Merge lp:codership-mysql/5.5 -r3961..3980.

cmake/wsrep.cmake

@@ -18,7 +18,7 @@
 # so WSREP_VERSION is produced regardless
 
 # Set the patch version
-SET(WSREP_PATCH_VERSION "9")
+SET(WSREP_PATCH_VERSION "10")
 
 # MariaDB addition: Revision number of the last revision merged from
 # codership branch visible in @@visible_comment.

scripts/CMakeLists.txt

@@ -328,7 +328,13 @@ IF(WIN32)
   ENDFOREACH()
 ELSE()
   IF(WITH_WSREP)
-    SET(WSREP_BINARIES wsrep_sst_common wsrep_sst_mysqldump wsrep_sst_rsync wsrep_sst_xtrabackup)
+    SET(WSREP_BINARIES
+      wsrep_sst_common
+      wsrep_sst_mysqldump
+      wsrep_sst_rsync
+      wsrep_sst_xtrabackup
+      wsrep_sst_xtrabackup-v2
+    )
   ENDIF()
 
   # Configure this one, for testing, but do not install it.

scripts/wsrep_sst_common

@@ -28,7 +28,7 @@ case "$1" in
         shift
         ;;
     '--auth')
-        readonly WSREP_SST_OPT_AUTH="$2"
+        WSREP_SST_OPT_AUTH="$2"
         shift
         ;;
     '--bypass')
@@ -87,13 +87,21 @@ shift
 done
 readonly WSREP_SST_OPT_BYPASS
 
-if [ -n "$WSREP_SST_OPT_DATA" ]
+# For Bug:1200727
+if my_print_defaults -c $WSREP_SST_OPT_CONF sst | grep -q "wsrep_sst_auth";then 
+    if [ -z $WSREP_SST_OPT_AUTH -o $WSREP_SST_OPT_AUTH = "(null)" ];then 
+            WSREP_SST_OPT_AUTH=$(my_print_defaults -c $WSREP_SST_OPT_CONF sst | grep -- "--wsrep_sst_auth" | cut -d= -f2)
+    fi
+fi
+
+if [ -n "${WSREP_SST_OPT_DATA:-}" ]
 then
     SST_PROGRESS_FILE="$WSREP_SST_OPT_DATA/sst_in_progress"
 else
     SST_PROGRESS_FILE=""
 fi
 
+
 wsrep_log()
 {
     # echo everything to stderr so that it gets into common error log

scripts/wsrep_sst_common.sh

@@ -28,7 +28,7 @@ case "$1" in
         shift
         ;;
     '--auth')
-        readonly WSREP_SST_OPT_AUTH="$2"
+        WSREP_SST_OPT_AUTH="$2"
         shift
         ;;
     '--bypass')
@@ -87,13 +87,21 @@ shift
 done
 readonly WSREP_SST_OPT_BYPASS
 
-if [ -n "$WSREP_SST_OPT_DATA" ]
+# For Bug:1200727
+if my_print_defaults -c $WSREP_SST_OPT_CONF sst | grep -q "wsrep_sst_auth";then 
+    if [ -z $WSREP_SST_OPT_AUTH -o $WSREP_SST_OPT_AUTH = "(null)" ];then 
+            WSREP_SST_OPT_AUTH=$(my_print_defaults -c $WSREP_SST_OPT_CONF sst | grep -- "--wsrep_sst_auth" | cut -d= -f2)
+    fi
+fi
+
+if [ -n "${WSREP_SST_OPT_DATA:-}" ]
 then
     SST_PROGRESS_FILE="$WSREP_SST_OPT_DATA/sst_in_progress"
 else
     SST_PROGRESS_FILE=""
 fi
 
+
 wsrep_log()
 {
     # echo everything to stderr so that it gets into common error log

scripts/wsrep_sst_rsync.sh

@@ -49,19 +49,21 @@ check_pid()
 check_pid_and_port()
 {
     local pid_file=$1
-    local rsync_pid=$(cat $pid_file)
-    local rsync_port=$2
-
-    if [ "$OS" == "Darwin" -o "$OS" == "FreeBSD" ]; then
-        # no netstat --program(-p) option in Darwin and FreeBSD
-        check_pid $pid_file && \
-        lsof -i -Pn 2>/dev/null | \
-        grep "(LISTEN)" | grep ":$rsync_port" | grep -w '^rsync[[:space:]]\+'"$rsync_pid" >/dev/null
-    else
-        check_pid $pid_file && \
-        netstat -lnpt 2>/dev/null | \
-        grep LISTEN | grep \:$rsync_port | grep $rsync_pid/rsync >/dev/null
+    local rsync_pid=$2
+    local rsync_port=$3
+
+    local port_info=$(lsof -i :$rsync_port -Pn 2>/dev/null | \
+        grep "(LISTEN)")
+    local is_rsync=$(echo $port_info | \
+        grep -w '^rsync[[:space:]]\+'"$rsync_pid" 2>/dev/null)
+
+    if [ -n "$port_info" -a -z "$is_rsync" ]; then
+        wsrep_log_error "rsync daemon port '$rsync_port' has been taken"
+        exit 16 # EBUSY
     fi
+    check_pid $pid_file && \
+        [ -n "$port_info" ] && [ -n "$is_rsync" ] && \
+        [ $(cat $pid_file) -eq $rsync_pid ]
 }
 
 MAGIC_FILE="$WSREP_SST_OPT_DATA/rsync_sst_complete"
@@ -234,9 +236,10 @@ EOF
 #    rm -rf "$DATA"/ib_logfile* # we don't want old logs around
 
     # listen at all interfaces (for firewalled setups)
-    rsync --daemon --port $RSYNC_PORT --config "$RSYNC_CONF"
+    rsync --daemon --no-detach --port $RSYNC_PORT --config "$RSYNC_CONF" &
+    RSYNC_REAL_PID=$!
 
-    until check_pid_and_port $RSYNC_PID $RSYNC_PORT
+    until check_pid_and_port $RSYNC_PID $RSYNC_REAL_PID $RSYNC_PORT
     do
         sleep 0.2
     done

sql/slave.cc

@@ -3456,12 +3456,18 @@ pthread_handler_t handle_slave_sql(void *arg)
   my_off_t UNINIT_VAR(saved_log_pos);
   my_off_t UNINIT_VAR(saved_master_log_pos);
   my_off_t saved_skip= 0;
+#ifdef WITH_WSREP
+  my_bool wsrep_node_dropped= FALSE;
+#endif /* WITH_WSREP */
   Relay_log_info* rli = &((Master_info*)arg)->rli;
   const char *errmsg;
 
   // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
   my_thread_init();
   DBUG_ENTER("handle_slave_sql");
+#ifdef WITH_WSREP
+ wsrep_restart_point:
+#endif /* WITH_WSREP */
 
   LINT_INIT(saved_master_log_pos);
   LINT_INIT(saved_log_pos);
@@ -3727,6 +3733,12 @@ log '%s' at position %s, relay log '%s' position: %s", RPL_LOG_NAME,
 Error running query, slave SQL thread aborted. Fix the problem, and restart \
 the slave SQL thread with \"SLAVE START\". We stopped at log \
 '%s' position %s", RPL_LOG_NAME, llstr(rli->group_master_log_pos, llbuff));
+#ifdef WITH_WSREP
+        if (WSREP_ON && last_errno == ER_UNKNOWN_COM_ERROR)
+        {
+	  wsrep_node_dropped= TRUE;
+	}
+#endif /* WITH_WSREP */
       }
       goto err;
     }
@@ -3791,6 +3803,27 @@ the slave SQL thread with \"SLAVE START\". We stopped at log \
   THD_CHECK_SENTRY(thd);
   delete thd;
   mysql_mutex_unlock(&LOCK_thread_count);
+#ifdef WITH_WSREP
+  /* if slave stopped due to node going non primary, we set global flag to
+     trigger automatic restart of slave when node joins back to cluster
+  */
+   if (wsrep_node_dropped && wsrep_restart_slave)
+   {
+     if (wsrep_ready)
+     {
+       WSREP_INFO("Slave error due to node temporarily non-primary"
+		  "SQL slave will continue");
+       wsrep_node_dropped= FALSE;
+       mysql_mutex_unlock(&rli->run_lock);
+       goto wsrep_restart_point;
+     } else {
+       WSREP_INFO("Slave error due to node going non-primary");
+       WSREP_INFO("wsrep_restart_slave was set and therefore slave will be "
+		  "automatically restarted when node joins back to cluster");
+       wsrep_restart_slave_activated= TRUE;
+     }
+   }
+#endif /* WITH_WSREP */
  /*
   Note: the order of the broadcast and unlock calls below (first broadcast, then unlock)
   is important. Otherwise a killer_thread can execute between the calls and

sql/sql_base.cc

@@ -5181,6 +5181,7 @@ bool open_tables(THD *thd, TABLE_LIST **start, uint *counter, uint flags,
        thd->lex->sql_command== SQLCOM_LOAD           ||
        thd->lex->sql_command== SQLCOM_DELETE)        &&
       wsrep_replicate_myisam                         &&
+      (*start)                                       &&
       (*start)->table && (*start)->table->file->ht->db_type == DB_TYPE_MYISAM)
     {
       WSREP_TO_ISOLATION_BEGIN(NULL, NULL, (*start));

sql/sys_vars.cc

@@ -3918,6 +3918,10 @@ static Sys_var_mybool Sys_wsrep_load_data_splitting(
        "transaction after every 10K rows inserted",
        GLOBAL_VAR(wsrep_load_data_splitting), 
        CMD_LINE(OPT_ARG), DEFAULT(TRUE));
+
+static Sys_var_mybool Sys_wsrep_restart_slave(
+       "wsrep_restart_slave", "Should MySQL slave be restarted automatically, when node joins back to cluster",
+       GLOBAL_VAR(wsrep_restart_slave), CMD_LINE(OPT_ARG), DEFAULT(FALSE));
 #endif /* WITH_WSREP */
 
 static Sys_var_charptr Sys_ignore_db_dirs(

sql/wsrep_binlog.cc

@@ -320,3 +320,59 @@ void wsrep_dump_rbr_buf(THD *thd, const void* rbr_buf, size_t buf_len)
   }
 }
 
+void wsrep_dump_rbr_direct(THD* thd, IO_CACHE* cache)
+{
+  char filename[PATH_MAX]= {0};
+  int len= snprintf(filename, PATH_MAX, "%s/GRA_%ld_%lld.log",
+                    wsrep_data_home_dir, thd->thread_id,
+                    (long long)wsrep_thd_trx_seqno(thd));
+  size_t bytes_in_cache = 0;
+  // check path
+  if (len >= PATH_MAX)
+  {
+    WSREP_ERROR("RBR dump path too long: %d, skipping dump.", len);
+    return ;
+  }
+  // init cache
+  my_off_t const saved_pos(my_b_tell(cache));
+  if (reinit_io_cache(cache, READ_CACHE, 0, 0, 0))
+  {
+    WSREP_ERROR("failed to initialize io-cache");
+    return ;
+  }
+  // open file
+  FILE* of = fopen(filename, "wb");
+  if (!of)
+  {
+    WSREP_ERROR("Failed to open file '%s': %d (%s)",
+                filename, errno, strerror(errno));
+    goto cleanup;
+  }
+  // ready to write
+  bytes_in_cache= my_b_bytes_in_cache(cache);
+  if (unlikely(bytes_in_cache == 0)) bytes_in_cache = my_b_fill(cache);
+  if (likely(bytes_in_cache > 0)) do
+  {
+    if (my_fwrite(of, cache->read_pos, bytes_in_cache,
+                  MYF(MY_WME | MY_NABP)) == (size_t) -1)
+    {
+      WSREP_ERROR("Failed to write file '%s'", filename);
+      goto cleanup;
+    }
+    cache->read_pos= cache->read_end;
+  } while ((cache->file >= 0) && (bytes_in_cache= my_b_fill(cache)));
+  if(cache->error == -1)
+  {
+    WSREP_ERROR("RBR inconsistent");
+    goto cleanup;
+  }
+cleanup:
+  // init back
+  if (reinit_io_cache(cache, WRITE_CACHE, saved_pos, 0, 0))
+  {
+    WSREP_ERROR("failed to reinitialize io-cache");
+  }
+  // close file
+  if (of) fclose(of);
+}
+

sql/wsrep_binlog.h

@@ -46,4 +46,7 @@ int wsrep_write_cache (wsrep_t*  wsrep,
 /* Dump replication buffer to disk */
 void wsrep_dump_rbr_buf(THD *thd, const void* rbr_buf, size_t buf_len);
 
+/* Dump replication buffer to disk without intermediate buffer */
+void wsrep_dump_rbr_direct(THD* thd, IO_CACHE* cache);
+
 #endif /* WSREP_BINLOG_H */

sql/wsrep_hton.cc

@@ -505,11 +505,17 @@ wsrep_run_wsrep_commit(THD *thd, handlerton *hton, bool all)
 
     DBUG_RETURN(WSREP_TRX_CERT_FAIL);
 
+  case WSREP_SIZE_EXCEEDED:
+    WSREP_ERROR("transaction size exceeded");
+    mysql_mutex_unlock(&thd->LOCK_wsrep_thd);
+    DBUG_RETURN(WSREP_TRX_SIZE_EXCEEDED);
   case WSREP_CONN_FAIL:
     WSREP_ERROR("connection failure");
+    mysql_mutex_unlock(&thd->LOCK_wsrep_thd);
     DBUG_RETURN(WSREP_TRX_ERROR);
   default:
     WSREP_ERROR("unknown connection failure");
+    mysql_mutex_unlock(&thd->LOCK_wsrep_thd);
     DBUG_RETURN(WSREP_TRX_ERROR);
   }
 

sql/wsrep_mysqld.cc

@@ -26,6 +26,7 @@
 #include <cstdio>
 #include <cstdlib>
 #include "log_event.h"
+#include <slave.h>
 
 Format_description_log_event *wsrep_format_desc = NULL;
 wsrep_t *wsrep                  = NULL;
@@ -59,7 +60,10 @@ ulong   wsrep_mysql_replication_bundle = 0;
 my_bool wsrep_desync                   = 0; // desynchronize the node from the
                                             // cluster
 my_bool wsrep_load_data_splitting      = 1; // commit load data every 10K intervals
-
+my_bool wsrep_restart_slave            = 0; // should mysql slave thread be
+                                            // restarted, if node joins back
+my_bool wsrep_restart_slave_activated  = 0; // node has dropped, and slave
+                                            // restart will be needed
 /*
  * End configuration options
  */
@@ -406,6 +410,25 @@ static void wsrep_synced_cb(void* app_ctx)
       // and wait for SE initialization
       wsrep_SE_init_wait();
   }
+  if (wsrep_restart_slave_activated)
+  {
+    int rcode;
+    WSREP_INFO("MySQL slave restart");
+    wsrep_restart_slave_activated= FALSE;
+
+    mysql_mutex_lock(&LOCK_active_mi);
+    if ((rcode = start_slave_threads(1 /* need mutex */,
+                            0 /* no wait for start*/,
+                            active_mi,
+                            master_info_file,
+                            relay_log_info_file,
+                       	    SLAVE_SQL)))
+    {
+      WSREP_WARN("Failed to create slave threads: %d", rcode);
+    }
+    mysql_mutex_unlock(&LOCK_active_mi);
+
+  }
 }
 
 static void wsrep_init_position()
@@ -627,6 +650,7 @@ int wsrep_init()
   {
     DBUG_PRINT("wsrep",("wsrep::init() failed: %d", rcode));
     WSREP_ERROR("wsrep::init() failed: %d, must shutdown", rcode);
+    wsrep->free(wsrep);
     free(wsrep);
     wsrep = NULL;
   } else {
@@ -1159,7 +1183,7 @@ create_view_query(THD *thd, uchar** buf, size_t* buf_len)
     }
     buff.append(STRING_WITH_LEN(" AS "));
     //buff.append(views->source.str, views->source.length);
-    buff.append(thd->lex->create_view_select.str, 
+    buff.append(thd->lex->create_view_select.str,
                 thd->lex->create_view_select.length);
     //int errcode= query_error_code(thd, TRUE);
     //if (thd->binlog_query(THD::STMT_QUERY_TYPE,
@@ -1168,7 +1192,7 @@ create_view_query(THD *thd, uchar** buf, size_t* buf_len)
 }
 
 static int wsrep_TOI_begin(THD *thd, char *db_, char *table_,
-                           const TABLE_LIST* table_list) 
+                           const TABLE_LIST* table_list)
 {
   wsrep_status_t ret(WSREP_WARNING);
   uchar* buf(0);
@@ -1193,7 +1217,7 @@ static int wsrep_TOI_begin(THD *thd, char *db_, char *table_,
     buf_err= wsrep_create_event_query(thd, &buf, &buf_len);
     break;
   default:
-    buf_err= wsrep_to_buf_helper(thd, thd->query(), thd->query_length(), &buf, 
+    buf_err= wsrep_to_buf_helper(thd, thd->query(), thd->query_length(), &buf,
                                  &buf_len);
     break;
   }
@@ -1243,7 +1267,7 @@ static void wsrep_TOI_end(THD *thd) {
   }
 }
 
-static int wsrep_RSU_begin(THD *thd, char *db_, char *table_) 
+static int wsrep_RSU_begin(THD *thd, char *db_, char *table_)
 {
   wsrep_status_t ret(WSREP_WARNING);
   WSREP_DEBUG("RSU BEGIN: %lld, %d : %s", (long long)wsrep_thd_trx_seqno(thd),
@@ -1325,9 +1349,9 @@ int wsrep_to_isolation_begin(THD *thd, char *db_, char *table_,
   int ret= 0;
   mysql_mutex_lock(&thd->LOCK_wsrep_thd);
 
-  if (thd->wsrep_conflict_state == MUST_ABORT) 
+  if (thd->wsrep_conflict_state == MUST_ABORT)
   {
-    WSREP_INFO("thread: %lu, %s has been aborted due to multi-master conflict", 
+    WSREP_INFO("thread: %lu, %s has been aborted due to multi-master conflict",
                thd->thread_id, thd->query());
     mysql_mutex_unlock(&thd->LOCK_wsrep_thd);
     return WSREP_TRX_FAIL;
@@ -1346,7 +1370,7 @@ int wsrep_to_isolation_begin(THD *thd, char *db_, char *table_,
 
   if (wsrep_debug && thd->mdl_context.has_locks())
   {
-    WSREP_DEBUG("thread holds MDL locks at TI begin: %s %lu", 
+    WSREP_DEBUG("thread holds MDL locks at TI begin: %s %lu",
                 thd->query(), thd->thread_id);
   }
 
@@ -1438,7 +1462,7 @@ wsrep_grant_mdl_exception(MDL_context *requestor_ctx,
       mysql_mutex_unlock(&granted_thd->LOCK_wsrep_thd);
       ret = TRUE;
     }
-    else if (request_thd->lex->sql_command == SQLCOM_DROP_TABLE) 
+    else if (request_thd->lex->sql_command == SQLCOM_DROP_TABLE)
     {
       WSREP_DEBUG("DROP caused BF abort");
       ticket->wsrep_report(wsrep_debug);
@@ -1446,7 +1470,7 @@ wsrep_grant_mdl_exception(MDL_context *requestor_ctx,
       wsrep_abort_thd((void*)request_thd, (void*)granted_thd, 1);
       ret = FALSE;
     }
-    else if (granted_thd->wsrep_query_state == QUERY_COMMITTING) 
+    else if (granted_thd->wsrep_query_state == QUERY_COMMITTING)
     {
       WSREP_DEBUG("mdl granted, but commiting thd abort scheduled");
       ticket->wsrep_report(wsrep_debug);
@@ -1454,7 +1478,7 @@ wsrep_grant_mdl_exception(MDL_context *requestor_ctx,
       wsrep_abort_thd((void*)request_thd, (void*)granted_thd, 1);
       ret = FALSE;
     }
-    else 
+    else
     {
       WSREP_MDL_LOG(DEBUG, "MDL conflict-> BF abort", request_thd, granted_thd);
       ticket->wsrep_report(wsrep_debug);

sql/wsrep_mysqld.h

@@ -95,6 +95,8 @@ extern my_bool     wsrep_replicate_myisam;
 extern my_bool     wsrep_log_conflicts;
 extern ulong       wsrep_mysql_replication_bundle;
 extern my_bool     wsrep_load_data_splitting;
+extern my_bool     wsrep_restart_slave;
+extern my_bool     wsrep_restart_slave_activated;
 
 enum enum_wsrep_OSU_method { WSREP_OSU_TOI, WSREP_OSU_RSU };
 

sql/wsrep_notify.cc

@@ -75,15 +75,18 @@ void wsrep_notify_status (wsrep_member_status_t    status,
     cmd_off += snprintf (cmd_ptr + cmd_off, cmd_len - cmd_off,
                          " --index %d", view->my_idx);
 
-    cmd_off += snprintf (cmd_ptr + cmd_off, cmd_len - cmd_off, " --members");
-
-    for (int i = 0; i < view->memb_num; i++)
+    if (view->memb_num)
     {
-      wsrep_uuid_print (&view->members[i].id, uuid_str, sizeof(uuid_str));
-      cmd_off += snprintf (cmd_ptr + cmd_off, cmd_len - cmd_off,
-                           "%c%s/%s/%s", i > 0 ? ',' : ' ',
-                           uuid_str, view->members[i].name,
-                           view->members[i].incoming);
+        cmd_off += snprintf (cmd_ptr + cmd_off, cmd_len - cmd_off, " --members");
+
+        for (int i = 0; i < view->memb_num; i++)
+        {
+            wsrep_uuid_print (&view->members[i].id, uuid_str, sizeof(uuid_str));
+            cmd_off += snprintf (cmd_ptr + cmd_off, cmd_len - cmd_off,
+                                 "%c%s/%s/%s", i > 0 ? ',' : ' ',
+                                 uuid_str, view->members[i].name,
+                                 view->members[i].incoming);
+        }
     }
   }
 

sql/wsrep_sst.cc

@@ -934,8 +934,9 @@ static void* sst_donor_thread (void* a)
     else
     {
       WSREP_ERROR("Failed to read from: %s", proc.cmd());
+      proc.wait();
     }
-    if (err && proc.error()) err= proc.error();
+    if (!err && proc.error()) err= proc.error();
   }
   else
   {

support-files/mysql.spec.sh

@@ -1207,6 +1207,7 @@ echo "====="                                     >> $STATUS_HISTORY
 %attr(755, root, root) %{_bindir}/wsrep_sst_rsync
 %attr(755, root, root) %{_bindir}/wsrep_sst_rsync_wan
 %attr(755, root, root) %{_bindir}/wsrep_sst_xtrabackup
+%attr(755, root, root) %{_bindir}/wsrep_sst_xtrabackup-v2
 %endif
 
 %attr(755, root, root) %{_sbindir}/mysqld

support-files/wsrep_notify.sh

@@ -87,6 +87,7 @@ do
     --members)
         MEMBERS=$2
         shift
+        fi
         ;;
     esac
     shift

wsrep/wsrep_api.h

@@ -137,7 +137,11 @@ typedef void (*wsrep_log_cb_t)(wsrep_log_level_t, const char *);
 typedef uint64_t wsrep_trx_id_t;  //!< application transaction ID
 typedef uint64_t wsrep_conn_id_t; //!< application connection ID
 typedef int64_t  wsrep_seqno_t;   //!< sequence number of a writeset, etc.
+#ifdef __cplusplus
+typedef bool     wsrep_bool_t;
+#else
 typedef _Bool    wsrep_bool_t;    //!< should be the same as standard (C99) bool
+#endif /* __cplusplus */
 
 /*! undefined seqno */
 #define WSREP_SEQNO_UNDEFINED (-1)

wsrep/wsrep_dummy.c

@@ -388,4 +388,3 @@ int wsrep_dummy_loader(wsrep_t* w)
 
     return 0;
 }
-