Add agent CLI command to remove a stale node (netdata#17691)

* Mark the node as ephemeral in the database Remove node command * Do not unregister the localhost Do not unregister a live node Make sure we rewrite the ephemeral value when a child reconnects
webfutureiorepo · May 20, 2024 · 9726ca1 · 9726ca1
1 parent 05dd823
commit 9726ca1
Show file tree

Hide file tree

Showing 5 changed files with 94 additions and 4 deletions.
diff --git a/src/collectors/plugins.d/pluginsd_parser.c b/src/collectors/plugins.d/pluginsd_parser.c
@@ -649,10 +649,12 @@ static inline PARSER_RC pluginsd_label(char **words, size_t num_words, PARSER *p
 
     if (strcmp(name,HOST_LABEL_IS_EPHEMERAL) == 0) {
         int is_ephemeral = appconfig_test_boolean_value((char *) value);
-        if (is_ephemeral) {
-            RRDHOST *host = pluginsd_require_scope_host(parser, PLUGINSD_KEYWORD_LABEL);
-            if (likely(host))
+        RRDHOST *host = pluginsd_require_scope_host(parser, PLUGINSD_KEYWORD_LABEL);
+        if (host) {
+            if (is_ephemeral)
                 rrdhost_option_set(host, RRDHOST_OPTION_EPHEMERAL_HOST);
+            else
+                rrdhost_option_clear(host, RRDHOST_OPTION_EPHEMERAL_HOST);
         }
     }
 

diff --git a/src/daemon/commands.c b/src/daemon/commands.c
@@ -47,6 +47,7 @@ static cmd_status_t cmd_ping_execute(char *args, char **message);
 static cmd_status_t cmd_aclk_state(char *args, char **message);
 static cmd_status_t cmd_version(char *args, char **message);
 static cmd_status_t cmd_dumpconfig(char *args, char **message);
+static cmd_status_t cmd_remove_node(char *args, char **message);
 
 static command_info_t command_info_array[] = {
         {"help", cmd_help_execute, CMD_TYPE_HIGH_PRIORITY},                  // show help menu
@@ -61,7 +62,8 @@ static command_info_t command_info_array[] = {
         {"ping", cmd_ping_execute, CMD_TYPE_ORTHOGONAL},
         {"aclk-state", cmd_aclk_state, CMD_TYPE_ORTHOGONAL},
         {"version", cmd_version, CMD_TYPE_ORTHOGONAL},
-        {"dumpconfig", cmd_dumpconfig, CMD_TYPE_ORTHOGONAL}
+        {"dumpconfig", cmd_dumpconfig, CMD_TYPE_ORTHOGONAL},
+        {"remove-stale-node", cmd_remove_node, CMD_TYPE_ORTHOGONAL}
 };
 
 /* Mutexes for commands of type CMD_TYPE_ORTHOGONAL */
@@ -129,6 +131,8 @@ static cmd_status_t cmd_help_execute(char *args, char **message)
              "    Returns current state of ACLK and Cloud connection. (optionally in json).\n"
              "dumpconfig\n"
              "    Returns the current netdata.conf on stdout.\n"
+             "remove-stale-node node_id|machine_guid\n"
+             "    Unregisters and removes a node from the cloud.\n"
              "version\n"
              "    Returns the netdata version.\n",
              MAX_COMMAND_LENGTH - 1);
@@ -326,6 +330,54 @@ static cmd_status_t cmd_dumpconfig(char *args, char **message)
     return CMD_STATUS_SUCCESS;
 }
 
+static cmd_status_t cmd_remove_node(char *args, char **message)
+{
+    (void)args;
+
+    BUFFER *wb = buffer_create(1024, NULL);
+    if (strlen(args) == 0) {
+        buffer_sprintf(wb, "Please specify a machine or node UUID");
+        goto done;
+    }
+
+    RRDHOST *host = NULL;
+    host = rrdhost_find_by_guid(args);
+    if (!host)
+        host = find_host_by_node_id(args);
+
+    if (!host)
+        buffer_sprintf(wb, "Node with machine or node UUID \"%s\" not found", args);
+    else {
+
+        if (host == localhost) {
+            buffer_sprintf(wb, "You cannot unregister the parent node");
+            goto done;
+        }
+
+        if (rrdhost_is_online(host)) {
+            buffer_sprintf(wb, "Cannot unregister a live node");
+            goto done;
+        }
+
+        if (!rrdhost_option_check(host, RRDHOST_OPTION_EPHEMERAL_HOST)) {
+            rrdhost_option_set(host, RRDHOST_OPTION_EPHEMERAL_HOST);
+            sql_set_host_label(&host->host_uuid, "_is_ephemeral", "true");
+            aclk_host_state_update(host, 0, 0);
+            unregister_node(host->machine_guid);
+            freez(host->node_id);
+            host->node_id = NULL;
+            buffer_sprintf(wb, "Unregistering node with machine guid %s, hostname = %s", host->machine_guid, rrdhost_hostname(host));
+        }
+        else
+            buffer_sprintf(wb, "Node with machine guid %s, hostname = %s is already unregistered", host->machine_guid, rrdhost_hostname(host));
+    }
+
+done:
+    *message = strdupz(buffer_tostring(wb));
+    buffer_free(wb);
+    return CMD_STATUS_SUCCESS;
+}
+
 static void cmd_lock_exclusive(unsigned index)
 {
     (void)index;

diff --git a/src/daemon/commands.h b/src/daemon/commands.h
@@ -20,6 +20,7 @@ typedef enum cmd {
     CMD_ACLK_STATE,
     CMD_VERSION,
     CMD_DUMPCONFIG,
+    CMD_REMOVE_NODE,
     CMD_TOTAL_COMMANDS
 } cmd_t;
 

diff --git a/src/database/sqlite/sqlite_metadata.c b/src/database/sqlite/sqlite_metadata.c
@@ -256,6 +256,40 @@ static inline void set_host_node_id(RRDHOST *host, nd_uuid_t *node_id)
         uuid_unparse_lower(*node_id, wc->node_id);
 }
 
+#define SQL_SET_HOST_LABEL                                                                                             \
+    "INSERT INTO host_label (host_id, source_type, label_key, label_value, date_created) "                             \
+    "VALUES (@host_id, @source_type, @label_key, @label_value, UNIXEPOCH()) ON CONFLICT (host_id, label_key) "         \
+    " DO UPDATE SET source_type = excluded.source_type, label_value=excluded.label_value, date_created=UNIXEPOCH()"
+
+bool sql_set_host_label(nd_uuid_t *host_id, const char *label_key, const char *label_value)
+{
+    sqlite3_stmt *res = NULL;
+    bool status = false;
+
+    if (!label_key || !label_value || !host_id)
+        return false;
+
+    if (!PREPARE_STATEMENT(db_meta, SQL_SET_HOST_LABEL, &res))
+        return 1;
+
+    int param = 0;
+    SQLITE_BIND_FAIL(done, sqlite3_bind_blob(res, ++param, host_id, sizeof(*host_id), SQLITE_STATIC));
+    SQLITE_BIND_FAIL(done, sqlite3_bind_int(res, ++param, RRDLABEL_SRC_AUTO));
+    SQLITE_BIND_FAIL(done, sqlite3_bind_text(res, ++param, label_key, -1, SQLITE_STATIC));
+    SQLITE_BIND_FAIL(done, sqlite3_bind_text(res, ++param, label_value, -1, SQLITE_STATIC));
+
+    param = 0;
+    int rc = execute_insert(res);
+    status = (rc == SQLITE_DONE);
+    if (false == status)
+        error_report("Failed to store node instance information, rc = %d", rc);
+done:
+    REPORT_BIND_FAIL(res, param);
+    SQLITE_FINALIZE(res);
+    return status;
+}
+
+
 #define SQL_UPDATE_NODE_ID  "UPDATE node_instance SET node_id = @node_id WHERE host_id = @host_id"
 
 int update_node_id(nd_uuid_t *host_id, nd_uuid_t *node_id)

diff --git a/src/database/sqlite/sqlite_metadata.h b/src/database/sqlite/sqlite_metadata.h
@@ -49,6 +49,7 @@ void sql_load_node_id(RRDHOST *host);
 void sql_build_host_system_info(nd_uuid_t *host_id, struct rrdhost_system_info *system_info);
 void invalidate_node_instances(nd_uuid_t *host_id, nd_uuid_t *claim_id);
 RRDLABELS *sql_load_host_labels(nd_uuid_t *host_id);
+bool sql_set_host_label(nd_uuid_t *host_id, const char *label_key, const char *label_value);
 
 uint64_t sqlite_get_meta_space(void);
 int sql_init_meta_database(db_check_action_type_t rebuild, int memory);