MDEV-34705: Binlog-in-engine: Implement RESET MASTER

Signed-off-by: Kristian Nielsen <knielsen@knielsen-hq.org>

mysql-test/suite/binlog_in_engine/binlog_flush_purge.test

@@ -1,6 +1,8 @@
 --source include/have_binlog_format_row.inc
 --source include/have_innodb_binlog.inc
 
+RESET MASTER;
+
 CREATE TABLE t1 (a INT PRIMARY KEY) ENGINE=InnoDB;
 CREATE TABLE t2 (a INT PRIMARY KEY, b VARCHAR(2048)) ENGINE=InnoDB;
 INSERT INTO t1 VALUES (1);
@@ -18,4 +20,16 @@ SHOW BINARY LOGS;
 FLUSH BINARY LOGS;
 SHOW BINARY LOGS;
 
+RESET MASTER;
+# ToDo: This will be racy, the second binlog file binlog-000001.ibb is
+# pre-allocated in the background, so will be visible or not depending on exact
+# timing. So just omit this SHOW BINARY LOGS or wait for both to be created
+# or something.
+SHOW BINARY LOGS;
+
+INSERT INTO t1 VALUES (100);
+INSERT INTO t2 VALUES (100, 'xyzzy');
+
 DROP TABLE t1, t2;
+
+--exec $MYSQL_BINLOG --read-from-remote-server --user=root --host=127.0.0.1 --port=$MASTER_MYPORT master-bin.000001 --start-position=0-1-1 > $MYSQLTEST_VARDIR/tmp/mysqlbinlog.txt

mysql-test/suite/binlog_in_engine/binlog_in_engine.test

@@ -6,6 +6,8 @@
 # files and only need to include the one.
 --source include/have_innodb_binlog.inc
 
+RESET MASTER;
+
 CREATE TABLE t1 (a INT PRIMARY KEY) ENGINE=InnoDB;
 --let $gtid_pos= `SELECT @@last_gtid`
 INSERT INTO t1 VALUES (1);

mysql-test/suite/binlog_in_engine/binlog_in_engine2.test

@@ -1,6 +1,9 @@
 --source include/have_binlog_format_mixed.inc
 --source include/have_innodb_binlog.inc
 
+# RESET MASTER deliberately omitted here for now; the test should be able to
+# work on top of any existing binlog.
+
 CREATE TABLE sbtest1(
   id INTEGER NOT NULL AUTO_INCREMENT,
   k INTEGER DEFAULT '0' NOT NULL,

mysql-test/suite/binlog_in_engine/binlog_in_engine_restart.test

@@ -6,6 +6,8 @@
 # Note: This test also tests the --binlog-directory option by putting it
 # in binlog_in_engine_restart.opt .
 
+RESET MASTER;
+
 CREATE TABLE t1 (a INT PRIMARY KEY, b INT) ENGINE=InnoDB;
 INSERT INTO t1 VALUES (1, 0);
 let $i= 0;

sql/handler.h

@@ -1570,6 +1570,8 @@ struct handlerton
     Used to implement FLUSH BINARY LOGS.
   */
   bool (*binlog_flush)();
+  /* Engine implementation of RESET MASTER. */
+  bool (*reset_binlogs)();
 
    /*
      Optional clauses in the CREATE/ALTER TABLE

sql/log.cc

@@ -4466,6 +4466,17 @@ bool MYSQL_BIN_LOG::reset_logs(THD *thd, bool create_new_log,
       DBUG_RETURN(1);
     }
 
+    if (opt_binlog_engine_hton)
+    {
+      if (next_log_number)
+      {
+        my_error(ER_ENGINE_BINLOG_NO_RESET_FILE_NUMBER, MYF(0));
+        DBUG_RETURN(true);
+      }
+      DBUG_ASSERT(create_new_log);
+      DBUG_RETURN(reset_engine_binlogs(thd, init_state, init_state_len));
+    }
+
     /*
       Mark that a RESET MASTER is in progress.
       This ensures that a binlog checkpoint will not try to write binlog
@@ -4728,6 +4739,29 @@ err:
 }
 
 
+bool
+MYSQL_BIN_LOG::reset_engine_binlogs(THD *thd, rpl_gtid *init_state,
+                                    uint32 init_state_len)
+{
+  bool err;
+  DBUG_ASSERT(!is_relay_log);
+
+  mysql_mutex_lock(&LOCK_log);
+  mysql_mutex_lock(&LOCK_index);
+
+  err= (*opt_binlog_engine_hton->reset_binlogs)();
+  if (init_state)
+    rpl_global_gtid_binlog_state.load(init_state, init_state_len);
+  else
+    rpl_global_gtid_binlog_state.reset();
+
+  mysql_mutex_unlock(&LOCK_index);
+  mysql_mutex_unlock(&LOCK_log);
+
+  return err;
+}
+
+
 void MYSQL_BIN_LOG::wait_for_last_checkpoint_event()
 {
   mysql_mutex_lock(&LOCK_xid_list);

sql/log.h

@@ -1100,6 +1100,8 @@ public:
   bool reset_logs(THD* thd, bool create_new_log,
                   rpl_gtid *init_state, uint32 init_state_len,
                   ulong next_log_number);
+  bool reset_engine_binlogs(THD *thd, rpl_gtid *init_state,
+                            uint32 init_state_len);
   void wait_for_last_checkpoint_event();
   void close(uint exiting);
   void clear_inuse_flag_when_closing(File file);

sql/share/errmsg-utf8.txt

@@ -12276,3 +12276,5 @@ ER_CANNOT_INIT_ENGINE_BINLOG_READER
         eng "Cannot initialize binlog reader from storage engine %s"
 ER_ENGINE_BINLOG_REQUIRES_GTID
         eng "GTID starting position is required on master with --binlog-storage-engine enabled"
+ER_ENGINE_BINLOG_NO_RESET_FILE_NUMBER
+	eng "RESET MASTER TO is not available when --binlog-storage-engine is enabled"

storage/innobase/fsp/fsp_binlog.cc

@@ -104,6 +104,14 @@ fsp_binlog_init()
 }
 
 
+void
+fsp_binlog_shutdown()
+{
+  pthread_cond_destroy(&active_binlog_cond);
+  mysql_mutex_destroy(&active_binlog_mutex);
+}
+
+
 /** Write out all pages, flush, and close/detach a binlog tablespace.
 @param[in] file_no	 Index of the binlog tablespace
 @return DB_SUCCESS or error code */
@@ -739,6 +747,7 @@ binlog_chunk_reader::fetch_current_page()
         actually maintaining that, to save unnecessary buffer pool
         lookup.
       */
+  goto_next_file:
       if (cur_file_handle >= (File)0)
       {
         my_close(cur_file_handle, MYF(0));
@@ -754,6 +763,8 @@ binlog_chunk_reader::fetch_current_page()
                          s.page_no << srv_page_size_shift, MYF(MY_WME));
     if (res == (size_t)-1)
       return CHUNK_READER_ERROR;
+    if (res == 0 && my_errno == HA_ERR_FILE_TOO_SHORT)
+      goto goto_next_file;
     page_ptr= page_buffer;
     return CHUNK_READER_FOUND;
   }

storage/innobase/handler/ha_innodb.cc

@@ -4137,6 +4137,7 @@ static int innodb_init(void* p)
         innobase_hton->get_binlog_reader= innodb_get_binlog_reader;
         innobase_hton->get_binlog_file_list= innodb_get_binlog_file_list;
         innobase_hton->binlog_flush= innodb_binlog_flush;
+        innobase_hton->reset_binlogs= innodb_reset_binlogs;
 
 	innodb_remember_check_sysvar_funcs();
 

storage/innobase/handler/innodb_binlog.cc

@@ -430,6 +430,37 @@ innodb_binlog_startup_init()
 }
 
 
+static void
+innodb_binlog_init_state()
+{
+  first_open_binlog_file_no= ~(uint64_t)0;
+  binlog_cur_end_offset[0].store(~(uint64_t)0, std::memory_order_relaxed);
+  binlog_cur_end_offset[1].store(~(uint64_t)0, std::memory_order_relaxed);
+  last_created_binlog_file_no= ~(uint64_t)0;
+  active_binlog_file_no.store(~(uint64_t)0, std::memory_order_release);
+  active_binlog_space= nullptr;
+  binlog_cur_page_no= 0;
+  binlog_cur_page_offset= FIL_PAGE_DATA;
+  current_binlog_state_interval= innodb_binlog_state_interval;
+}
+
+
+/* Start the thread that pre-allocates new binlog files. */
+static void
+start_binlog_prealloc_thread()
+{
+  prealloc_thread_end= false;
+  binlog_prealloc_thr_obj= std::thread{innodb_binlog_prealloc_thread};
+
+  mysql_mutex_lock(&active_binlog_mutex);
+  while (last_created_binlog_file_no == ~(uint64_t)0) {
+    /* Wait for the first binlog file to be available. */
+    my_cond_wait(&active_binlog_cond, &active_binlog_mutex.m_mutex);
+  }
+  mysql_mutex_unlock(&active_binlog_mutex);
+}
+
+
 /*
   Open the InnoDB binlog implementation.
   This is called from server binlog layer if the user configured the binlog to
@@ -462,15 +493,8 @@ innodb_binlog_init(size_t binlog_size, const char *directory)
   }
   innodb_binlog_directory= directory;
 
-  first_open_binlog_file_no= ~(uint64_t)0;
-  binlog_cur_end_offset[0].store(~(uint64_t)0, std::memory_order_relaxed);
-  binlog_cur_end_offset[1].store(~(uint64_t)0, std::memory_order_relaxed);
-  last_created_binlog_file_no= ~(uint64_t)0;
-  active_binlog_file_no.store(~(uint64_t)0, std::memory_order_release);
-  active_binlog_space= nullptr;
-  binlog_cur_page_no= 0;
-  binlog_cur_page_offset= FIL_PAGE_DATA;
-  current_binlog_state_interval= innodb_binlog_state_interval;
+  innodb_binlog_init_state();
+
   /* Find any existing binlog files and continue writing in them. */
   int res= innodb_binlog_discover();
   if (res < 0)
@@ -485,15 +509,7 @@ innodb_binlog_init(size_t binlog_size, const char *directory)
       return true;
   }
 
-  /* Start pre-allocating new binlog files. */
-  binlog_prealloc_thr_obj= std::thread{innodb_binlog_prealloc_thread};
-
-  mysql_mutex_lock(&active_binlog_mutex);
-  while (last_created_binlog_file_no == ~(uint64_t)0) {
-    /* Wait for the first binlog file to be available. */
-    my_cond_wait(&active_binlog_cond, &active_binlog_mutex.m_mutex);
-  }
-  mysql_mutex_unlock(&active_binlog_mutex);
+  start_binlog_prealloc_thread();
 
   return false;
 }
@@ -763,7 +779,7 @@ innodb_binlog_discover()
 }
 
 
-void innodb_binlog_close()
+void innodb_binlog_close(bool shutdown)
 {
   if (binlog_prealloc_thr_obj.joinable()) {
     mysql_mutex_lock(&active_binlog_mutex);
@@ -782,17 +798,11 @@ void innodb_binlog_close()
       }
     }
   }
-  /*
-    ToDo: This doesn't seem to free all memory. I'm still getting leaks in eg. --valgrind. Find out why and fix. Example:
-==3464576==    at 0x48407B4: malloc (vg_replace_malloc.c:381)
-==3464576==    by 0x15318CD: mem_strdup(char const*) (mem0mem.inl:452)
-==3464576==    by 0x15321DF: fil_space_t::add(char const*, pfs_os_file_t, unsigned int, bool, bool, unsigned int) (fil0fil.cc:306)
-==3464576==    by 0x1558445: fsp_binlog_tablespace_create(unsigned long) (fsp0fsp.cc:3900)
-==3464576==    by 0x1558C70: fsp_binlog_write_cache(st_io_cache*, unsigned long, mtr_t*) (fsp0fsp.cc:4013)
-  */
-  binlog_diff_state.free();
-  pthread_cond_destroy(&active_binlog_cond);
-  mysql_mutex_destroy(&active_binlog_mutex);
+  if (shutdown)
+  {
+    binlog_diff_state.free();
+    fsp_binlog_shutdown();
+  }
 }
 
 
@@ -2176,3 +2186,49 @@ innodb_find_binlogs(uint64_t *out_first, uint64_t *out_last)
   *out_last= last_file_no;
   return false;
 }
+
+
+bool
+innodb_reset_binlogs()
+{
+  bool err= false;
+
+  /* Close existing binlog tablespaces and stop the pre-alloc thread. */
+  innodb_binlog_close(false);
+
+  /* Delete all binlog files in the directory. */
+  MY_DIR *dir= my_dir(innodb_binlog_directory, MYF(MY_WME));
+  if (!dir)
+  {
+    ib::error() << "Could not read the binlog directory '" <<
+      innodb_binlog_directory << "', error code " << my_errno << ".";
+    err= true;
+  }
+  else
+  {
+    size_t num_entries= dir->number_of_files;
+    fileinfo *entries= dir->dir_entry;
+    for (size_t i= 0; i < num_entries; ++i) {
+      const char *name= entries[i].name;
+      uint64_t file_no;
+      if (!is_binlog_name(name, &file_no))
+        continue;
+      char full_path[OS_FILE_MAX_PATH];
+      binlog_name_make(full_path, file_no);
+      if (my_delete(full_path, MYF(MY_WME)))
+        err= true;
+    }
+    my_dirend(dir);
+  }
+  /*
+    If we get an error deleting any of the existing files, we report the error
+    back up. But we still try to initialize an empty binlog state, better than
+    leaving a non-functional binlog with corrupt internal state.
+  */
+
+  /* Re-initialize empty binlog state and start the pre-alloc thread. */
+  innodb_binlog_init_state();
+  start_binlog_prealloc_thread();
+
+  return err;
+}

storage/innobase/include/fsp_binlog.h

@@ -202,6 +202,7 @@ extern std::atomic<uint64_t> binlog_cur_written_offset[2];
 extern std::atomic<uint64_t> binlog_cur_end_offset[2];
 
 extern void fsp_binlog_init();
+extern void fsp_binlog_shutdown();
 extern dberr_t fsp_binlog_tablespace_close(uint64_t file_no);
 extern fil_space_t *fsp_binlog_open(const char *file_name, pfs_os_file_t fh,
                                     uint64_t file_no, size_t file_size,

storage/innobase/include/innodb_binlog.h

@@ -85,7 +85,7 @@ binlog_name_make(char name_buf[OS_FILE_MAX_PATH], uint64_t file_no)
 
 extern void innodb_binlog_startup_init();
 extern bool innodb_binlog_init(size_t binlog_size, const char *directory);
-extern void innodb_binlog_close();
+extern void innodb_binlog_close(bool shutdown);
 extern bool binlog_gtid_state(rpl_binlog_state_base *state, mtr_t *mtr,
                               buf_block_t * &block, uint32_t &page_no,
                               uint32_t &page_offset, fil_space_t *space);
@@ -98,5 +98,6 @@ extern bool innobase_binlog_write_direct
   (IO_CACHE *cache, handler_binlog_event_group_info *binlog_info,
    const rpl_gtid *gtid);
 extern bool innodb_find_binlogs(uint64_t *out_first, uint64_t *out_last);
+extern bool innodb_reset_binlogs();
 
 #endif /* innodb_binlog_h */

storage/innobase/srv/srv0start.cc

@@ -2066,7 +2066,7 @@ void innodb_shutdown()
 		logs_empty_and_mark_files_at_shutdown();
 	}
 
-        innodb_binlog_close();
+        innodb_binlog_close(true);
 	os_aio_free();
 	fil_space_t::close_all();
 	/* Exit any remaining threads. */