Christopher B. Browne's Home Page
cbbrowne@acm.org

8.64. failednode(integer, integer)

Function Properties

PLPGSQLinteger
Initiate failover from failed_node to backup_node. This function must be called on all nodes, and then waited for the restart of all node daemons.
    declare
    	p_failed_node		alias for $1;
    	p_backup_node		alias for $2;
    	v_row				record;
    	v_row2				record;
    	v_n					int4;
    begin
    	-- ----
    	-- Grab the central configuration lock
    	-- ----
    	lock table sl_config_lock;
    
    	-- ----
    	-- All consistency checks first
    	-- Check that every node that has a path to the failed node
    	-- also has a path to the backup node.
    	-- ----
    	for v_row in select P.pa_client
    			from sl_path P
    			where P.pa_server = p_failed_node
    				and P.pa_client <> p_backup_node
    				and not exists (select true from sl_path PP
    							where PP.pa_server = p_backup_node
    								and PP.pa_client = P.pa_client)
    	loop
    		raise exception 'Slony-I: cannot failover - node % has no path to the backup node',
    				v_row.pa_client;
    	end loop;
    
    	-- ----
    	-- Check all sets originating on the failed node
    	-- ----
    	for v_row in select set_id
    			from sl_set
    			where set_origin = p_failed_node
    	loop
    		-- ----
    		-- Check that the backup node is subscribed to all sets
    		-- that originate on the failed node
    		-- ----
    		select into v_row2 sub_forward, sub_active
    				from sl_subscribe
    				where sub_set = v_row.set_id
    					and sub_receiver = p_backup_node;
    		if not found then
    			raise exception 'Slony-I: cannot failover - node % is not subscribed to set %',
    					p_backup_node, v_row.set_id;
    		end if;
    
    		-- ----
    		-- Check that the subscription is active
    		-- ----
    		if not v_row2.sub_active then
    			raise exception 'Slony-I: cannot failover - subscription for set % is not active',
    					v_row.set_id;
    		end if;
    
    		-- ----
    		-- If there are other subscribers, the backup node needs to
    		-- be a forwarder too.
    		-- ----
    		select into v_n count(*)
    				from sl_subscribe
    				where sub_set = v_row.set_id
    					and sub_receiver <> p_backup_node;
    		if v_n > 0 and not v_row2.sub_forward then
    			raise exception 'Slony-I: cannot failover - node % is not a forwarder of set %',
    					p_backup_node, v_row.set_id;
    		end if;
    	end loop;
    
    	-- ----
    	-- Terminate all connections of the failed node the hard way
    	-- ----
    	perform terminateNodeConnections(p_failed_node);
    
    	-- ----
    	-- Move the sets
    	-- ----
    	for v_row in select S.set_id, (select count(*)
    					from sl_subscribe SUB
    					where S.set_id = SUB.sub_set
    						and SUB.sub_receiver <> p_backup_node
    						and SUB.sub_provider = p_failed_node)
    					as num_direct_receivers 
    			from sl_set S
    			where S.set_origin = p_failed_node
    			for update
    	loop
    		-- ----
    		-- If the backup node is the only direct subscriber ...
    		-- ----
    		if v_row.num_direct_receivers = 0 then
    		        raise notice 'failedNode: set % has no other direct receivers - move now', v_row.set_id;
    			-- ----
    			-- backup_node is the only direct subscriber, move the set
    			-- right now. On the backup node itself that includes restoring
    			-- all user mode triggers, removing the protection trigger,
    			-- adding the log trigger, removing the subscription and the
    			-- obsolete setsync status.
    			-- ----
    			if p_backup_node = getLocalNodeId('_schemadoc') then
    				update sl_set set set_origin = p_backup_node
    						where set_id = v_row.set_id;
    
    				delete from sl_setsync
    						where ssy_setid = v_row.set_id;
    
    				for v_row2 in select * from sl_table
    						where tab_set = v_row.set_id
    						order by tab_id
    				loop
    					perform alterTableConfigureTriggers(v_row2.tab_id);
    				end loop;
    			end if;
    
    			delete from sl_subscribe
    					where sub_set = v_row.set_id
    						and sub_receiver = p_backup_node;
    		else
    			raise notice 'failedNode: set % has other direct receivers - change providers only', v_row.set_id;
    			-- ----
    			-- Backup node is not the only direct subscriber or not
    			-- a direct subscriber at all. 
    			-- This means that at this moment, we redirect all possible
    			-- direct subscribers to receive from the backup node, and the
    			-- backup node itself to receive from another one.
    			-- The admin utility will wait for the slon engine to
    			-- restart and then call failedNode2() on the node with
    			-- the highest SYNC and redirect this to it on
    			-- backup node later.
    			-- ----
    			update sl_subscribe
    					set sub_provider = (select min(SS.sub_receiver)
    							from sl_subscribe SS
    							where SS.sub_set = v_row.set_id
    								and SS.sub_receiver <> p_backup_node
    								and SS.sub_forward
    								and exists (
    									select 1 from sl_path
    										where pa_server = SS.sub_receiver
    										  and pa_client = p_backup_node
    								))
    					where sub_set = v_row.set_id
    						and sub_receiver = p_backup_node;
    			update sl_subscribe
    					set sub_provider = (select min(SS.sub_receiver)
    							from sl_subscribe SS
    							where SS.sub_set = v_row.set_id
    								and SS.sub_receiver <> p_failed_node
    								and SS.sub_forward
    								and exists (
    									select 1 from sl_path
    										where pa_server = SS.sub_receiver
    										  and pa_client = sl_subscribe.sub_receiver
    								))
    					where sub_set = v_row.set_id
    						and sub_receiver <> p_backup_node;
    			update sl_subscribe
    					set sub_provider = p_backup_node
    					where sub_set = v_row.set_id
    						and sub_receiver <> p_backup_node
    						and exists (
    							select 1 from sl_path
    								where pa_server = p_backup_node
    								  and pa_client = sl_subscribe.sub_receiver
    						);
    		end if;
    	end loop;
    
    	-- Rewrite sl_listen table
    	perform RebuildListenEntries();
    
    	-- Run addPartialLogIndices() to try to add indices to unused sl_log_? table
    	perform addPartialLogIndices();
    
    	-- ----
    	-- Make sure the node daemon will restart
    	-- ----
    	notify "_schemadoc_Restart";
    
    	-- ----
    	-- That is it - so far.
    	-- ----
    	return p_failed_node;
    end;

Google

If this was useful, let others know by an Affero rating

Contact me at cbbrowne@acm.org