Removed unnecessary 'worker' field from shardman.nodes.

Author: arssher

init.sql

@@ -179,14 +179,15 @@ BEGIN
 		SELECT pg_settings.setting INTO lord_connstring FROM pg_settings
 			WHERE NAME = 'shardman.shardlord_connstring';
 		EXECUTE format(
-			'INSERT INTO @extschema@.nodes VALUES (DEFAULT, %L, NULL, false, true)
+			'INSERT INTO shardman.nodes VALUES (DEFAULT, %L, NULL, true)
 			RETURNING id', lord_connstring) INTO lord_id;
 		PERFORM shardman.set_node_id(lord_id);
 		init_lord := true;
 	ELSE
-		EXECUTE 'SELECT NOT (SELECT lord FROM shardman.nodes WHERE id = $1)'
+		EXECUTE 'SELECT NOT (SELECT shardlord FROM shardman.nodes WHERE id = $1)'
 			INTO init_lord USING lord_id;
-		EXECUTE 'UPDATE shardman.nodes SET lord = true WHERE id = $1' USING lord_id;
+		EXECUTE 'UPDATE shardman.nodes SET shardlord = true WHERE id = $1'
+			USING lord_id;
 	END IF;
 	IF init_lord THEN
 		-- TODO: set up lr channels

membership.sql

@@ -6,23 +6,25 @@
  * ------------------------------------------------------------------------
  */
 
--- active is the normal mode, others needed only for proper node add and removal
+-- active is the normal mode, removed means node removed, others needed only
+-- for proper node add and removal
 CREATE TYPE worker_node_status AS ENUM (
 	'active', 'add_in_progress', 'rm_in_progress', 'removed');
 
 -- list of nodes present in the cluster
 CREATE TABLE nodes (
 	id serial PRIMARY KEY,
-	connstring text NOT NULL UNIQUE,
-	worker_status worker_node_status,
+	connstring text NOT NULL,
 	-- While currently we don't support lord and worker roles on one node,
-	-- potentially node can be either worker, lord or both, so we need 2 bits.
-	-- One bool with NULL might be fine, but it seems a bit counter-intuitive.
-	worker bool NOT NULL DEFAULT true,
-	lord bool NOT NULL DEFAULT false,
+	-- potentially node can be either worker, lord or both.
+	worker_status worker_node_status, -- NULL if this node is not a worker
+	-- is this node shardlord?
+	shardlord bool NOT NULL DEFAULT false,
 	-- cmd by which node was added
 	added_by bigint REFERENCES shardman.cmd_log(id)
 );
+CREATE UNIQUE INDEX unique_node_connstr ON shardman.nodes (connstring)
+	WHERE (worker_status = 'active' OR shardlord);
 
 -- Lord is removing us, so reset our state, removing all subscriptions. A bit
 -- tricky part: we can't DROP SUBSCRIPTION here, because that would mean
@@ -88,7 +90,7 @@ BEGIN
 	  	INTO n_id;
 	IF n_id IS NULL THEN
 		INSERT INTO shardman.nodes
-			VALUES (DEFAULT, connstring, 'add_in_progress', true, false, cmd_id)
+			VALUES (DEFAULT, connstring, 'add_in_progress', false, cmd_id)
 			RETURNING id INTO n_id;
 	END IF;
 	RETURN n_id;
@@ -110,7 +112,8 @@ END $$ LANGUAGE plpgsql;
 -- one.
 CREATE FUNCTION get_worker_node_connstr(node_id int) RETURNS text AS $$
 DECLARE
-	connstr text := connstring FROM shardman.nodes WHERE id = node_id AND worker;
+	connstr text := connstring FROM shardman.nodes WHERE id = node_id AND
+				worker_status IS NOT NULL;
 BEGIN
 	IF connstr IS NULL THEN
 		RAISE EXCEPTION 'Worker node with id % not found', node_id;

readme.txt

@@ -91,17 +91,17 @@ Remove node from the cluster. Its shardman state will be reset. We don't delete
 tables with data and foreign tables though.
 
 You can see all cluster nodes at any time by examining shardman.nodes table:
--- active is the normal mode, others needed only for proper node add and removal
+-- active is the normal mode, removed means node removed, others needed only
+-- for proper node add and removal
 CREATE TYPE worker_node_status AS ENUM (
 	'active', 'add_in_progress', 'rm_in_progress', 'removed');
 CREATE TABLE nodes (
 	id serial PRIMARY KEY,
 	connstring text NOT NULL UNIQUE,
-	worker_status worker_node_status,
 	-- While currently we don't support lord and worker roles on one node,
-	-- potentially node can be either worker, lord or both, so we need 2 bits.
-	-- One bool with NULL might be fine, but it seems a bit counter-intuitive.
-	worker bool NOT NULL DEFAULT true,
+	-- potentially node can be either worker, lord or both.
+	worker_status worker_node_status, -- NULL if this node is not a worker
+	-- is this node shardlord?
 	lord bool NOT NULL DEFAULT false,
 	-- cmd by which node was added
 	added_by bigint REFERENCES shardman.cmd_log(id)

src/pg_shardman.c

@@ -634,14 +634,14 @@ insert_node(const char *connstr, int64 cmd_id)
 }
 
 /*
- * Returns true, if node 'id' is in cluster and not in add_in_progress state
+ * Returns true, if node 'id' is active node in cluster or rm in progress
  */
 static bool
 node_in_cluster(int id)
 {
 	char *sql = psprintf(
-		"select id from shardman.nodes where id = %d and (shardlord OR"
-		" worker_status != 'add_in_progress');",
+		"select id from shardman.nodes where id = %d and (shardlord or"
+		" worker_status = 'active' or worker_status = 'rm_in_progress');",
 		id);
 	bool res;
 
@@ -695,7 +695,7 @@ rm_node(Cmd *cmd)
 	 * fixed.
 	 */
 	sql = psprintf(
-		"select shardman.drop_repslot('shardman_meta_sub_%d');"
+		"select shardman.drop_repslot('shardman_meta_sub_%d', true);"
 		"update shardman.nodes set worker_status = 'removed' where id = %d;"
 		"update shardman.cmd_log set status = 'success' where id = %ld;",
 		node_id, node_id, cmd->id);
@@ -708,14 +708,14 @@ rm_node(Cmd *cmd)
 
 /*
  * Get connstr of worker node with id node_id. Memory is palloc'ed.
- * NULL is returned, if there is no such node.
+ * NULL is returned, if there is no such worker.
  */
 char *
 get_worker_node_connstr(int32 node_id)
 {
 	MemoryContext oldcxt = CurrentMemoryContext;
 	char *sql = psprintf("select connstring from shardman.nodes where id = %d"
-						 " and worker", node_id);
+						 " and worker_status is not null", node_id);
 	char *res;
 
 	SPI_PROLOG;