Merge lp:~vkolesnikov/pbxt/pbxt-recovery-freeze-and-partitioned-tables into lp:pbxt

[Vladimir Kolesnikov]

=== modified file 'ChangeLog'
--- ChangeLog	2009-05-28 15:35:31 +0000
+++ ChangeLog	2009-06-03 14:53:00 +0000
@@ -3,6 +3,10 @@
 
 ------- 1.0.08 RC - Not yet released
 
+RN243: Fixed a bug that caused a recovery failure if partitioned pbxt tables where present. This happended because the recovery used a MySQL function to open tables and the PBXT handler was not yet registered
+
+RN242: Fixed a bug that caused a deadlock if pbxt initialization failed. This happened because pbxt ceanup was done from pbxt_init() with PLUGIN_lock being held by MySQL which lead to a deadlock in the freeer thread
+
 RN241: Fixed a heap corruption bug (writing to a freed memory location). It happened only when memory mapped files were used leading to heap inconsistency and program crash or termination by heap checker. Likely to happen right after or during DROP TABLE but possible in other cases too.
 
 RN240: Load the record cache on read when no using memory mapped files.

=== modified file 'src/ha_pbxt.cc'
--- src/ha_pbxt.cc	2009-05-21 10:27:57 +0000
+++ src/ha_pbxt.cc	2009-06-03 14:07:54 +0000
@@ -105,6 +105,8 @@
 static void		ha_release_exclusive_use(XTThreadPtr self, XTSharePtr share);
 static void		ha_close_open_tables(XTThreadPtr self, XTSharePtr share, ha_pbxt *mine);
 
+extern void		xt_xres_start_database_recovery(XTThreadPtr self);
+
 #ifdef TRACE_STATEMENTS
 
 #ifdef PRINT_STATEMENTS
@@ -167,7 +169,7 @@
 /* Variables for pbxt share methods */
 static xt_mutex_type	pbxt_database_mutex;		// Prevent a database from being opened while it is being dropped
 static XTHashTabPtr		pbxt_share_tables;			// Hash used to track open tables
-static XTDatabaseHPtr	pbxt_database = NULL;		// The global open database
+XTDatabaseHPtr			pbxt_database = NULL;		// The global open database
 static char				*pbxt_index_cache_size;
 static char				*pbxt_record_cache_size;
 static char				*pbxt_log_cache_size;
@@ -1100,7 +1102,7 @@
 #endif
 		self = xt_init_threading(pbxt_max_threads);				/* Create the main self: */
 		if (!self)
-			goto error_4;
+			goto error_3;
 
  		pbxt_inited = true;
 
@@ -1164,9 +1166,7 @@
 							xt_throw(self);
 					}
 
-					xt_open_database(self, mysql_real_data_home, TRUE);
-					pbxt_database = self->st_database;
-					xt_heap_reference(self, pbxt_database);
+					xt_xres_start_database_recovery(self);
 				}
 				catch_(b) {
 					if (!curr_thd && thd)
@@ -1221,28 +1221,29 @@
 			 * I have to stop the freeer here because it was
 			 * started before opening the database.
 			 */
-			pbxt_call_exit(self);
-			pbxt_inited = FALSE;
-			xt_exit_threading(self);
-			goto error_4;
+
+			/* {FREEER-HANG-ON-INIT-ERROR}
+			 * pbxt_init is called with LOCK_plugin and if it fails and tries to exit
+			 * the freeer here it hangs because the freeer calls THD::~THD which tries
+			 * to aquire the same lock and hangs. OTOH MySQL calls pbxt_end() after
+			 * an unsuccessful call to pbxt_init, so we defer cleaup, except 
+			 * releasing 'self'
+			 */
+			xt_free_thread(self);
+			goto error_3;
 		}
 		xt_free_thread(self);
  	}
 	XT_RETURN(init_err);
 
-	error_4:
-	xt_exit_memory();
-
 	error_3:
 #ifdef XT_STREAMING
 	xt_exit_streaming();
 
 	error_2:
 #endif
-	xt_exit_logging();
 
 	error_1:
-	xt_p_mutex_destroy(&pbxt_database_mutex);		
 	XT_RETURN(1);
 }
 
@@ -1261,7 +1262,7 @@
 		XTExceptionRec	e;
 
 		/* This flag also means "shutting down". */
-		pbxt_inited = FALSE;
+		pbxt_inited = FALSE; 
 		self = xt_create_thread("TempForEnd", FALSE, TRUE, &e);
 		if (self) {
 			self->t_main = TRUE;

=== modified file 'src/restart_xt.cc'
--- src/restart_xt.cc	2009-05-19 14:13:25 +0000
+++ src/restart_xt.cc	2009-06-03 14:07:54 +0000
@@ -3188,3 +3188,36 @@
 	done:
 	db->db_xlog.xlog_seq_exit(&seq);
 }
+
+/* ----------------------------------------------------------------------
+ * D A T A B A S E   R E C O V E R Y   T H R E A D
+ */
+
+extern XTDatabaseHPtr pbxt_database;
+
+static void *xn_xres_run_recovery_thread(XTThreadPtr self)
+{
+	THD *mysql_thread;
+
+	mysql_thread = (THD *)myxt_create_thread();
+
+	while(!ha_resolve_by_legacy_type(mysql_thread, DB_TYPE_PBXT))
+		xt_sleep_milli_second(1);
+
+	xt_open_database(self, mysql_real_data_home, TRUE);
+	pbxt_database = self->st_database;
+	xt_heap_reference(self, pbxt_database);
+	myxt_destroy_thread(mysql_thread, TRUE);
+
+	return NULL;
+}
+
+xtPublic void xt_xres_start_database_recovery(XTThreadPtr self)
+{
+	char name[PATH_MAX];
+
+	sprintf(name, "DB-RECOVERY-%s", xt_last_directory_of_path(mysql_real_data_home));
+	xt_remove_dir_char(name);
+	XTThreadPtr thread = xt_create_daemon(self, name);
+	xt_run_thread(self, thread, xn_xres_run_recovery_thread);
+}

=== modified file 'src/restart_xt.h'
--- src/restart_xt.h	2009-02-23 19:34:56 +0000
+++ src/restart_xt.h	2009-06-03 13:13:32 +0000
@@ -131,4 +131,6 @@
 void xt_print_log_record(xtLogID log, off_t offset, XTXactLogBufferDPtr record);
 void xt_dump_xlogs(struct XTDatabase *db, xtLogID start_log);
 
+xtPublic void xt_xres_start_database_recovery(XTThreadPtr self, const char *path);
+
 #endif