for active resources during startup", NULL },
-	{ "enable-container-probes", NULL, "boolean", NULL, "true", &check_boolean,
-	  "Should the cluster check for active resources on container nodes during startup", NULL },
 
 	/* Stonith Options */
 	{ "stonith-enabled", "stonith_enabled", "boolean", NULL, "true", &check_boolean,

lib/pengine/unpack.c

@@ -88,11 +88,6 @@ unpack_config(xmlNode * config, pe_working_set_t * data_set)
         crm_info("Startup probes: disabled (dangerous)");
     }
 
-    set_config_flag(data_set, "enable-container-probes", pe_flag_container_probes);
-    if(is_not_set(data_set->flags, pe_flag_container_probes)) {
-        crm_info("Container probes: disabled");
-    }
-
     value = pe_pref(data_set->config_hash, "stonith-timeout");
     data_set->stonith_timeout = crm_get_msec(value);
     crm_debug("STONITH timeout: %d", data_set->stonith_timeout);
@@ -2062,6 +2057,9 @@ unpack_lrm_resources(node_t * node, xmlNode * lrm_rsc_list, pe_working_set_t * d
 {
     xmlNode *rsc_entry = NULL;
     gboolean found_orphaned_container_filler = FALSE;
+    GListPtr unexpected_containers = NULL;
+    GListPtr gIter = NULL;
+    resource_t *remote = NULL;
 
     CRM_CHECK(node != NULL, return FALSE);
 
@@ -2073,9 +2071,29 @@ unpack_lrm_resources(node_t * node, xmlNode * lrm_rsc_list, pe_working_set_t * d
         if (crm_str_eq((const char *)rsc_entry->name, XML_LRM_TAG_RESOURCE, TRUE)) {
             resource_t *rsc;
             rsc = unpack_lrm_rsc_state(node, rsc_entry, data_set);
-            if (rsc && is_set(rsc->flags, pe_rsc_orphan_container_filler)) {
+            if (!rsc) {
+                continue;
+            }
+            if (is_set(rsc->flags, pe_rsc_orphan_container_filler)) {
                 found_orphaned_container_filler = TRUE;
             }
+            if (is_set(rsc->flags, pe_rsc_unexpectedly_running)) {
+                remote = rsc_contains_remote_node(data_set, rsc);
+                if (remote) {
+                    unexpected_containers = g_list_append(unexpected_containers, remote);
+                }
+            }
+        }
+    }
+
+    /* If a container resource is unexpectedly up... and the remote-node
+     * connection resource for that container is not up, the entire container
+     * must be recovered. */
+    for (gIter = unexpected_containers; gIter != NULL; gIter = gIter->next) {
+        remote = (resource_t *) gIter->data;
+        if (remote->role != RSC_ROLE_STARTED) {
+            crm_warn("Recovering container resource %s. Resource is unexpectedly running and involves a remote-node.");
+            set_bit(remote->container->flags, pe_rsc_failed);
         }
     }
 
@@ -2085,7 +2103,7 @@ unpack_lrm_resources(node_t * node, xmlNode * lrm_rsc_list, pe_working_set_t * d
     if (found_orphaned_container_filler) {
         handle_orphaned_container_fillers(lrm_rsc_list, data_set);
     }
-
+    g_list_free(unexpected_containers);
     return TRUE;
 }
 
@@ -2469,6 +2487,7 @@ determine_op_status(
                 result = PCMK_LRM_OP_DONE;
                 pe_rsc_info(rsc, "Operation %s found resource %s active on %s",
                             task, rsc->id, node->details->uname);
+                set_bit(rsc->flags, pe_rsc_unexpectedly_running);
 
                 /* legacy code for pre-0.6.5 operations */
             } else if (target_rc < 0 && interval > 0 && rsc->role == RSC_ROLE_MASTER) {

lib/pengine/utils.c

@@ -1805,11 +1805,11 @@ is_remote_node(node_t *node)
     return FALSE;
 }
 
-gboolean
+resource_t *
 rsc_contains_remote_node(pe_working_set_t * data_set, resource_t *rsc)
 {
     if (is_set(data_set->flags, pe_flag_have_remote_nodes) == FALSE) {
-        return FALSE;
+        return NULL;
     }
 
     if (rsc->fillers) {
@@ -1818,11 +1818,11 @@ rsc_contains_remote_node(pe_working_set_t * data_set, resource_t *rsc)
             resource_t *filler = (resource_t *) gIter->data;
 
             if (filler->is_remote_node) {
-                return TRUE;
+                return filler;
             }
         }
     }
-    return FALSE;
+    return NULL;
 }
 
 gboolean

pengine/allocate.c

@@ -850,8 +850,8 @@ probe_resources(pe_working_set_t * data_set)
              * it is being shutdown. */
             continue;
 
-        } else if (is_container_remote_node(node) && is_not_set(data_set->flags, pe_flag_container_probes)) {
-            /* skip container nodes if container node probes are disabled */
+        } else if (is_container_remote_node(node)) {
+            /* TODO enable container node probes once ordered probing is implemented. */
             continue;
 
         } else if (probe_complete == NULL) {

pengine/native.c

@@ -2414,7 +2414,7 @@ native_create_probe(resource_t * rsc, node_t * node, action_t * complete,
     if (force == FALSE && is_not_set(data_set->flags, pe_flag_startup_probes)) {
         pe_rsc_trace(rsc, "Skipping active resource detection for %s", rsc->id);
         return FALSE;
-    } else if (force == FALSE && is_container_remote_node(node) && is_not_set(data_set->flags, pe_flag_container_probes)) {
+    } else if (force == FALSE && is_container_remote_node(node)) {
         pe_rsc_trace(rsc, "Skipping active resource detection for %s on container %s",
                      rsc->id, node->details->id);
         return FALSE;

pengine/regression.sh

@@ -714,7 +714,7 @@ do_test whitebox-ms-ordering "Verify promote/demote can not occur before connect
 do_test whitebox-ms-ordering "Verify promote/demote can not occur before connection is established"
 do_test whitebox-orphaned    "Properly shutdown orphaned whitebox container"
 do_test whitebox-orphan-ms   "Properly tear down orphan ms resources on remote-nodes"
-do_test whitebox-disable-probes "Verify probes are not executed when container probes are disabled"
+do_test whitebox-unexpectedly-running "Recover container nodes the cluster did not start."
 
 echo ""
 do_test remote-startup-probes  "Baremetal remote-node startup probes"

pengine/test10/master-partially-demoted-group.xml

@@ -23,7 +23,6 @@
         <nvpair name="election-timeout" value="2m" id="cib-bootstrap-options-election-timeout"/>
         <nvpair name="enable-acl" value="false" id="cib-bootstrap-options-enable-acl"/>
         <nvpair name="enable-startup-probes" value="true" id="cib-bootstrap-options-enable-startup-probes"/>
-        <nvpair name="enable-container-probes" value="true" id="cib-bootstrap-options-enable-container-probes"/>
         <nvpair name="migration-limit" value="4" id="cib-bootstrap-options-migration-limit"/>
         <nvpair name="node-health-green" value="0" id="cib-bootstrap-options-node-health-green"/>
         <nvpair name="node-health-red" value="-INFINITY" id="cib-bootstrap-options-node-health-red"/>

pengine/test10/whitebox-asymmetric.dot

@@ -1,26 +1,10 @@
  digraph "g" {
 "18node2_monitor_30000 18builder" [ style=bold color="green" fontcolor="black"]
 "18node2_start_0 18builder" -> "18node2_monitor_30000 18builder" [ style = bold]
-"18node2_start_0 18builder" -> "nfs_mount_monitor_0 18node2" [ style = bold]
 "18node2_start_0 18builder" -> "nfs_mount_monitor_10000 18node2" [ style = bold]
 "18node2_start_0 18builder" -> "nfs_mount_start_0 18node2" [ style = bold]
-"18node2_start_0 18builder" -> "vg_tags_dup_monitor_0 18node2" [ style = bold]
-"18node2_start_0 18builder" -> "vg_tags_monitor_0 18node2" [ style = bold]
-"18node2_start_0 18builder" -> "webserver_monitor_0 18node2" [ style = bold]
 "18node2_start_0 18builder" [ style=bold color="green" fontcolor="black"]
-"nfs_mount_monitor_0 18node2" -> "probe_complete 18node2" [ style = bold]
-"nfs_mount_monitor_0 18node2" [ style=bold color="green" fontcolor="black"]
 "nfs_mount_monitor_10000 18node2" [ style=bold color="green" fontcolor="black"]
 "nfs_mount_start_0 18node2" -> "nfs_mount_monitor_10000 18node2" [ style = bold]
 "nfs_mount_start_0 18node2" [ style=bold color="green" fontcolor="black"]
-"probe_complete 18node2" -> "probe_complete" [ style = bold]
-"probe_complete 18node2" [ style=bold color="green" fontcolor="black"]
-"probe_complete" -> "nfs_mount_start_0 18node2" [ style = bold]
-"probe_complete" [ style=bold color="green" fontcolor="orange"]
-"vg_tags_dup_monitor_0 18node2" -> "probe_complete 18node2" [ style = bold]
-"vg_tags_dup_monitor_0 18node2" [ style=bold color="green" fontcolor="black"]
-"vg_tags_monitor_0 18node2" -> "probe_complete 18node2" [ style = bold]
-"vg_tags_monitor_0 18node2" [ style=bold color="green" fontcolor="black"]
-"webserver_monitor_0 18node2" -> "probe_complete 18node2" [ style = bold]
-"webserver_monitor_0 18node2" [ style=bold color="green" fontcolor="black"]
 }