Merge bk-internal.mysql.com:/home/bk/mysql-maria

into  mysql.com:/home/my/mysql-maria


storage/maria/ha_maria.cc:
  Auto merged
storage/maria/ma_bitmap.c:
  Auto merged
storage/maria/ma_checkpoint.c:
  Auto merged
storage/maria/ma_close.c:
  Auto merged
storage/maria/ma_loghandler.c:
  Auto merged
storage/maria/ma_loghandler.h:
  Auto merged
storage/maria/ma_open.c:
  Auto merged
storage/maria/ma_pagecache.h:
  Auto merged
storage/maria/ma_write.c:
  Auto merged
storage/maria/maria_def.h:
  Auto merged
storage/maria/unittest/ma_pagecache_single.c:
  Auto merged
storage/maria/ma_blockrec.c:
  Manual merge
storage/maria/ma_page.c:
  Manual merge
storage/maria/ma_pagecache.c:
  Manual merge
storage/maria/ma_preload.c:
  Manual merge
storage/maria/ma_recovery.c:
  Manual merge
  Add _ma_unpin_all_pages() to all new UNDO redo_exec_hook's

mysql-test/include/maria_empty_logs.inc

+# Maria help script.
+# Cleans up all logs to give recovery a fresh start.
+
+# API: none, just uses vardir, port and socket.
+
+connection admin;
+
+-- echo * shut down mysqld, removed logs, restarted it
+append_file $MYSQLTEST_VARDIR/tmp/master0.expect;
+wait-maria_empty_logs.inc
+EOF
+
+--exec $MYSQLADMIN --no-defaults -S $MASTER_MYSOCK -P $MASTER_MYPORT  -u root --password= shutdown 2>&1;
+
+remove_file $MYSQLTEST_VARDIR/master-data/maria_log_control;
+remove_file $MYSQLTEST_VARDIR/master-data/maria_log.00000001;
+-- error 0,1 # maybe there is just one log
+remove_file $MYSQLTEST_VARDIR/master-data/maria_log.00000002;
+# Hope there were not more than these logs.
+
+-- error 0,1
+remove_file $MYSQLTEST_VARDIR/master-data/maria_recovery.trace;
+
+append_file $MYSQLTEST_VARDIR/tmp/master0.expect;
+restart-maria_empty_logs.inc
+EOF
+
+--source include/wait_until_connected_again.inc
+
+connection default;
+# the effect of "use" is lost after a restart so we are back into db "test",
+# because connection 'default' was created with db "test".
+use mysqltest;

mysql-test/include/maria_make_snapshot.inc

+# Maria helper script
+# Copies table' data and index file to other directory, or back, or compares.
+# The other directory looks like a database directory, so that we can
+# read copies from inside mysqld, that's also why we copy the frm.
+
+# "mms" is a namespace for Maria_Make_Snapshot
+
+# API:
+# 1) set one of
+# $mms_copy : to copy table from database to spare directory
+# $mms_reverse : to copy it back
+# $mms_compare_physically : to compare both byte-for-byte
+# 2) set $mms_table_to_use to a number N: table will be mysqltest.tN
+# 3) set $mms_purpose to say what this copy is for (influences the naming
+# of the spare directory).
+
+if ($mms_copy)
+{
+  --echo * copied t$mms_table_to_use for $mms_purpose
+  copy_file $MYSQLTEST_VARDIR/master-data/mysqltest/t$mms_table_to_use.MAD $MYSQLTEST_VARDIR/master-data/mysqltest_for_$mms_purpose/t$mms_table_to_use.MAD;
+  copy_file $MYSQLTEST_VARDIR/master-data/mysqltest/t$mms_table_to_use.MAI $MYSQLTEST_VARDIR/master-data/mysqltest_for_$mms_purpose/t$mms_table_to_use.MAI;
+  copy_file $MYSQLTEST_VARDIR/master-data/mysqltest/t$mms_table_to_use.frm $MYSQLTEST_VARDIR/master-data/mysqltest_for_$mms_purpose/t$mms_table_to_use.frm;
+}
+
+if ($mms_reverse_copy)
+{
+  # do not call this without flushing target table first!
+  --echo * copied t$mms_table_to_use back for $mms_purpose
+ -- error 0,1
+  remove_file $MYSQLTEST_VARDIR/master-data/mysqltest/t$mms_table_to_use.MAD;
+  copy_file $MYSQLTEST_VARDIR/master-data/mysqltest_for_$mms_purpose/t$mms_table_to_use.MAD $MYSQLTEST_VARDIR/master-data/mysqltest/t$mms_table_to_use.MAD;
+  -- error 0,1
+  remove_file $MYSQLTEST_VARDIR/master-data/mysqltest/t$mms_table_to_use.MAI;
+  copy_file $MYSQLTEST_VARDIR/master-data/mysqltest_for_$mms_purpose/t$mms_table_to_use.MAI $MYSQLTEST_VARDIR/master-data/mysqltest/t$mms_table_to_use.MAI;
+}
+
+if ($mms_compare_physically)
+{
+  # After the UNDO phase this is normally impossible
+  # (UNDO execution has created new log records => pages have new LSNs).
+  # So, do this only when testing REDO phase.
+  # If UNDO phase, we nevertheless compare checksums
+  # (see maria_verify_recovery.inc).
+  --echo * compared t$mms_table_to_use to old version
+  diff_files $MYSQLTEST_VARDIR/master-data/mysqltest/t$mms_table_to_use.MAD $MYSQLTEST_VARDIR/master-data/mysqltest_for_$mms_purpose/t$mms_table_to_use.MAD;
+# index file not yet recovered
+#  diff_files $MYSQLTEST_VARDIR/master-data/mysqltest/t$mms_table_to_use.MAI $MYSQLTEST_VARDIR/master-data/mysqltest_for_$mms_purpose/t$mms_table_to_use.MAI;
+}

mysql-test/include/maria_make_snapshot_for_comparison.inc

+# Maria helper script
+# Copies clean tables' data and index file to other directory
+# Tables are t1...t[$mms_tables]
+# They are later used as a reference to see if recovery works.
+
+# API:
+# set $mms_tables to N, the script will cover tables mysqltest.t1,...tN
+
+connection admin;
+
+let $mms_table_to_use=$mms_tables;
+let $mms_purpose=comparison;
+let $mms_copy=1;
+
+--disable_query_log
+--disable_warnings
+eval drop database if exists mysqltest_for_$mms_purpose;
+--enable_warnings
+eval create database mysqltest_for_$mms_purpose;
+--enable_query_log
+
+while ($mms_table_to_use)
+{
+  # to serve as a reference, table must be in a clean state
+  eval flush table t$mms_table_to_use;
+  -- source include/maria_make_snapshot.inc
+  dec $mms_table_to_use;
+}
+let $mms_copy=0;
+connection default;

mysql-test/include/maria_make_snapshot_for_feeding_recovery.inc

+# Maria helper script
+# Copies tables' data and index file to other directory, and control file.
+# Tables are t1...t[$mms_tables].
+# Later, mysqld is shutdown, and that snapshot is put back into the
+# datadir, control file too ("flashing recovery's brain"), and recovery is let
+# to run on it (see maria_verify_recovery.inc).
+
+# API:
+# set $mms_tables to N, the script will cover tables mysqltest.t1,...tN
+
+connection admin;
+
+let $mms_table_to_use=$mms_tables;
+let $mms_purpose=feeding_recovery;
+let $mms_copy=1;
+
+--disable_query_log
+--disable_warnings
+eval drop database if exists mysqltest_for_$mms_purpose;
+--enable_warnings
+eval create database mysqltest_for_$mms_purpose;
+--enable_query_log
+
+while ($mms_table_to_use)
+{
+  -- source include/maria_make_snapshot.inc
+  dec $mms_table_to_use;
+}
+let $mms_copy=0;
+
+-- error 0,1
+remove_file $MYSQLTEST_VARDIR/tmp/mms_for_$mms_purpose.maria_log_control;
+copy_file $MYSQLTEST_VARDIR/master-data/maria_log_control $MYSQLTEST_VARDIR/tmp/mms_for_$mms_purpose.maria_log_control;
+
+connection default;

mysql-test/include/maria_verify_recovery.inc

+# Maria helper script.
+# Runs recovery, compare with expected table data.
+
+# API:
+# 1) set $mms_tables to N, the script will cover tables mysqltest.t1,...tN
+# 2) set $mvr_debug_option to the crash way
+# 3) set $mvr_restore_old_snapshot to 1 if you want recovery to run on
+# an old copy of tables and of the control file, 0 for normal recovery.
+# 4) set $mms_compare_physically to 1 if you want a physical byte-for-byte
+# comparison with expected table. Checksum comparison is always done.
+
+# "mvr" is a namespace for Maria_Verify_Recovery
+
+connection admin;
+
+# we may do a copy-back  of tables before comparison, so save comparison
+# request made by caller:
+let $mms_compare_physically_save=$mms_compare_physically;
+let $mms_compare_physically=0;
+
+# warn mtr that mysqld is going to die and should not be restarted immediately
+#append_file $MYSQLTEST_VARDIR/tmp/master0.expect;
+#wait-maria_verify_recovery.inc
+#EOF
+# todo: remove this "system" and uncomment above when BUG#32296 is fixed
+system echo wait-maria_verify_recovery.inc >> $MYSQLTEST_VARDIR/tmp/master0.expect;
+
+# flush page cache and log, only log, or nothing, and kill mysqld with
+# abort().
+# When we restore an old snapshot, we could just kill mysqld nicely,
+# but that would implicitely commit all work, which the tester may
+# not want (tester may want to observe rollback happening).
+
+eval SET SESSION debug=$mvr_debug_option;
+--echo * crashing mysqld intentionally
+--error 2013
+set global maria_checkpoint_interval=1; # this will crash (DBUG magic)
+
+if ($mvr_restore_old_snapshot)
+{
+
+  # copy snapshot made by maria_make_snapshot_for_feeding_recovery back
+  # into datadir.
+
+  let $mms_table_to_use=$mms_tables;
+  let $mms_purpose=feeding_recovery;
+  let $mms_reverse_copy=1;
+  while ($mms_table_to_use)
+  {
+    -- source include/maria_make_snapshot.inc
+    dec $mms_table_to_use;
+  }
+  let $mms_reverse_copy=0;
+
+  # also copy back control file, to force recovery to start from an early
+  # point, ignoring further checkpoints.
+  -- error 0,1
+  remove_file $MYSQLTEST_VARDIR/master-data/maria_log_control;
+  copy_file $MYSQLTEST_VARDIR/tmp/mms_for_$mms_purpose.maria_log_control $MYSQLTEST_VARDIR/master-data/maria_log_control;
+}
+
+--echo * recovery happens
+# let mtr restart mysqld (and thus execute the maria log)
+#append_file $MYSQLTEST_VARDIR/tmp/master0.expect;
+#restart-maria_verify_recovery.inc
+#EOF
+system echo restart-maria_verify_recovery.inc >> $MYSQLTEST_VARDIR/tmp/master0.expect;
+
+--source include/wait_until_connected_again.inc
+
+# Compare that tables of $mms_tables are identical to old.
+# We always compare with CHECKSUM TABLE, and if requested (which makes sense
+# only for testing the REDO phase, as UNDO phase generates new records so new
+# LSNs on pages.) with a physical byte-for-byte comparison.
+let $mms_table_to_use=$mms_tables;
+let $mms_purpose=comparison;
+let $mms_compare_physically=$mms_compare_physically_save;
+while ($mms_table_to_use)
+{
+  # Todo: remove this REPAIR when we have index recovery working.
+  # It is a quick repair, so that it will fail if data file is corrupted.
+  --echo * rebuilding index (until we have recovery of index)
+  eval repair table t$mms_table_to_use quick;
+  --echo * testing that checksum after recovery is as expected
+  let $new_checksum=`CHECKSUM TABLE t$mms_table_to_use`;
+  let $old_checksum=`CHECKSUM TABLE mysqltest_for_$mms_purpose.t$mms_table_to_use`;
+  # the $ text variables above are of the form "db.tablename\tchecksum",
+  # as db differs, we use substring().
+  eval select if(substring("$new_checksum",instr("$new_checksum",".t1")) = substring("$old_checksum",instr("$old_checksum",".t1")),"ok","failure");
+  # this script may compare physically or do nothing
+  -- source include/maria_make_snapshot.inc
+  dec $mms_table_to_use;
+}
+
+connection default;
+# the effect of "use" is lost after a restart so we are back into db "test"
+use mysqltest;

mysql-test/include/wait_until_connected_again.inc

 #
 # Include this script to wait until the connection to the
-# server has been restored or timeout occurs
+# server has been restored or timeout occurs.
+# You should have done --enable_reconnect first
 --disable_result_log
 --disable_query_log
 let $counter= 500;
+let $mysql_errno= 1;
 while ($mysql_errno)
 {
-  --error 0,2002,2006
+  --error 0,2002,2003,2006
   show status;
 
   dec $counter;

mysql-test/lib/mtr_process.pl

@@ -886,15 +886,33 @@ sub check_expected_crash_and_restart($)
       mtr_verbose("$mysqld->{'type'} $mysqld->{'idx'} exited, pid: $ret_pid");
       $mysqld->{'pid'}= 0;
 
-      # Check if crash expected and restart if it was
+      # Check if crash expected, and restart if it was
       my $expect_file= "$::opt_vardir/tmp/" . "$mysqld->{'type'}" .
 	"$mysqld->{'idx'}" . ".expect";
+      while ( 1 )
+      {
         if ( -f $expect_file )
         {
           mtr_verbose("Crash was expected, file $expect_file exists");
+          my $expect_file_handler;
+          open($expect_file_handler, "<$expect_file") or die;
+          my @expect_lines= <$expect_file_handler>;
+          close $expect_file_handler;
+          # look at most recent order by the test
+          my $expect_content= pop @expect_lines;
+          chomp $expect_content;
+          if ( $expect_content =~ /^wait/ )
+          {
+            mtr_verbose("Test asks that we wait before restart");
+            # Millisceond sleep emulated with select
+            select(undef, undef, undef, (0.1));
+            next;
+          }
+          unlink($expect_file);
           mysqld_start($mysqld, $mysqld->{'start_opts'},
                        $mysqld->{'start_slave_master_info'});
-	unlink($expect_file);
+        }
+        last;
       }
 
       return;
@@ -914,8 +932,8 @@ sub check_expected_crash_and_restart($)
       if ( -f $expect_file )
       {
 	mtr_verbose("Crash was expected, file $expect_file exists");
-	ndbmgmd_start($cluster);
 	unlink($expect_file);
+	ndbmgmd_start($cluster);
       }
       return;
     }
@@ -933,9 +951,9 @@ sub check_expected_crash_and_restart($)
 	if ( -f $expect_file )
 	{
 	  mtr_verbose("Crash was expected, file $expect_file exists");
+	  unlink($expect_file);
 	  ndbd_start($cluster, $ndbd->{'idx'},
 		     $ndbd->{'start_extra_args'});
-	  unlink($expect_file);
 	}
 	return;
       }

mysql-test/r/maria-recovery.result

+drop database if exists mysqltest;
+create database mysqltest;
+use mysqltest;
+* shut down mysqld, removed logs, restarted it
+use mysqltest;
+create table t1 (a varchar(1000)) engine=maria;
+* TEST of REDO: see if recovery can reconstruct if we give it an old table
+* copied t1 for feeding_recovery
+insert into t1 values ("00000000");
+flush table t1;
+* copied t1 for comparison
+SET SESSION debug="+d,maria_flush_whole_log,maria_crash";
+* crashing mysqld intentionally
+set global maria_checkpoint_interval=1;
+ERROR HY000: Lost connection to MySQL server during query
+* copied t1 back for feeding_recovery
+* recovery happens
+* rebuilding index (until we have recovery of index)
+repair table t1 quick;
+Table	Op	Msg_type	Msg_text
+mysqltest.t1	repair	status	OK
+* testing that checksum after recovery is as expected
+select if(substring("mysqltest.t1	488070860",instr("mysqltest.t1	488070860",".t1")) = substring("mysqltest_for_comparison.t1	488070860",instr("mysqltest_for_comparison.t1	488070860",".t1")),"ok","failure");
+if(substring("mysqltest.t1	488070860",instr("mysqltest.t1	488070860",".t1")) = substring("mysqltest_for_comparison.t1	488070860",instr("mysqltest_for_comparison.t1	488070860",".t1")),"ok","failure")
+ok
+* compared t1 to old version
+use mysqltest;
+select * from t1;
+a
+00000000
+* TEST of REDO+UNDO: normal recovery test (no moving tables under its feet)
+insert into t1 values ("00000000");
+flush table t1;
+* copied t1 for comparison
+lock tables t1 write;
+insert into t1 values ("aaaaaaaaa");
+SET SESSION debug="+d,maria_crash";
+* crashing mysqld intentionally
+set global maria_checkpoint_interval=1;
+ERROR HY000: Lost connection to MySQL server during query
+* recovery happens
+* rebuilding index (until we have recovery of index)
+repair table t1 quick;
+Table	Op	Msg_type	Msg_text
+mysqltest.t1	repair	status	OK
+* testing that checksum after recovery is as expected
+select if(substring("mysqltest.t1	976141720",instr("mysqltest.t1	976141720",".t1")) = substring("mysqltest_for_comparison.t1	976141720",instr("mysqltest_for_comparison.t1	976141720",".t1")),"ok","failure");
+if(substring("mysqltest.t1	976141720",instr("mysqltest.t1	976141720",".t1")) = substring("mysqltest_for_comparison.t1	976141720",instr("mysqltest_for_comparison.t1	976141720",".t1")),"ok","failure")
+ok
+use mysqltest;
+select * from t1;
+a
+00000000
+00000000
+insert into t1 values ("00000000");
+flush table t1;
+* copied t1 for comparison
+lock tables t1 write;
+insert into t1 values ("aaaaaaaaa");
+SET SESSION debug="+d,maria_flush_whole_page_cache,maria_crash";
+* crashing mysqld intentionally
+set global maria_checkpoint_interval=1;
+ERROR HY000: Lost connection to MySQL server during query
+* recovery happens
+* rebuilding index (until we have recovery of index)
+repair table t1 quick;
+Table	Op	Msg_type	Msg_text
+mysqltest.t1	repair	status	OK
+* testing that checksum after recovery is as expected
+select if(substring("mysqltest.t1	1464212580",instr("mysqltest.t1	1464212580",".t1")) = substring("mysqltest_for_comparison.t1	1464212580",instr("mysqltest_for_comparison.t1	1464212580",".t1")),"ok","failure");
+if(substring("mysqltest.t1	1464212580",instr("mysqltest.t1	1464212580",".t1")) = substring("mysqltest_for_comparison.t1	1464212580",instr("mysqltest_for_comparison.t1	1464212580",".t1")),"ok","failure")
+ok
+use mysqltest;
+select * from t1;
+a
+00000000
+00000000
+00000000
+insert into t1 values ("00000000");
+flush table t1;
+* copied t1 for comparison
+lock tables t1 write;
+insert into t1 values ("aaaaaaaaa");
+SET SESSION debug="+d,maria_flush_states,maria_flush_whole_log,maria_crash";
+* crashing mysqld intentionally
+set global maria_checkpoint_interval=1;
+ERROR HY000: Lost connection to MySQL server during query
+* recovery happens
+* rebuilding index (until we have recovery of index)
+repair table t1 quick;
+Table	Op	Msg_type	Msg_text
+mysqltest.t1	repair	status	OK
+* testing that checksum after recovery is as expected
+select if(substring("mysqltest.t1	1952283440",instr("mysqltest.t1	1952283440",".t1")) = substring("mysqltest_for_comparison.t1	1952283440",instr("mysqltest_for_comparison.t1	1952283440",".t1")),"ok","failure");
+if(substring("mysqltest.t1	1952283440",instr("mysqltest.t1	1952283440",".t1")) = substring("mysqltest_for_comparison.t1	1952283440",instr("mysqltest_for_comparison.t1	1952283440",".t1")),"ok","failure")
+ok
+use mysqltest;
+select * from t1;
+a
+00000000
+00000000
+00000000
+00000000
+insert into t1 values ("00000000");
+flush table t1;
+* copied t1 for comparison
+lock tables t1 write;
+insert into t1 values ("aaaaaaaaa");
+SET SESSION debug="+d,maria_flush_whole_log,maria_crash";
+* crashing mysqld intentionally
+set global maria_checkpoint_interval=1;
+ERROR HY000: Lost connection to MySQL server during query
+* recovery happens
+* rebuilding index (until we have recovery of index)
+repair table t1 quick;
+Table	Op	Msg_type	Msg_text
+mysqltest.t1	repair	status	OK
+* testing that checksum after recovery is as expected
+select if(substring("mysqltest.t1	2440354300",instr("mysqltest.t1	2440354300",".t1")) = substring("mysqltest_for_comparison.t1	2440354300",instr("mysqltest_for_comparison.t1	2440354300",".t1")),"ok","failure");
+if(substring("mysqltest.t1	2440354300",instr("mysqltest.t1	2440354300",".t1")) = substring("mysqltest_for_comparison.t1	2440354300",instr("mysqltest_for_comparison.t1	2440354300",".t1")),"ok","failure")
+ok
+use mysqltest;
+select * from t1;
+a
+00000000
+00000000
+00000000
+00000000
+00000000
+drop table t1;
+* shut down mysqld, removed logs, restarted it
+use mysqltest;
+CREATE TABLE t1 (
+i int,
+b blob default NULL,
+c varchar(6000) default NULL
+) ENGINE=MARIA CHECKSUM=1;
+* copied t1 for feeding_recovery
+INSERT INTO t1 VALUES (1, REPEAT('a', 5000), REPEAT('b', 5000));
+UPDATE t1 SET i=3, b=CONCAT(b,'c') WHERE i=1;
+SELECT LENGTH(b) FROM t1 WHERE i=3;
+LENGTH(b)
+5001
+flush table t1;
+* copied t1 for comparison
+SET SESSION debug="+d,maria_flush_whole_log,maria_crash";
+* crashing mysqld intentionally
+set global maria_checkpoint_interval=1;
+ERROR HY000: Lost connection to MySQL server during query
+* copied t1 back for feeding_recovery
+* recovery happens
+* rebuilding index (until we have recovery of index)
+repair table t1 quick;
+Table	Op	Msg_type	Msg_text
+mysqltest.t1	repair	status	OK
+* testing that checksum after recovery is as expected
+select if(substring("mysqltest.t1	3472399915",instr("mysqltest.t1	3472399915",".t1")) = substring("mysqltest_for_comparison.t1	3472399915",instr("mysqltest_for_comparison.t1	3472399915",".t1")),"ok","failure");
+if(substring("mysqltest.t1	3472399915",instr("mysqltest.t1	3472399915",".t1")) = substring("mysqltest_for_comparison.t1	3472399915",instr("mysqltest_for_comparison.t1	3472399915",".t1")),"ok","failure")
+ok
+use mysqltest;
+SELECT LENGTH(b) FROM t1 WHERE i=3;
+LENGTH(b)
+5001
+drop table t1;
+drop database mysqltest_for_feeding_recovery;
+drop database mysqltest_for_comparison;
+drop database mysqltest;

mysql-test/t/maria-recovery-master.opt

+--skip-stack-trace --skip-core-file
+

mysql-test/t/maria-recovery.test

+--source include/not_embedded.inc
+# Don't test this under valgrind, memory leaks will occur as we crash
+--source include/not_valgrind.inc
+# Binary must be compiled with debug for crash to occur
+--source include/have_debug.inc
+--source include/have_maria.inc
+
+--disable_warnings
+drop database if exists mysqltest;
+--enable_warnings
+create database mysqltest;
+
+# Include scripts can perform SQL. For it to not influence the main test
+# they use a separate connection. This way if they use a DDL it would
+# not autocommit in the main test.
+connect (admin, 127.0.0.1, root,,mysqltest,,);
+--enable_reconnect
+
+connection default;
+use mysqltest;
+--enable_reconnect
+
+# A sample test
+-- source include/maria_empty_logs.inc
+let $mms_tables=1;
+create table t1 (a varchar(1000)) engine=maria;
+
+--echo * TEST of REDO: see if recovery can reconstruct if we give it an old table
+
+-- source include/maria_make_snapshot_for_feeding_recovery.inc
+# Your committed statements here, which we expect to
+# be reconstructed from the log
+insert into t1 values ("00000000");
+-- source include/maria_make_snapshot_for_comparison.inc
+# we want recovery to run on the first snapshot made above
+let $mvr_restore_old_snapshot=1;
+# As we did only committed work, we test REDO applying, which could
+# produce a physically identical table.
+let $mms_compare_physically=1;
+let $mvr_debug_option="+d,maria_flush_whole_log,maria_crash";
+# the script below will trigger recovery and compare checksums
+-- source include/maria_verify_recovery.inc
+let $mms_compare_physically=0;
+# so a SELECT like this is pure visual effect, brings nothing.
+select * from t1;
+
+--echo * TEST of REDO+UNDO: normal recovery test (no moving tables under its feet)
+
+# different types of crash => a loop; here are loop control variables
+let $crash_no_flush=1;
+let $crash_flush_whole_page_cache=0;
+let $crash_flush_states=0;
+let $crash_flush_whole_log=0;
+let $crash_loop=1;
+
+# we want recovery to use the tables as they were at time of crash
+let $mvr_restore_old_snapshot=0;
+# UNDO phase prevents physical comparison, normally,
+# so we'll only use checksums to compare.
+let $mms_compare_physically=0;
+
+# Note that we don't remove logs between iterations. Test is
+# cumulative (each new recovery processes more log records than the previous).
+
+while ($crash_loop)
+{
+  if ($crash_flush_whole_log)
+  {
+     let $mvr_debug_option="+d,maria_flush_whole_log,maria_crash";
+     # set up what next iteration should do:
+     let $crash_flush_whole_log=0;
+     let $crash_loop=0;
+  }
+  if ($crash_flush_states)
+  {
+     let $mvr_debug_option="+d,maria_flush_states,maria_flush_whole_log,maria_crash";
+     let $crash_flush_states=0;
+     let $crash_flush_whole_log=1;
+  }
+  if ($crash_flush_whole_page_cache)
+  {
+     let $mvr_debug_option="+d,maria_flush_whole_page_cache,maria_crash";
+     let $crash_flush_whole_page_cache=0;
+     let $crash_flush_states=1;
+  }
+  if ($crash_no_flush)
+  {
+     let $mvr_debug_option="+d,maria_crash";
+     let $crash_no_flush=0;
+     let $crash_flush_whole_page_cache=1;
+  }
+  # Your committed statements here
+  insert into t1 values ("00000000");
+  -- source include/maria_make_snapshot_for_comparison.inc
+  # Your statements which we expect to be rolled back
+  lock tables t1 write;
+  insert into t1 values ("aaaaaaaaa");
+  -- source include/maria_verify_recovery.inc
+  select * from t1;
+}
+
+drop table t1;
+
+# what did we compare above:
+# - checksum: tells that the tables contain the same amount of rows
+# and same data in rows
+# - index: no, neither state nor pages were compared
+# - bitmap pages: the REPAIR QUICK done above very probably checks
+# that bitmap reflects page occupation; do we need to do physical
+# compare?
+# - page LSN: not compared; we should compare that page's LSN in new
+# table is >= page's LSN in old table (it can be >, due to UNDO phase)
+# we had a bug where new page's LSN was 0... todo.
+
+#
+# Test for this bug: an UPDATE purges and rewrites a tail page, and
+# recovery applied the purge, stamped page with UNDO's LSN, thus
+# the rewrite was ignored.
+#
+
+-- source include/maria_empty_logs.inc
+let $mms_tables=1;
+CREATE TABLE t1 (
+  i int,
+  b blob default NULL,
+  c varchar(6000) default NULL
+) ENGINE=MARIA CHECKSUM=1;
+-- source include/maria_make_snapshot_for_feeding_recovery.inc
+INSERT INTO t1 VALUES (1, REPEAT('a', 5000), REPEAT('b', 5000));
+UPDATE t1 SET i=3, b=CONCAT(b,'c') WHERE i=1;
+SELECT LENGTH(b) FROM t1 WHERE i=3;
+-- source include/maria_make_snapshot_for_comparison.inc
+# we want recovery to run on the first snapshot made above
+let $mvr_restore_old_snapshot=1;
+let $mms_compare_physically=0;
+let $mvr_debug_option="+d,maria_flush_whole_log,maria_crash";
+-- source include/maria_verify_recovery.inc
+SELECT LENGTH(b) FROM t1 WHERE i=3;
+drop table t1;
+
+# clean up everything
+let $mms_purpose=feeding_recovery;
+eval drop database mysqltest_for_$mms_purpose;
+let $mms_purpose=comparison;
+eval drop database mysqltest_for_$mms_purpose;
+drop database mysqltest;

storage/maria/ha_maria.cc

@@ -1243,6 +1243,11 @@ int ha_maria::repair(THD *thd, HA_CHECK &param, bool do_optimize)
       thd->proc_info= "Repair with keycache";
       param.testflag &= ~T_REP_BY_SORT;
       error= maria_repair(&param, file, fixed_name, param.testflag & T_QUICK);
+      /**
+         @todo RECOVERY BUG we do things with the index file
+         (maria_sort_index() after the above which already has logged the
+         record and bumped create_rename_lsn. Is it ok?
+      */
     }
     param.testflag= testflag;
     optimize_done= 1;
@@ -1311,6 +1316,11 @@ int ha_maria::repair(THD *thd, HA_CHECK &param, bool do_optimize)
   thd->proc_info= old_proc_info;
   if (!thd->locked_tables)
   {
+    /**
+       @todo RECOVERY BUG find why this is needed. Monty says it's because a
+       new non-transactional table is created by maria_repair(): find how this
+       new table's state influences the old one's.
+    */
     _ma_reenable_logging_for_table(file->s);
     maria_lock_database(file, F_UNLCK);
   }
@@ -1991,20 +2001,6 @@ int ha_maria::external_lock(THD *thd, int lock_type)
     goto skip_transaction;
   if (lock_type != F_UNLCK)
   {
-    if (!thd->transaction.on)
-    {
-      /*
-        No need to log REDOs/UNDOs. If this is an internal temporary table
-        which will be renamed to a permanent table (like in ALTER TABLE),
-        the rename happens after unlocking so will be durable (and the table
-        will get its create_rename_lsn).
-        Note: if we wanted to enable users to have an old backup and apply
-        tons of archived logs to roll-forward, we could then not disable
-        REDOs/UNDOs in this case.
-      */
-      DBUG_PRINT("info", ("Disabling logging for table"));
-      _ma_tmp_disable_logging_for_table(file->s);
-    }
     if (!trn)  /* no transaction yet - open it now */
     {
       trn= trnman_new_trn(& thd->mysys_var->mutex,
@@ -2025,6 +2021,20 @@ int ha_maria::external_lock(THD *thd, int lock_type)
       trans_register_ha(thd, FALSE, maria_hton);
       trnman_new_statement(trn);
     }
+    if (!thd->transaction.on)
+    {
+      /*
+        No need to log REDOs/UNDOs. If this is an internal temporary table
+        which will be renamed to a permanent table (like in ALTER TABLE),
+        the rename happens after unlocking so will be durable (and the table
+        will get its create_rename_lsn).
+        Note: if we wanted to enable users to have an old backup and apply
+        tons of archived logs to roll-forward, we could then not disable
+        REDOs/UNDOs in this case.
+      */
+      DBUG_PRINT("info", ("Disabling logging for table"));
+      _ma_tmp_disable_logging_for_table(file, TRUE);
+    }
   }
   else
   {

storage/maria/ma_bitmap.c

@@ -152,7 +152,8 @@ static inline my_bool write_changed_bitmap(MARIA_SHARE *share,
                           (uchar*) bitmap->map, PAGECACHE_PLAIN_PAGE,
                           PAGECACHE_LOCK_LEFT_UNLOCKED,
                           PAGECACHE_PIN_LEFT_UNPINNED,
-                          PAGECACHE_WRITE_DELAY, 0));
+                          PAGECACHE_WRITE_DELAY, 0,
+                          LSN_IMPOSSIBLE));
 }
 
 /*
@@ -571,6 +572,26 @@ static my_bool _ma_read_bitmap_page(MARIA_SHARE *share,
       Inexistent or half-created page (could be crash in the middle of
       _ma_bitmap_create_first(), before appending maria_bitmap_marker).
     */
+    /**
+       @todo RECOVERY BUG
+       We are updating data_file_length before writing any log record for the
+       row operation. What if now state is flushed by a checkpoint with the
+       new value, and crash before the checkpoint record is written, recovery
+       may not even open the table (no log records) so not fix
+       data_file_length ("WAL violation")?
+       Scenario: assume share->id==0, then:
+       thread 1 (here)                thread 2 (checkpoint)
+       update data_file_length
+                                      copy state to memory, flush log
+       set share->id and write FILE_ID (not flushed)
+                                      see share->id!=0 so flush state
+                                      crash
+       FILE_ID will be missing, Recovery will not open table and not fix
+       data_file_length. This bug should be fixed with other "checkpoint vs
+       bitmap" bugs.
+       One possibility will be logging a standalone LOGREC_CREATE_BITMAP in a
+       separate transaction (using dummy_transaction_object).
+    */
     share->state.state.data_file_length= end_of_page;
     bzero(bitmap->map, bitmap->block_size);
     memcpy(bitmap->map + bitmap->block_size - sizeof(maria_bitmap_marker),
@@ -641,6 +662,12 @@ static my_bool _ma_change_bitmap_page(MARIA_HA *info,
 
   if (bitmap->changed)
   {
+    /**
+       @todo RECOVERY BUG this is going to flush the bitmap page possibly to
+       disk even though it could be over-allocated with not yet any REDO-UNDO
+       complete group (WAL violation: no way to undo the over-allocation if
+       crash). See also collect_tables().
+    */
     if (write_changed_bitmap(info->s, bitmap))
       DBUG_RETURN(1);
     bitmap->changed= 0;

storage/maria/ma_blockrec.c

@@ -1433,7 +1433,15 @@ static my_bool write_tail(MARIA_HA *info,
   /* Increase data file size, if extended */
   position= (my_off_t) block->page * block_size;
   if (info->state->data_file_length <= position)
+  {
+    /*
+      We are modifying a state member before writing the UNDO; this is a WAL
+      violation. But for data_file_length this is ok, as long as we change
+      data_file_length after writing any log record (FILE_ID/REDO/UNDO) (see
+      collect_tables()).
+    */
     info->state->data_file_length= position + block_size;
+  }
 
   DBUG_ASSERT(share->pagecache->block_size == block_size);
   if (!(res= pagecache_write(share->pagecache,
@@ -1443,7 +1451,8 @@ static my_bool write_tail(MARIA_HA *info,
                              PAGECACHE_LOCK_READ,
                              block_is_read ? PAGECACHE_PIN_LEFT_PINNED :
                              PAGECACHE_PIN,
-                             PAGECACHE_WRITE_DELAY, &page_link.link)))
+                             PAGECACHE_WRITE_DELAY, &page_link.link,
+                             LSN_IMPOSSIBLE)))
   {
     page_link.unlock= PAGECACHE_LOCK_READ_UNLOCK;
     page_link.changed= 1;
@@ -1547,7 +1556,7 @@ static my_bool write_full_pages(MARIA_HA *info,
                         PAGECACHE_LOCK_LEFT_UNLOCKED,
                         PAGECACHE_PIN_LEFT_UNPINNED,
                         PAGECACHE_WRITE_DELAY,
-                        0))
+                        0, LSN_IMPOSSIBLE))
       DBUG_RETURN(1);
     page++;
     block->used= BLOCKUSED_USED;
@@ -2351,7 +2360,8 @@ static my_bool write_block_record(MARIA_HA *info,
                       PAGECACHE_LOCK_READ,
                       head_block_is_read ? PAGECACHE_PIN_LEFT_PINNED :
                       PAGECACHE_PIN,
-                      PAGECACHE_WRITE_DELAY, &page_link.link))
+                      PAGECACHE_WRITE_DELAY, &page_link.link,
+                      LSN_IMPOSSIBLE))
     goto disk_err;
   page_link.unlock= PAGECACHE_LOCK_READ_UNLOCK;
   page_link.changed= 1;
@@ -3172,7 +3182,8 @@ static my_bool delete_head_or_tail(MARIA_HA *info,
                         buff, share->page_type,
                         lock_at_write,
                         PAGECACHE_PIN_LEFT_PINNED,
-                        PAGECACHE_WRITE_DELAY, &page_link.link))
+                        PAGECACHE_WRITE_DELAY, &page_link.link,
+                        LSN_IMPOSSIBLE))
       DBUG_RETURN(1);
   }
   else /* page is now empty */
@@ -3196,7 +3207,8 @@ static my_bool delete_head_or_tail(MARIA_HA *info,
                         buff, share->page_type,
                         lock_at_write,
                         PAGECACHE_PIN_LEFT_PINNED,
-                        PAGECACHE_WRITE_DELAY, &page_link.link))
+                        PAGECACHE_WRITE_DELAY, &page_link.link,
+                        LSN_IMPOSSIBLE))
       DBUG_RETURN(1);
 
     DBUG_ASSERT(empty_space >= info->s->bitmap.sizes[0]);
@@ -3257,6 +3269,7 @@ my_bool _ma_delete_block_record(MARIA_HA *info, const uchar *record)
   ulonglong page;
   uint record_number;
   MARIA_SHARE *share= info->s;
+  LSN lsn= LSN_IMPOSSIBLE;
   DBUG_ENTER("_ma_delete_block_record");
 
   page=          ma_recordpos_to_page(info->cur_row.lastpos);
@@ -3273,7 +3286,6 @@ my_bool _ma_delete_block_record(MARIA_HA *info, const uchar *record)
 
   if (share->now_transactional)
   {
-    LSN lsn;
     uchar log_data[LSN_STORE_SIZE + FILEID_STORE_SIZE + PAGE_STORE_SIZE +
                    DIRPOS_STORE_SIZE + HA_CHECKSUM_STORE_SIZE];
     size_t row_length;
@@ -3311,7 +3323,7 @@ my_bool _ma_delete_block_record(MARIA_HA *info, const uchar *record)
 
   }
 
-  _ma_unpin_all_pages_and_finalize_row(info, info->trn->undo_lsn);
+  _ma_unpin_all_pages_and_finalize_row(info, lsn);
   DBUG_RETURN(0);
 
 err:
@@ -5078,8 +5090,8 @@ uint _ma_apply_redo_insert_row_head_or_tail(MARIA_HA *info, LSN lsn,
     DBUG_ASSERT(rownr == 0);
     if (rownr != 0)
       goto err;
-    unlock_method= PAGECACHE_LOCK_LEFT_UNLOCKED;
-    unpin_method=  PAGECACHE_PIN_LEFT_UNPINNED;
+    unlock_method= PAGECACHE_LOCK_WRITE;
+    unpin_method=  PAGECACHE_PIN;
 
     buff= info->keyread_buff;
     info->keyread_buff_used= 1;
@@ -5120,8 +5132,8 @@ uint _ma_apply_redo_insert_row_head_or_tail(MARIA_HA *info, LSN lsn,
         DBUG_RETURN(my_errno);
       DBUG_RETURN(0);
     }
-    unlock_method= PAGECACHE_LOCK_WRITE_UNLOCK;
-    unpin_method=  PAGECACHE_UNPIN;
+    unlock_method= PAGECACHE_LOCK_LEFT_WRITELOCKED;
+    unpin_method=  PAGECACHE_PIN_LEFT_PINNED;
 
     if (((buff[PAGE_TYPE_OFFSET] & PAGE_TYPE_MASK) != page_type))
     {
@@ -5189,15 +5201,23 @@ uint _ma_apply_redo_insert_row_head_or_tail(MARIA_HA *info, LSN lsn,
   empty_space-= data_length;
   int2store(buff + EMPTY_SPACE_OFFSET, empty_space);
 
-  /* Write modified page */
-  lsn_store(buff, lsn);
+  /*
+    Write modified page. We don't update its LSN, and keep it pinned. When we
+    have processed all REDOs for this page in the current REDO's group, we
+    will stamp page with UNDO's LSN (if we stamped it now, a next REDO, in
+    this group, for this page, would be skipped) and unpin then.
+  */
   if (pagecache_write(share->pagecache,
                       &info->dfile, page, 0,
                       buff, PAGECACHE_PLAIN_PAGE,
                       unlock_method, unpin_method,
-                      PAGECACHE_WRITE_DELAY, 0))
+                      PAGECACHE_WRITE_DELAY, &page_link.link,
+                      LSN_IMPOSSIBLE))
     DBUG_RETURN(my_errno);
 
+  page_link.unlock= PAGECACHE_LOCK_WRITE_UNLOCK;
+  push_dynamic(&info->pinned_pages, (void*) &page_link);
+
   /* Fix bitmap */
   if (_ma_bitmap_set(info, page, page_type == HEAD_PAGE, empty_space))
     DBUG_RETURN(my_errno);
@@ -5215,7 +5235,7 @@ uint _ma_apply_redo_insert_row_head_or_tail(MARIA_HA *info, LSN lsn,
   DBUG_RETURN(0);
 
 err:
-  if (unlock_method == PAGECACHE_LOCK_WRITE_UNLOCK)
+  if (unlock_method == PAGECACHE_LOCK_LEFT_WRITELOCKED)
     pagecache_unlock_by_link(share->pagecache, page_link.link,
                              PAGECACHE_LOCK_WRITE_UNLOCK,
                              PAGECACHE_UNPIN, LSN_IMPOSSIBLE,
@@ -5302,18 +5322,23 @@ uint _ma_apply_redo_purge_row_head_or_tail(MARIA_HA *info, LSN lsn,
   if (delete_dir_entry(buff, block_size, rownr, &empty_space) < 0)
     goto err;
 
-  lsn_store(buff, lsn);
   result= 0;
   if (pagecache_write(share->pagecache,
                       &info->dfile, page, 0,
                       buff, PAGECACHE_PLAIN_PAGE,
-                      PAGECACHE_LOCK_WRITE_UNLOCK, PAGECACHE_UNPIN,
-                      PAGECACHE_WRITE_DELAY, 0))
+                      PAGECACHE_LOCK_LEFT_WRITELOCKED,
+                      PAGECACHE_PIN_LEFT_PINNED,
+                      PAGECACHE_WRITE_DELAY, 0,
+                      LSN_IMPOSSIBLE))
     result= my_errno;
-
+  else
+  {
+    page_link.unlock= PAGECACHE_LOCK_WRITE_UNLOCK;
+    push_dynamic(&info->pinned_pages, (void*) &page_link);
     /* This will work even if the page was marked as UNALLOCATED_PAGE */
     if (_ma_bitmap_set(info, page, page_type == HEAD_PAGE, empty_space))
       result= my_errno;
+  }
 
   DBUG_RETURN(result);
 
@@ -5360,7 +5385,12 @@ uint _ma_apply_redo_free_blocks(MARIA_HA *info,
 
     start_page= page= page_korr(header);
     header+= PAGE_STORE_SIZE;
-    page_range= pagerange_korr(header);
+    /* Page range may have this bit set to indicate a tail page */
+    page_range= pagerange_korr(header) & ~TAIL_BIT;
+    /** @todo RECOVERY BUG enable this assertion when newer tree pulled */
+#if 0
+    DBUG_ASSERT(page_range > 0);
+#endif
     header+= PAGERANGE_STORE_SIZE;
 
     DBUG_PRINT("info", ("page: %lu  pages: %u", (long) page, page_range));

storage/maria/ma_checkpoint.c

@@ -176,7 +176,6 @@ static int really_execute_checkpoint(void)
                      LSN_IN_PARTS(checkpoint_start_log_horizon)));
   lsn_store(checkpoint_start_log_horizon_char, checkpoint_start_log_horizon);
 
-
   /*
     STEP 2: fetch information about transactions.
     We must fetch transactions before dirty pages. Indeed, a transaction
@@ -346,6 +345,43 @@ int ma_checkpoint_init(ulong interval)
 }
 
 
+#ifndef DBUG_OFF
+/**
+   Function used to test recovery: flush some table pieces and then caller
+   crashes.
+
+   @param  what_to_flush   0: current bitmap and all data pages
+                           1: state
+*/
+static void flush_all_tables(int what_to_flush)
+{
+  LIST *pos; /**< to iterate over open tables */
+  pthread_mutex_lock(&THR_LOCK_maria);
+  for (pos= maria_open_list; pos; pos= pos->next)
+  {
+    MARIA_HA *info= (MARIA_HA*)pos->data;
+    if (info->s->now_transactional)
+    {
+      switch (what_to_flush)
+      {
+      case 0:
+        _ma_flush_table_files(info, MARIA_FLUSH_DATA | MARIA_FLUSH_INDEX,
+                              FLUSH_KEEP, FLUSH_KEEP);
+        break;
+      case 1:
+        _ma_state_info_write(info->s, 1|4);
+        DBUG_PRINT("maria_flush_states",
+                   ("is_of_horizon: LSN (%lu,0x%lx)",
+                    LSN_IN_PARTS(info->s->state.is_of_horizon)));
+        break;
+      }
+    }
+  }
+  pthread_mutex_unlock(&THR_LOCK_maria);
+}
+#endif
+
+
 /**
    @brief Destroys the checkpoint module
 */
@@ -353,6 +389,32 @@ int ma_checkpoint_init(ulong interval)
 void ma_checkpoint_end(void)
 {
   DBUG_ENTER("ma_checkpoint_end");
+  DBUG_EXECUTE_IF("maria_flush_whole_page_cache",
+                  {
+                    DBUG_PRINT("maria_flush_whole_page_cache", ("now"));
+                    flush_all_tables(0);
+                  });
+  DBUG_EXECUTE_IF("maria_flush_whole_log",
+                  {
+                    DBUG_PRINT("maria_flush_whole_log", ("now"));
+                    translog_flush(translog_get_horizon());
+                  });
+  /*
+    Note that for WAL reasons, maria_flush_states requires
+    maria_flush_whole_log.
+  */
+  DBUG_EXECUTE_IF("maria_flush_states",
+                  {
+                    DBUG_PRINT("maria_flush_states", ("now"));
+                    flush_all_tables(1);
+                  });
+  DBUG_EXECUTE_IF("maria_crash",
+                  {
+                    DBUG_PRINT("maria_crash", ("now"));
+                    fflush(DBUG_FILE);
+                    abort();
+                  });
+
   if (checkpoint_inited)
   {
     pthread_mutex_lock(&LOCK_checkpoint);
@@ -501,11 +563,6 @@ filter_flush_data_file_evenly(enum pagecache_page_type type,
 
    @note MikaelR questioned why the same thread does two different jobs, the
    risk could be that while a checkpoint happens no LRD flushing happens.
-
-   @note MikaelR noted that he observed that Linux's file cache may never
-   fsync to  disk until this cache is full, at which point it decides to empty
-   the cache, making the machine very slow. A solution was to fsync after
-   writing 2 MB.
 */
 
 pthread_handler_t ma_checkpoint_background(void *arg)
@@ -622,6 +679,13 @@ pthread_handler_t ma_checkpoint_background(void *arg)
           if (filter_param.max_pages == 0) /* bunch all flushed, sleep */
             break; /* and we will continue with the same file */
           dfile++; /* otherwise all this file is flushed, move to next file */
+          /*
+            MikaelR noted that he observed that Linux's file cache may never
+            fsync to  disk until this cache is full, at which point it decides
+            to empty the cache, making the machine very slow. A solution was
+            to fsync after writing 2 MB. So we might want to fsync() here if
+            we wrote enough pages.
+          */
         }
         filter_param.is_data_file= FALSE;
         while (kfile != kfiles_end)
@@ -866,9 +930,32 @@ static int collect_tables(LEX_STRING *str, LSN checkpoint_start_log_horizon)
         */
       }
       translog_unlock();
+      /**
+         We are going to flush these states.
+         Before, all records describing how to undo such state must be
+         in the log (WAL). Usually this means UNDOs. In the special case of
+         data|key_file_length, recovery just needs to open the table to fix the
+         length, so any LOGREC_FILE_ID/REDO/UNDO allowing recovery to
+         understand it must open a table, is enough; so as long as
+         data|key_file_length is updated after writing any log record it's ok:
+         if we copied new value above, it means the record was before
+         state_copies_horizon and we flush such record below.
+         Apart from data|key_file_length which are easily recoverable from the
+         real file's size, all other state members must be updated only when
+         writing the UNDO; otherwise, if updated before, if their new value is
+         flushed by a checkpoint and there is a crash before UNDO is written,
+         their REDO group will be missing or at least incomplete and skipped
+         by recovery, so bad state value will stay. For example, setting
+         key_root before writing the UNDO: the table would have old index
+         pages (they were pinned at time of crash) and a new, thus wrong,
+         key_root.
+         @todo RECOVERY BUG check that all code honours that.
+      */
+      if (translog_flush(state_copies_horizon))
+        goto err;
+      /* now we have cached states and they are WAL-safe*/
       state_copies_end= state_copy;
       state_copy= state_copies;
-      /* so now we have cached states */
     }
 
     /* locate our state among these cached ones */
@@ -979,6 +1066,12 @@ static int collect_tables(LEX_STRING *str, LSN checkpoint_start_log_horizon)
           each checkpoint if the table was once written and then not anymore.
         */
       }
+      /**
+         @todo RECOVERY BUG this is going to flush the bitmap page possibly to
+         disk even though it could be over-allocated with not yet any
+         REDO-UNDO complete group (WAL violation: no way to undo the
+         over-allocation if crash); see also _ma_change_bitmap_page().
+      */
       sync_error|=
         _ma_flush_bitmap(share); /* after that, all is in page cache */
       DBUG_ASSERT(share->pagecache == maria_pagecache);

storage/maria/ma_close.c

@@ -86,8 +86,9 @@ int maria_close(register MARIA_HA *info)
         may be using the file at this point
         IF using --external-locking, which does not apply to Maria.
       */
-      if ((share->changed && share->base.born_transactional) ||
-          (share->mode != O_RDONLY && maria_is_crashed(info)))
+      if (share->mode != O_RDONLY &&
+          ((share->changed && share->base.born_transactional) ||
+           maria_is_crashed(info)))
       {
         /*
           State must be written to file as it was not done at table's

storage/maria/ma_loghandler.c

@@ -552,6 +552,11 @@ static LOG_DESC INIT_LOGREC_LONG_TRANSACTION_ID=
 {LOGRECTYPE_FIXEDLENGTH, 6, 6, NULL, NULL, NULL, 0,
  "long_transaction_id", LOGREC_IS_GROUP_ITSELF, NULL, NULL};
 
+static LOG_DESC INIT_LOGREC_INCOMPLETE_LOG=
+{LOGRECTYPE_FIXEDLENGTH, FILEID_STORE_SIZE, FILEID_STORE_SIZE,
+ NULL, NULL, NULL, 0,
+ "incomplete_log", LOGREC_IS_GROUP_ITSELF, NULL, NULL};
+
 const myf log_write_flags= MY_WME | MY_NABP | MY_WAIT_IF_FULL;
 
 static void loghandler_init()
@@ -627,12 +632,14 @@ static void loghandler_init()
     INIT_LOGREC_FILE_ID;
   log_record_type_descriptor[LOGREC_LONG_TRANSACTION_ID]=
     INIT_LOGREC_LONG_TRANSACTION_ID;
-  for (i= LOGREC_LONG_TRANSACTION_ID + 1;
+  log_record_type_descriptor[LOGREC_INCOMPLETE_LOG]=
+    INIT_LOGREC_INCOMPLETE_LOG;
+  for (i= LOGREC_INCOMPLETE_LOG + 1;
        i < LOGREC_NUMBER_OF_TYPES;
        i++)
     log_record_type_descriptor[i].class= LOGRECTYPE_NOT_ALLOWED;
   DBUG_EXECUTE("info",
-               check_translog_description_table(LOGREC_LONG_TRANSACTION_ID););
+               check_translog_description_table(LOGREC_INCOMPLETE_LOG););
 };
 
 
@@ -2083,6 +2090,7 @@ static my_bool translog_buffer_flush(struct st_translog_buffer *buffer)
                         PAGECACHE_PLAIN_PAGE,
                         PAGECACHE_LOCK_LEFT_UNLOCKED,
                         PAGECACHE_PIN_LEFT_UNPINNED, 0,
+                        LSN_IMPOSSIBLE,
                         &translog_page_validator, (uchar*) &data))
     {
       UNRECOVERABLE_ERROR(("Can't write page (%lu,0x%lx) to pagecache",
@@ -6570,7 +6578,8 @@ my_bool translog_flush(LSN lsn)
     struct st_translog_buffer *buffer= log_descriptor.bc.buffer;
     /* we can't flush in future */
     DBUG_ASSERT(cmp_translog_addr(log_descriptor.horizon, lsn) >= 0);
-    if (cmp_translog_addr(log_descriptor.flushed, lsn) >= 0)
+    if (cmp_translog_addr(log_descriptor.flushed, lsn) >= 0 ||
+        full_circle)
     {
       DBUG_PRINT("info", ("already flushed: (%lu,0x%lx)",
                           LSN_IN_PARTS(log_descriptor.flushed)));

storage/maria/ma_loghandler.h

@@ -137,6 +137,7 @@ enum translog_record_type
   LOGREC_REDO_REPAIR_TABLE,
   LOGREC_FILE_ID,
   LOGREC_LONG_TRANSACTION_ID,
+  LOGREC_INCOMPLETE_LOG,
   LOGREC_RESERVED_FUTURE_EXTENSION= 63
 };
 #define LOGREC_NUMBER_OF_TYPES 64              /* Maximum, can't be extended */

storage/maria/ma_open.c

@@ -1058,6 +1058,8 @@ uint _ma_state_info_write(MARIA_SHARE *share, uint pWrite)
       is too new). Recovery does it by itself.
     */
     share->state.is_of_horizon= translog_get_horizon();
+    DBUG_PRINT("info", ("is_of_horizon set to LSN (%lu,0x%lx)",
+                        LSN_IN_PARTS(share->state.is_of_horizon)));
   }
   res= _ma_state_info_write_sub(share->kfile.file, &share->state, pWrite);
   if (pWrite & 4)

storage/maria/ma_page.c

@@ -137,7 +137,8 @@ int _ma_write_keypage(register MARIA_HA *info, register MARIA_KEYDEF *keyinfo,
                        lock == PAGECACHE_LOCK_LEFT_WRITELOCKED ?
                        PAGECACHE_PIN_LEFT_PINNED :
                        PAGECACHE_PIN,
-                       PAGECACHE_WRITE_DELAY, &page_link.link);
+                       PAGECACHE_WRITE_DELAY, &page_link.link,
+		       LSN_IMPOSSIBLE);
 
   if (lock == PAGECACHE_LOCK_WRITE)
   {
@@ -192,7 +193,6 @@ int _ma_dispose(register MARIA_HA *info, my_off_t pos, my_bool page_not_read)
   _ma_store_keynr(info, buff, (uchar) MARIA_DELETE_KEY_NR);
   mi_sizestore(buff + share->keypage_header, old_link);
   share->state.changed|= STATE_NOT_SORTED_PAGES;
-
   if (info->s->now_transactional)
   {
     LSN lsn;
@@ -235,6 +235,7 @@ int _ma_dispose(register MARIA_HA *info, my_off_t pos, my_bool page_not_read)
                            share->page_type,
                            lock_method, pin_method,
                            PAGECACHE_WRITE_DELAY, &page_link.link,
+			   LSN_IMPOSSIBLE,
                            0, share->keypage_header+8, 0, 0))
     result= 1;
 

storage/maria/ma_pagecache.c

@@ -633,6 +633,24 @@ static uint pagecache_fwrite(PAGECACHE *pagecache,
            (pageno)<<(pagecache->shift), flags)
 
 
+/**
+  @brief set rec_lsn of pagecache block (if it is needed)
+
+  @param block                   block where to set rec_lsn
+  @param first_REDO_LSN_for_page the LSN to set
+*/
+
+static inline void pagecache_set_block_rec_lsn(PAGECACHE_BLOCK_LINK *block,
+                                               LSN first_REDO_LSN_for_page)
+{
+  if (block->rec_lsn == LSN_MAX)
+    block->rec_lsn= first_REDO_LSN_for_page;
+  else
+    DBUG_ASSERT(cmp_translog_addr(block->rec_lsn,
+                                  first_REDO_LSN_for_page) <= 0);
+}
+
+
 /*
   next_power(value) is 2 at the power of (1+floor(log2(value)));
   e.g. next_power(2)=4, next_power(3)=4.
@@ -2488,7 +2506,12 @@ static void check_and_set_lsn(PAGECACHE *pagecache,
 {
   LSN old;
   DBUG_ENTER("check_and_set_lsn");
-  DBUG_ASSERT(block->type == PAGECACHE_LSN_PAGE);
+  /*
+    In recovery, we can _ma_unpin_all_pages() to put a LSN on page, though
+    page would be PAGECACHE_PLAIN_PAGE (transactionality temporarily disabled
+    to not log REDOs).
+  */
+  DBUG_ASSERT((block->type == PAGECACHE_LSN_PAGE) || maria_in_recovery);
   old= lsn_korr(block->buffer + PAGE_LSN_OFFSET);
   DBUG_PRINT("info", ("old lsn: (%lu, 0x%lx)  new lsn: (%lu, 0x%lx)",
                       LSN_IN_PARTS(old), LSN_IN_PARTS(lsn)));
@@ -2570,12 +2593,7 @@ void pagecache_unlock(PAGECACHE *pagecache,
   {
     DBUG_ASSERT(lock == PAGECACHE_LOCK_WRITE_UNLOCK);
     DBUG_ASSERT(pin == PAGECACHE_UNPIN);
-    if (block->rec_lsn == LSN_MAX)
-      block->rec_lsn= first_REDO_LSN_for_page;
-    else
-      DBUG_ASSERT(cmp_translog_addr(block->rec_lsn,
-                                    first_REDO_LSN_for_page) <= 0);
-
+    pagecache_set_block_rec_lsn(block, first_REDO_LSN_for_page);
   }
   if (lsn != LSN_IMPOSSIBLE)
     check_and_set_lsn(pagecache, lsn, block);
@@ -2755,11 +2773,7 @@ void pagecache_unlock_by_link(PAGECACHE *pagecache,
       DBUG_ASSERT(lock == PAGECACHE_LOCK_WRITE_UNLOCK ||
                   lock == PAGECACHE_LOCK_READ_UNLOCK);
       DBUG_ASSERT(pin == PAGECACHE_UNPIN);
-      if (block->rec_lsn == LSN_MAX)
-        block->rec_lsn= first_REDO_LSN_for_page;
-      else
-        DBUG_ASSERT(cmp_translog_addr(block->rec_lsn,
-                                      first_REDO_LSN_for_page) <= 0);
+      pagecache_set_block_rec_lsn(block, first_REDO_LSN_for_page);
     }
     if (lsn != LSN_IMPOSSIBLE)
       check_and_set_lsn(pagecache, lsn, block);
@@ -2785,6 +2799,7 @@ void pagecache_unlock_by_link(PAGECACHE *pagecache,
                         (ulong) block));
 
   }
+    pagecache_set_block_rec_lsn(block, first_REDO_LSN_for_page);
 
   if (make_lock_and_pin(pagecache, block, lock, pin, 0))
     DBUG_ASSERT(0);                           /* should not happend */
@@ -3214,25 +3229,27 @@ my_bool pagecache_delete_pages(PAGECACHE *pagecache,
 }
 
 
-/*
-  Write a buffer into a cached file.
-
-  SYNOPSIS
-
-    pagecache_write_part()
-      pagecache           pointer to a page cache data structure
-      file                handler for the file to write data to
-      pageno              number of the block of data in the file
-      level               determines the weight of the data
-      buff                buffer with the data
-      type                type of the page
-      lock                lock change
-      pin                 pin page
-      write_mode          how to write page
-      link                link to the page if we pin it
+/**
+  @brief Writes a buffer into a cached file.
 
-  RETURN VALUE
-    0 if a success, 1 - otherwise.
+  @param pagecache       pointer to a page cache data structure
+  @param file            handler for the file to write data to
+  @param pageno          number of the block of data in the file
+  @param level           determines the weight of the data
+  @param buff            buffer with the data
+  @param type            type of the page
+  @param lock            lock change
+  @param pin             pin page
+  @param write_mode      how to write page
+  @param link            link to the page if we pin it
+  @param first_REDO_LSN_for_page the lsn to set rec_lsn
+  @param offset          offset in the page
+  @param size            size of data
+  @param validator       read page validator
+  @param validator_data  the validator data
+
+  @retval 0 if a success.
+  @retval 1 Error.
 */
 
 /* description of how to change lock before and after write */
@@ -3296,6 +3313,7 @@ my_bool pagecache_write_part(PAGECACHE *pagecache,
                              enum pagecache_page_pin pin,
                              enum pagecache_write_mode write_mode,
                              PAGECACHE_BLOCK_LINK **page_link,
+                             LSN first_REDO_LSN_for_page,
                              uint offset, uint size,
                              pagecache_disk_read_validator validator,
                              uchar* validator_data)
@@ -3430,6 +3448,16 @@ my_bool pagecache_write_part(PAGECACHE *pagecache,
         block->status&= ~PCBLOCK_ERROR;
     }
 
+    if (first_REDO_LSN_for_page)
+    {
+      /* single write action of the last write action */
+      DBUG_ASSERT(lock == PAGECACHE_LOCK_WRITE_UNLOCK ||
+                  lock == PAGECACHE_LOCK_LEFT_UNLOCKED);
+      DBUG_ASSERT(pin == PAGECACHE_UNPIN ||
+                  pin == PAGECACHE_PIN_LEFT_UNPINNED);
+      pagecache_set_block_rec_lsn(block, first_REDO_LSN_for_page);
+    }
+
     if (need_lock_change)
     {
       /*
@@ -3612,6 +3640,10 @@ static int flush_cached_blocks(PAGECACHE *pagecache,
                         PCBLOCK_NUMBER(pagecache, block), (ulong)block,
                         block->pins));
     DBUG_ASSERT(block->pins == 1);
+    /**
+       @todo If page is contiguous with next page to flush, group flushes in
+       one single my_pwrite().
+    */
     error= pagecache_fwrite(pagecache, file,
                             block->buffer,
                             block->hash_link->pageno,

storage/maria/ma_pagecache.h

@@ -204,12 +204,12 @@ extern uchar *pagecache_valid_read(PAGECACHE *pagecache,
                                   pagecache_disk_read_validator validator,
                                   uchar* validator_data);
 
-#define  pagecache_write(P,F,N,L,B,T,O,I,M,K) \
-   pagecache_write_part(P,F,N,L,B,T,O,I,M,K,0,(P)->block_size,0,0)
+#define  pagecache_write(P,F,N,L,B,T,O,I,M,K,R) \
+   pagecache_write_part(P,F,N,L,B,T,O,I,M,K,R,0,(P)->block_size,0,0)
 
-#define  pagecache_inject(P,F,N,L,B,T,O,I,K,V,D) \
+#define  pagecache_inject(P,F,N,L,B,T,O,I,K,R,V,D) \
    pagecache_write_part(P,F,N,L,B,T,O,I,PAGECACHE_WRITE_DONE, \
-                        K,0,(P)->block_size,V,D)
+                        K,R,0,(P)->block_size,V,D)
 
 extern my_bool pagecache_write_part(PAGECACHE *pagecache,
                                     PAGECACHE_FILE *file,
@@ -221,6 +221,7 @@ extern my_bool pagecache_write_part(PAGECACHE *pagecache,
                                     enum pagecache_page_pin pin,
                                     enum pagecache_write_mode write_mode,
                                     PAGECACHE_BLOCK_LINK **link,
+                                    LSN first_REDO_LSN_for_page,
                                     uint offset,
                                     uint size,
                                     pagecache_disk_read_validator validator,

storage/maria/ma_preload.c

@@ -86,7 +86,8 @@ int maria_preload(MARIA_HA *info, ulonglong key_map, my_bool ignore_leaves)
                             PAGECACHE_PLAIN_PAGE,
                             PAGECACHE_LOCK_LEFT_UNLOCKED,
                             PAGECACHE_PIN_LEFT_UNPINNED,
-                            PAGECACHE_WRITE_DONE, 0))
+                            PAGECACHE_WRITE_DONE, 0,
+			    LSN_IMPOSSIBLE))
           goto err;
       }
       pos+= block_length;

storage/maria/ma_recovery.c

@@ -49,6 +49,10 @@ static HASH all_dirty_pages;
 static struct st_dirty_page *dirty_pages_pool;
 static LSN current_group_end_lsn,
   checkpoint_start= LSN_IMPOSSIBLE;
+#ifndef DBUG_OFF
+/** Current group of REDOs is about this table and only this one */
+static MARIA_HA *current_group_table;
+#endif
 static TrID max_long_trid= 0; /**< max long trid seen by REDO phase */
 static FILE *tracef; /**< trace file for debugging */
 static my_bool skip_DDLs; /**< if REDO phase should skip DDL records */
@@ -56,6 +60,7 @@ static my_bool skip_DDLs; /**< if REDO phase should skip DDL records */
 static my_bool checkpoint_useful;
 static ulonglong now; /**< for tracking execution time of phases */
 static char preamble[]= "Maria engine: starting recovery; ";
+uint warnings; /**< count of warnings */
 
 #define prototype_redo_exec_hook(R)                                          \
   static int exec_REDO_LOGREC_ ## R(const TRANSLOG_HEADER_BUFFER *rec)
@@ -74,6 +79,7 @@ prototype_redo_exec_hook(REDO_RENAME_TABLE);
 prototype_redo_exec_hook(REDO_REPAIR_TABLE);
 prototype_redo_exec_hook(REDO_DROP_TABLE);
 prototype_redo_exec_hook(FILE_ID);
+prototype_redo_exec_hook(INCOMPLETE_LOG);
 prototype_redo_exec_hook(REDO_INSERT_ROW_HEAD);
 prototype_redo_exec_hook(REDO_INSERT_ROW_TAIL);
 prototype_redo_exec_hook(REDO_INSERT_ROW_BLOBS);
@@ -175,22 +181,36 @@ int maria_recover(void)
 {
   int res= 1;
   FILE *trace_file;
+  uint warnings_count;
   DBUG_ENTER("maria_recover");
 
   DBUG_ASSERT(!maria_in_recovery);
   maria_in_recovery= TRUE;
 
 #ifdef EXTRA_DEBUG
-  trace_file= fopen("maria_recovery.trace", "w");
+  trace_file= fopen("maria_recovery.trace", "a+");
 #else
   trace_file= NULL; /* no trace file for being fast */
 #endif
   tprint(trace_file, "TRACE of the last MARIA recovery from mysqld\n");
   DBUG_ASSERT(maria_pagecache->inited);
   res= maria_apply_log(LSN_IMPOSSIBLE, MARIA_LOG_APPLY, trace_file,
-                       TRUE, TRUE, TRUE);
+                       TRUE, TRUE, TRUE, &warnings_count);
   if (!res)
+  {
+    if (warnings_count == 0)
       tprint(trace_file, "SUCCESS\n");
+    else
+    {
+      tprint(trace_file, "DOUBTFUL (%u warnings, check previous output)\n",
+             warnings_count);
+      /*
+        We asked for execution of UNDOs, and skipped DDLs, so shouldn't get
+        any warnings.
+      */
+      DBUG_ASSERT(0);
+    }
+  }
   if (trace_file)
     fclose(trace_file);
   maria_in_recovery= FALSE;
@@ -208,6 +228,7 @@ int maria_recover(void)
    @param  skip_DDLs_arg   Should DDL records (CREATE/RENAME/DROP/REPAIR)
                            be skipped by the REDO phase or not
    @param  take_checkpoints Should we take checkpoints or not.
+   @param[out] warnings_count Count of warnings will be put there
 
    @todo This trace_file thing is primitive; soon we will make it similar to
    ma_check_print_warning() etc, and a successful recovery does not need to
@@ -221,7 +242,7 @@ int maria_recover(void)
 int maria_apply_log(LSN from_lsn, enum maria_apply_log_way apply,
                     FILE *trace_file,
                     my_bool should_run_undo_phase, my_bool skip_DDLs_arg,
-                    my_bool take_checkpoints)
+                    my_bool take_checkpoints, uint *warnings_count)
 {
   int error= 0;
   uint unfinished_trans;
@@ -230,6 +251,7 @@ int maria_apply_log(LSN from_lsn, enum maria_apply_log_way apply,
 
   DBUG_ASSERT(apply == MARIA_LOG_APPLY || !should_run_undo_phase);
   DBUG_ASSERT(!maria_multi_threaded);
+  warnings= 0;
   /* checkpoints can happen only if TRNs have been built */
   DBUG_ASSERT(should_run_undo_phase || !take_checkpoints);
   all_active_trans= (struct st_trn_for_recovery *)
@@ -246,31 +268,7 @@ int maria_apply_log(LSN from_lsn, enum maria_apply_log_way apply,
 
   recovery_message_printed= REC_MSG_NONE;
   tracef= trace_file;
-  if (!(skip_DDLs= skip_DDLs_arg))
-  {
-    /*
-      Example of what can go wrong when replaying DDLs:
-      CREATE TABLE t (logged); INSERT INTO t VALUES(1) (logged);
-      ALTER TABLE t ... which does
-        CREATE a temporary table #sql... (logged)
-        INSERT data from t into #sql... (not logged)
-        RENAME #sql TO t (logged)
-      Removing tables by hand and replaying the log will leave in the
-      end an empty table "t": missing records. If after the RENAME an INSERT
-      into t was done, that row had number 1 in its page, executing the
-      REDO_INSERT_ROW_HEAD on the recreated empty t will fail (assertion
-      failure in _ma_apply_redo_insert_row_head_or_tail(): new data page is
-      created whereas rownr is not 0).
-      Another issue is that replaying of DDLs is not correct enough to work if
-      there was a crash during a DDL (see comment in execution of
-      REDO_RENAME_TABLE ).
-    */
-    tprint(tracef, "WARNING: MySQL server currently disables log records"
-           " about insertion of data by ALTER TABLE"
-           " (copy_data_between_tables()), applying of log records may"
-           " well not work. Additionally, applying of DDL records will"
-           " cause damage if there are tables left by a crash of a DDL.\n");
-  }
+  skip_DDLs= skip_DDLs_arg;
 
   if (from_lsn == LSN_IMPOSSIBLE)
   {
@@ -316,7 +314,7 @@ int maria_apply_log(LSN from_lsn, enum maria_apply_log_way apply,
      start from the checkpoint and never from before, wrongly skipping REDOs
      (tested).
 
-     @todo fix this.
+     @todo fix this; pagecache_write() now can have a rec_lsn argument.
   */
 #if 0
   if (take_checkpoints && checkpoint_useful)
@@ -337,8 +335,11 @@ int maria_apply_log(LSN from_lsn, enum maria_apply_log_way apply,
       goto err;
   }
   else if (unfinished_trans > 0)
-    tprint(tracef, "WARNING: %u unfinished transactions; some tables may be"
-           " left inconsistent!\n", unfinished_trans);
+  {
+    tprint(tracef, "***WARNING: %u unfinished transactions; some tables may"
+           " be left inconsistent!***\n", unfinished_trans);
+    warnings++;
+  }
 
   old_now= now;
   now= my_getsystime();
@@ -389,6 +390,7 @@ int maria_apply_log(LSN from_lsn, enum maria_apply_log_way apply,
   log_record_buffer.str= NULL;
   log_record_buffer.length= 0;
   ma_checkpoint_end();
+  *warnings_count= warnings;
   if (recovery_message_printed != REC_MSG_NONE)
   {
     /** @todo RECOVERY BUG all prints to stderr should go to error log */
@@ -496,6 +498,48 @@ prototype_redo_exec_hook_dummy(CHECKPOINT)
 }
 
 
+prototype_redo_exec_hook(INCOMPLETE_LOG)
+{
+  MARIA_HA *info;
+  if (skip_DDLs)
+  {
+    tprint(tracef, "we skip DDLs\n");
+    return 0;
+  }
+  if ((info= get_MARIA_HA_from_REDO_record(rec)) == NULL)
+  {
+    /* no such table, don't need to warn */
+    return 0;
+  }
+  /*
+    Example of what can go wrong when replaying DDLs:
+    CREATE TABLE t (logged); INSERT INTO t VALUES(1) (logged);
+    ALTER TABLE t ... which does
+    CREATE a temporary table #sql... (logged)
+    INSERT data from t into #sql... (not logged)
+    RENAME #sql TO t (logged)
+    Removing tables by hand and replaying the log will leave in the
+    end an empty table "t": missing records. If after the RENAME an INSERT
+    into t was done, that row had number 1 in its page, executing the
+    REDO_INSERT_ROW_HEAD on the recreated empty t will fail (assertion
+    failure in _ma_apply_redo_insert_row_head_or_tail(): new data page is
+    created whereas rownr is not 0).
+    So when the server disables logging for ALTER TABLE or CREATE SELECT, it
+    logs LOGREC_INCOMPLETE_LOG to warn maria_read_log and then the user.
+
+    Another issue is that replaying of DDLs is not correct enough to work if
+    there was a crash during a DDL (see comment in execution of
+    REDO_RENAME_TABLE ).
+  */
+  tprint(tracef, "***WARNING: MySQL server currently logs no records"
+         " about insertion of data by ALTER TABLE and CREATE SELECT,"
+         " as they are not necessary for recovery;"
+         " present applying of log records may well not work.***\n");
+  warnings++;
+  return 0;
+}
+
+
 prototype_redo_exec_hook(REDO_CREATE_TABLE)
 {
   File dfile= -1, kfile= -1;
@@ -711,7 +755,7 @@ prototype_redo_exec_hook(REDO_RENAME_TABLE)
     scratch. It means that "maria_read_log -a" should not be used on a
     database which just crashed during a DDL. And also ALTER TABLE does not
     log insertions of records into the temporary table, so replaying may
-    fail (see comment and warning in maria_apply_log()).
+    fail (grep for INCOMPLETE_LOG in files).
   */
   info= maria_open(old_name, O_RDONLY, HA_OPEN_FOR_REPAIR);
   if (info)
@@ -1048,7 +1092,11 @@ static int new_table(uint16 sid, const char *name,
   if (share->reopen != 1)
   {
     tprint(tracef, ", is already open (reopen=%u)\n", share->reopen);
-    ALERT_USER();
+    /*
+      It could be that we have in the log
+      FILE_ID(t1,10) ... (t1 was flushed) ... FILE_ID(t1,12);
+    */
+    if (close_one_table(share->open_file_name, lsn_of_file_id))
       goto end;
   }
   DBUG_ASSERT(share->now_transactional == share->base.born_transactional);
@@ -1069,7 +1117,9 @@ static int new_table(uint16 sid, const char *name,
     goto end;
   }
   /* don't log any records for this work */
-  _ma_tmp_disable_logging_for_table(share);
+  _ma_tmp_disable_logging_for_table(info, FALSE);
+  /* _ma_unpin_all_pages() reads info->trn: */
+  info->trn= &dummy_transaction_object;
   /* execution of some REDO records relies on data_file_length */
   my_off_t dfile_len= my_seek(info->dfile.file, 0, SEEK_END, MYF(MY_WME));
   my_off_t kfile_len= my_seek(info->s->kfile.file, 0, SEEK_END, MYF(MY_WME));
@@ -1079,8 +1129,16 @@ static int new_table(uint16 sid, const char *name,
     tprint(tracef, ", length unknown\n");
     goto end;
   }
+  if (share->state.state.data_file_length != dfile_len)
+  {
+    tprint(tracef, ", has wrong state.data_file_length (fixing it)");
     share->state.state.data_file_length= dfile_len;
+  }
+  if (share->state.state.key_file_length != kfile_len)
+  {
+    tprint(tracef, ", has wrong state.key_file_length (fixing it)");
     share->state.state.key_file_length= kfile_len;
+  }
   if ((dfile_len % share->block_size) > 0)
   {
     tprint(tracef, ", has too short last page\n");
@@ -1416,7 +1474,8 @@ prototype_redo_exec_hook(UNDO_ROW_INSERT)
   set_undo_lsn_for_active_trans(rec->short_trid, rec->lsn);
   if (cmp_translog_addr(rec->lsn, share->state.is_of_horizon) >= 0)
   {
-    tprint(tracef, "   state older than record, updating rows' count\n");
+    tprint(tracef, "   state has LSN (%lu,0x%lx) older than record, updating"
+           " rows' count\n", LSN_IN_PARTS(share->state.is_of_horizon));
     share->state.state.records++;
     if (share->calc_checksum)
     {
@@ -1438,6 +1497,8 @@ prototype_redo_exec_hook(UNDO_ROW_INSERT)
     info->s->state.changed|= STATE_CHANGED | STATE_NOT_ANALYZED;
   }
   tprint(tracef, "   rows' count %lu\n", (ulong)info->s->state.state.records);
+  /* Unpin all pages, stamp them with UNDO's LSN */
+  _ma_unpin_all_pages(info, rec->lsn);
   return 0;
 }
 
@@ -1471,6 +1532,7 @@ prototype_redo_exec_hook(UNDO_ROW_DELETE)
     share->state.changed|= STATE_CHANGED | STATE_NOT_ANALYZED;
   }
   tprint(tracef, "   rows' count %lu\n", (ulong)share->state.state.records);
+  _ma_unpin_all_pages(info, rec->lsn);
   return 0;
 }
 
@@ -1500,21 +1562,27 @@ prototype_redo_exec_hook(UNDO_ROW_UPDATE)
     }
     share->state.changed|= STATE_CHANGED | STATE_NOT_ANALYZED;
   }
+  _ma_unpin_all_pages(info, rec->lsn);
   return 0;
 }
 
+
 prototype_redo_exec_hook(UNDO_KEY_INSERT)
 {
   set_undo_lsn_for_active_trans(rec->short_trid, rec->lsn);
+  _ma_unpin_all_pages(info, rec->lsn);
   return 0;
 }
 
+
 prototype_redo_exec_hook(UNDO_KEY_DELETE)
 {
   set_undo_lsn_for_active_trans(rec->short_trid, rec->lsn);
+  _ma_unpin_all_pages(info, rec->lsn);
   return 0;
 }
 
+
 prototype_redo_exec_hook(UNDO_KEY_DELETE_WITH_ROOT)
 {
   MARIA_HA *info= get_MARIA_HA_from_UNDO_record(rec);
@@ -1534,6 +1602,7 @@ prototype_redo_exec_hook(UNDO_KEY_DELETE_WITH_ROOT)
                                     HA_OFFSET_ERROR :
                                     page * share->block_size);
   }
+  _ma_unpin_all_pages(info, rec->lsn);
   return 0;
 }
 
@@ -1626,6 +1695,7 @@ prototype_redo_exec_hook(CLR_END)
     share->state.changed|= STATE_CHANGED | STATE_NOT_ANALYZED;
   }
   tprint(tracef, "   rows' count %lu\n", (ulong)share->state.state.records);
+  _ma_unpin_all_pages(info, rec->lsn);
   return 0;
 }
 
@@ -1918,6 +1988,7 @@ static int run_redo_phase(LSN lsn, enum maria_apply_log_way apply)
   install_redo_exec_hook(REDO_REPAIR_TABLE);
   install_redo_exec_hook(REDO_DROP_TABLE);
   install_redo_exec_hook(FILE_ID);
+  install_redo_exec_hook(INCOMPLETE_LOG);
   install_redo_exec_hook(REDO_INSERT_ROW_HEAD);
   install_redo_exec_hook(REDO_INSERT_ROW_TAIL);
   install_redo_exec_hook(REDO_INSERT_ROW_BLOBS);
@@ -1945,6 +2016,9 @@ static int run_redo_phase(LSN lsn, enum maria_apply_log_way apply)
   install_undo_exec_hook(UNDO_KEY_DELETE_WITH_ROOT);
 
   current_group_end_lsn= LSN_IMPOSSIBLE;
+#ifndef DBUG_OFF
+  current_group_table= NULL;
+#endif
 
   if (unlikely(lsn == LSN_IMPOSSIBLE || lsn == translog_get_horizon()))
   {
@@ -2060,6 +2134,9 @@ static int run_redo_phase(LSN lsn, enum maria_apply_log_way apply)
       if (apply == MARIA_LOG_APPLY &&
           display_and_apply_record(log_desc, &rec))
         goto err;
+#ifndef DBUG_OFF
+      current_group_table= NULL;
+#endif
     }
     else /* record does not end group */
     {
@@ -2268,6 +2345,7 @@ static void prepare_table_for_close(MARIA_HA *info, TRANSLOG_ADDRESS horizon)
     _ma_state_info_write_sub(share->kfile.file, &share->state, 1);
   }
   _ma_reenable_logging_for_table(share);
+  info->trn= NULL; /* safety */
 }
 
 
@@ -2299,6 +2377,10 @@ static MARIA_HA *get_MARIA_HA_from_REDO_record(const
     break;
   }
   info= all_tables[sid].info;
+#ifndef DBUG_OFF
+  DBUG_ASSERT(current_group_table == NULL || current_group_table == info);
+  current_group_table= info;
+#endif
   if (info == NULL)
   {
     tprint(tracef, ", table skipped, so skipping record\n");
@@ -2361,6 +2443,10 @@ static MARIA_HA *get_MARIA_HA_from_UNDO_record(const
   sid= fileid_korr(rec->header + LSN_STORE_SIZE);
   tprint(tracef, "   For table of short id %u", sid);
   info= all_tables[sid].info;
+#ifndef DBUG_OFF
+  DBUG_ASSERT(current_group_table == NULL || current_group_table == info);
+  current_group_table= info;
+#endif
   if (info == NULL)
   {
     tprint(tracef, ", table skipped, so skipping record\n");
@@ -2627,6 +2713,39 @@ static my_bool close_one_table(const char *name, TRANSLOG_ADDRESS addr)
 }
 
 
+/**
+   Temporarily disables logging for this table.
+
+   If that makes the log incomplete, writes a LOGREC_INCOMPLETE_LOG to the log
+   to warn log readers.
+
+   @param  info            table
+   @param  log_incomplete  if that disabling makes the log incomplete
+
+   @note for example in the REDO phase we disable logging but that does not
+   make the log incomplete.
+*/
+void _ma_tmp_disable_logging_for_table(MARIA_HA *info,
+                                       my_bool log_incomplete)
+{
+  MARIA_SHARE *share= info->s;
+  if (log_incomplete)
+  {
+    uchar log_data[FILEID_STORE_SIZE];
+    LEX_STRING log_array[TRANSLOG_INTERNAL_PARTS + 1];
+    LSN lsn;
+    log_array[TRANSLOG_INTERNAL_PARTS + 0].str=    (char*) log_data;
+    log_array[TRANSLOG_INTERNAL_PARTS + 0].length= sizeof(log_data);
+    translog_write_record(&lsn, LOGREC_INCOMPLETE_LOG,
+                          info->trn, info, sizeof(log_data),
+                          TRANSLOG_INTERNAL_PARTS + 1, log_array,
+                          log_data, NULL);
+  }
+  /* if we disabled before writing the record, record wouldn't reach log */
+  share->now_transactional= FALSE;
+  share->page_type= PAGECACHE_PLAIN_PAGE;
+}
+
 static void print_redo_phase_progress(TRANSLOG_ADDRESS addr)
 {
   static int end_logno= FILENO_IMPOSSIBLE, end_offset, percentage_printed= 0;

storage/maria/ma_recovery.h

@@ -30,5 +30,5 @@ int maria_recover(void);
 int maria_apply_log(LSN lsn, enum maria_apply_log_way apply,
                     FILE *trace_file,
                     my_bool execute_undo_phase, my_bool skip_DDLs,
-                    my_bool take_checkpoints);
+                    my_bool take_checkpoints, uint *warnings_count);
 C_MODE_END

storage/maria/ma_write.c

@@ -218,8 +218,15 @@ int maria_write(MARIA_HA *info, uchar *record)
       info->cur_row.checksum;
   }
   if (share->base.auto_key)
+  {
+    /**
+       @todo RECOVERY BUG
+       if updated here, it's not recoverable (no mutex => checkpoint may see a
+       crazy value and flush it into the table's state on disk).
+    */
     set_if_bigger(info->s->state.auto_increment,
                   ma_retrieve_auto_increment(info, record));
+  }
   info->update= (HA_STATE_CHANGED | HA_STATE_AKTIV | HA_STATE_WRITTEN |
 		 HA_STATE_ROW_CHANGED);
   info->state->records+= !share->now_transactional; /*otherwise already done*/

storage/maria/maria_def.h

@@ -1033,8 +1033,8 @@ int _ma_update_create_rename_lsn(MARIA_SHARE *share,
 int _ma_update_create_rename_lsn_sub(MARIA_SHARE *share,
                                      LSN lsn, my_bool do_sync);
 
-#define _ma_tmp_disable_logging_for_table(S) \
-  { (S)->now_transactional= FALSE; (S)->page_type= PAGECACHE_PLAIN_PAGE; }
+void _ma_tmp_disable_logging_for_table(MARIA_HA *info,
+                                       my_bool log_incomplete);
 #define _ma_reenable_logging_for_table(S)                        \
   { if (((S)->now_transactional= (S)->base.born_transactional))  \
       (S)->page_type= PAGECACHE_LSN_PAGE; }

storage/maria/maria_read_log.c

@@ -38,6 +38,7 @@ int main(int argc, char **argv)
 {
   LSN lsn;
   char **default_argv;
+  uint warnings_count;
   MY_INIT(argv[0]);
   my_progname_short= my_progname+dirname_length(my_progname);
 
@@ -106,9 +107,13 @@ int main(int argc, char **argv)
   if (maria_apply_log(lsn, opt_apply ?  MARIA_LOG_APPLY :
                       (opt_check ? MARIA_LOG_CHECK :
                        MARIA_LOG_DISPLAY_HEADER), opt_silent ? NULL : stdout,
-                      opt_apply_undo, FALSE, FALSE))
+                      opt_apply_undo, FALSE, FALSE, &warnings_count))
     goto err;
+  if (warnings_count == 0)
     fprintf(stdout, "%s: SUCCESS\n", my_progname_short);
+  else
+    fprintf(stdout, "%s: DOUBTFUL (%u warnings, check previous output)\n",
+            my_progname_short, warnings_count);
 
   goto end;
 err:
@@ -130,7 +135,8 @@ int main(int argc, char **argv)
 static struct my_option my_long_options[] =
 {
   {"apply", 'a',
-   "Apply log to tables. Will display a lot of information if not run with --silent",
+   "Apply log to tables: modifies tables! you should make a backup first! "
+   " Displays a lot of information if not run with --silent",
    (uchar **) &opt_apply, (uchar **) &opt_apply, 0,
    GET_BOOL, NO_ARG, 0, 0, 0, 0, 0, 0},
   {"check", 'c',
@@ -143,7 +149,7 @@ static struct my_option my_long_options[] =
 #endif
   {"help", '?', "Display this help and exit.",
    0, 0, 0, GET_NO_ARG, NO_ARG, 0, 0, 0, 0, 0, 0},
-  {"only-display", 'o', "display brief info about records's header",
+  {"only-display", 'o', "display brief info read from records' header",
    (uchar **) &opt_only_display, (uchar **) &opt_only_display, 0, GET_BOOL,
    NO_ARG,0, 0, 0, 0, 0, 0},
   { "page_buffer_size", 'P', "",
@@ -154,7 +160,7 @@ static struct my_option my_long_options[] =
   {"silent", 's', "Print less information during apply/undo phase",
    (uchar **) &opt_silent, (uchar **) &opt_silent, 0,
    GET_BOOL, NO_ARG, 0, 0, 0, 0, 0, 0},
-  {"undo", 'u', "Apply undos to tables. (disable with --disable-undo)",
+  {"undo", 'u', "Apply UNDO records to tables. (disable with --disable-undo)",
    (uchar **) &opt_apply_undo, (uchar **) &opt_apply_undo, 0,
    GET_BOOL, NO_ARG, 1, 0, 0, 0, 0, 0},
   {"version", 'V', "Print version and exit.",

storage/maria/unittest/ma_pagecache_consist.c

@@ -237,7 +237,7 @@ void writer(int num)
                     PAGECACHE_LOCK_WRITE_UNLOCK,
                     PAGECACHE_UNPIN,
                     PAGECACHE_WRITE_DELAY,
-                    0);
+                    0, LSN_IMPOSSIBLE);
 
     if (i % flush_divider == 0)
       flush_pagecache_blocks(&pagecache, &file1, FLUSH_FORCE_WRITE);
@@ -380,7 +380,7 @@ int main(int argc __attribute__((unused)),
                       PAGECACHE_LOCK_LEFT_UNLOCKED,
                       PAGECACHE_PIN_LEFT_UNPINNED,
                       PAGECACHE_WRITE_DELAY,
-                      0);
+                      0, LSN_IMPOSSIBLE);
     }
     flush_pagecache_blocks(&pagecache, &file1, FLUSH_FORCE_WRITE);
     free(buffr);

storage/maria/unittest/ma_pagecache_single.c

@@ -104,7 +104,7 @@ int simple_read_write_test()
                   PAGECACHE_LOCK_LEFT_UNLOCKED,
                   PAGECACHE_PIN_LEFT_UNPINNED,
                   PAGECACHE_WRITE_DELAY,
-                  0);
+                  0, LSN_IMPOSSIBLE);
   pagecache_read(&pagecache, &file1, 0, 3, (char*)buffr,
                  PAGECACHE_PLAIN_PAGE,
                  PAGECACHE_LOCK_LEFT_UNLOCKED,
@@ -140,7 +140,7 @@ int simple_read_change_write_read_test()
                   PAGECACHE_LOCK_LEFT_UNLOCKED,
                   PAGECACHE_PIN_LEFT_UNPINNED,
                   PAGECACHE_WRITE_DELAY,
-                  0);
+                  0, LSN_IMPOSSIBLE);
   flush_pagecache_blocks(&pagecache, &file1, FLUSH_FORCE_WRITE);
   /* test */
   pagecache_read(&pagecache, &file1, 0, 3, (char*)buffw,
@@ -153,7 +153,7 @@ int simple_read_change_write_read_test()
                   PAGECACHE_LOCK_WRITE_UNLOCK,
                   PAGECACHE_UNPIN,
                   PAGECACHE_WRITE_DELAY,
-                  0);
+                  0, LSN_IMPOSSIBLE);
 
   pagecache_read(&pagecache, &file1, 0, 3, (char*)buffr,
                  PAGECACHE_PLAIN_PAGE,
@@ -194,7 +194,7 @@ int simple_pin_test()
                   PAGECACHE_LOCK_LEFT_UNLOCKED,
                   PAGECACHE_PIN_LEFT_UNPINNED,
                   PAGECACHE_WRITE_DELAY,
-                  0);
+                  0, LSN_IMPOSSIBLE);
   /* test */
   if (flush_pagecache_blocks(&pagecache, &file1, FLUSH_FORCE_WRITE))
   {
@@ -210,14 +210,14 @@ int simple_pin_test()
                   PAGECACHE_LOCK_LEFT_UNLOCKED,
                   PAGECACHE_PIN_LEFT_UNPINNED,
                   PAGECACHE_WRITE_DELAY,
-                  0);
+                  0, LSN_IMPOSSIBLE);
   bfill(buffw + PAGE_SIZE/2, PAGE_SIZE/2, ((unsigned char) 129));
   pagecache_write(&pagecache, &file1, 0, 3, (char*)buffw,
                   PAGECACHE_PLAIN_PAGE,
                   PAGECACHE_LOCK_WRITE_TO_READ,
                   PAGECACHE_PIN_LEFT_PINNED,
                   PAGECACHE_WRITE_DELAY,
-                  0);
+                  0, LSN_IMPOSSIBLE);
   /*
     We have to get error because one page of the file is pinned,
     other page should be flushed
@@ -272,7 +272,7 @@ int simple_delete_forget_test()
                   PAGECACHE_LOCK_LEFT_UNLOCKED,
                   PAGECACHE_PIN_LEFT_UNPINNED,
                   PAGECACHE_WRITE_DELAY,
-                  0);
+                  0, LSN_IMPOSSIBLE);
   flush_pagecache_blocks(&pagecache, &file1, FLUSH_FORCE_WRITE);
   /* test */
   bfill(buffw, PAGE_SIZE, '\2');
@@ -281,7 +281,7 @@ int simple_delete_forget_test()
                   PAGECACHE_LOCK_LEFT_UNLOCKED,
                   PAGECACHE_PIN_LEFT_UNPINNED,
                   PAGECACHE_WRITE_DELAY,
-                  0);
+                  0, LSN_IMPOSSIBLE);
   pagecache_delete(&pagecache, &file1, 0,
                    PAGECACHE_LOCK_WRITE, 0);
   flush_pagecache_blocks(&pagecache, &file1, FLUSH_FORCE_WRITE);
@@ -314,7 +314,7 @@ int simple_delete_flush_test()
                   PAGECACHE_LOCK_WRITE,
                   PAGECACHE_PIN,
                   PAGECACHE_WRITE_DELAY,
-                  0);
+                  0, LSN_IMPOSSIBLE);
   flush_pagecache_blocks(&pagecache, &file1, FLUSH_FORCE_WRITE);
   /* test */
   bfill(buffw, PAGE_SIZE, '\2');
@@ -323,7 +323,7 @@ int simple_delete_flush_test()
                   PAGECACHE_LOCK_LEFT_WRITELOCKED,
                   PAGECACHE_PIN_LEFT_PINNED,
                   PAGECACHE_WRITE_DELAY,
-                  0);
+                  0, LSN_IMPOSSIBLE);
   pagecache_delete(&pagecache, &file1, 0,
                    PAGECACHE_LOCK_LEFT_WRITELOCKED, 1);
   flush_pagecache_blocks(&pagecache, &file1, FLUSH_FORCE_WRITE);
@@ -362,7 +362,7 @@ int simple_big_test()
                     PAGECACHE_LOCK_LEFT_UNLOCKED,
                     PAGECACHE_PIN_LEFT_UNPINNED,
                     PAGECACHE_WRITE_DELAY,
-                    0);
+                    0, LSN_IMPOSSIBLE);
   }
   desc[i].length= 0;
   desc[i].content= '\0';

storage/maria/unittest/ma_test_loghandler-t.c

@@ -185,7 +185,7 @@ int main(int argc __attribute__((unused)), char *argv[])
   /* Suppressing of automatic record writing */
   trn->first_undo_lsn|= TRANSACTION_LOGGED_LONG_ID;
 
-  plan(((ITERATIONS - 1) * 4 + 1)*2 + ITERATIONS - 1);
+  plan(((ITERATIONS - 1) * 4 + 1)*2 + ITERATIONS - 1 + 1);
 
   srandom(122334817L);
 
@@ -335,6 +335,15 @@ int main(int argc __attribute__((unused)), char *argv[])
     ok(1, "flush");
   }
 
+  if (translog_flush(translog_get_horizon()))
+  {
+    fprintf(stderr, "Can't flush up to horizon\n", (ulong) i);
+    translog_destroy();
+    ok(0, "flush");
+    exit(1);
+  }
+  ok(1, "flush");
+
   srandom(122334817L);
 
   rc= 1;

storage/maria/unittest/ma_test_loghandler_pagecache-t.c

@@ -129,7 +129,7 @@ int main(int argc __attribute__((unused)), char *argv[])
                     PAGECACHE_LOCK_LEFT_UNLOCKED,
                     PAGECACHE_PIN_LEFT_UNPINNED,
                     PAGECACHE_WRITE_DELAY,
-                    0);
+                    0, LSN_IMPOSSIBLE);
     flush_pagecache_blocks(&pagecache, &file1, FLUSH_FORCE_WRITE);
   }
   if ((stat= my_stat(first_translog_file, &st,  MYF(0))) == 0)