diff --git a/src/include/pool.h b/src/include/pool.h
index 65907dcf1..fea5744f3 100644
--- a/src/include/pool.h
+++ b/src/include/pool.h
@@ -487,6 +487,13 @@ typedef struct
 	int			count;			/* request node ids count */
 } POOL_REQUEST_NODE;
 
+/*
+ * Maximum number of pcp worker child process.  * Since pcp worker process is
+ * forked whenever failover/failback request is made, it should be equal to
+ * MAX_REQUEST_QUEUE_SIZE + some room. 10 is an arbitrary number.
+*/
+#define	MAX_PCP_WORKER_PIDS		MAX_REQUEST_QUEUE_SIZE + 10
+
 typedef struct
 {
 	POOL_REQUEST_NODE request[MAX_REQUEST_QUEUE_SIZE];
@@ -524,6 +531,12 @@ typedef struct
 	bool		query_cache_invalidate_request; /* true if
 												 * pcp_invalidate_query_cache
 												 * requested */
+
+	/*
+	 * pcp worker child pids. This is inherited to new pcp main process to
+	 * track pcp worker child when new pcp worker child starts.
+	 */
+	pid_t		pcp_worker_pids[MAX_PCP_WORKER_PIDS];
 } POOL_REQUEST_INFO;
 
 /* description of row. corresponding to RowDescription message */
diff --git a/src/main/pgpool_main.c b/src/main/pgpool_main.c
index 32bcb0a1f..4112074e2 100644
--- a/src/main/pgpool_main.c
+++ b/src/main/pgpool_main.c
@@ -3200,6 +3200,8 @@ initialize_shared_mem_objects(bool clear_memcache_oidmaps)
 		wd_ipc_initialize_data();
 	}
 
+	/* initialize pcp worker child pids */
+	memset(Req_info->pcp_worker_pids, 0, sizeof(Req_info->pcp_worker_pids));
 }
 
 /*
diff --git a/src/pcp_con/pcp_child.c b/src/pcp_con/pcp_child.c
index e07c8897e..fc1cda311 100644
--- a/src/pcp_con/pcp_child.c
+++ b/src/pcp_con/pcp_child.c
@@ -154,6 +154,16 @@ pcp_main(int *fds)
 	/* We can now handle ereport(ERROR) */
 	PG_exception_stack = &local_sigjmp_buf;
 
+	/*
+	 * Restore pcp woker child pids from shmem
+	 */
+	for (int i = 0; i < MAX_PCP_WORKER_PIDS; i++)
+	{
+		pid_t	pid = Req_info->pcp_worker_pids[i];
+		if (pid != 0)
+			pcp_worker_children = lappend_int(pcp_worker_children, (int) pid);
+	}
+
 	/*
 	 * Unblock signals
 	 */
@@ -326,6 +336,7 @@ start_pcp_command_processor_process(int port, int *fds)
 	}
 	else						/* parent */
 	{
+		int		i;
 		if (pool_config->log_pcp_processes)
 			ereport(LOG,
 					(errmsg("forked new pcp worker, pid=%d socket=%d",
@@ -334,6 +345,18 @@ start_pcp_command_processor_process(int port, int *fds)
 		close(port);
 		/* Add it to the list */
 		pcp_worker_children = lappend_int(pcp_worker_children, (int) pid);
+		/* save it to shmem */
+		for (i = 0; i < MAX_PCP_WORKER_PIDS; i++)
+		{
+			if (Req_info->pcp_worker_pids[i] == 0)
+			{
+				Req_info->pcp_worker_pids[i] = pid;
+				break;
+			}
+		}
+		if (i == MAX_PCP_WORKER_PIDS)
+			ereport(WARNING,
+					(errmsg("no empty slot in pcp worker table")));
 	}
 }
 
@@ -378,6 +401,16 @@ reaper(void)
 
 	while ((pid = pool_waitpid(&status)) > 0)
 	{
+		/* remove the pid from shmem */
+		for (int i = 0; i < MAX_PCP_WORKER_PIDS; i++)
+		{
+			if (Req_info->pcp_worker_pids[i] == pid)
+			{
+				Req_info->pcp_worker_pids[i] = 0;
+				break;
+			}
+		}
+
 		if (WIFEXITED(status))
 		{
 			if (WEXITSTATUS(status) == POOL_EXIT_FATAL)
@@ -405,6 +438,17 @@ reaper(void)
 				(errmsg("going to remove pid: %d from pid list having %d elements", pid, list_length(pcp_worker_children))));
 		/* remove the pid of process from the list */
 		pcp_worker_children = list_delete_int(pcp_worker_children, pid);
+
+		/* remove the pid from shmem */
+		for (int i = 0; i < MAX_PCP_WORKER_PIDS; i++)
+		{
+			if (Req_info->pcp_worker_pids[i] == pid)
+			{
+				Req_info->pcp_worker_pids[i] = 0;
+				break;
+			}
+		}
+
 		ereport(DEBUG2,
 				(errmsg("new list have %d elements", list_length(pcp_worker_children))));
 	}