Update progress.t, option_sanity.t, and standard_options.t. Make test env always have slave2 (12347, slave of 12346). Add PerconaTest::count_checksum_results(). Add util/wait-to-exec.

lib/PerconaTest.pm

@@ -224,11 +224,11 @@ sub wait_until {
 
    my $slept = 0;
    while ( $slept <= $max_t ) {
-      return if $code->();
+      return 1 if $code->();
       sleep $t;
       $slept += $t;
    }
-   return;
+   return 0;
 }
 
 # Wait t seconds for code to return.
@@ -558,6 +558,41 @@ sub test_bash_tool {
    return;
 }
 
+my %checksum_result_col = (
+   ts      => 0,
+   errors  => 1,
+   diffs   => 2,
+   rows    => 3,
+   chunks  => 4,
+   skipped => 5,
+   time    => 6,
+   table   => 7,
+);
+sub count_checksum_results {
+   my ($output, $column, $table) = @_;
+
+   my (@res) = map {
+      my $line = $_;
+      my (@cols) = $line =~ m/(\S+)/g;
+      \@cols;
+   }
+   grep {
+      my $line = $_;
+      if ( !$table ) {
+         $line;
+      }
+      else {
+         $line =~ m/$table$/m ? $line : '';
+      }
+   }
+   grep { m/^\d+\-\d+T/ } split /\n/, $output;
+   my $colno = $checksum_result_col{lc $column};
+   die "Invalid checksum result column: $column" unless defined $colno;
+   my $total = 0;
+   map { $total += $_->[$colno] } @res;
+   return $total;
+}
+
 1;
 }
 # ###########################################################################

sandbox/load-sakila-db

@@ -46,7 +46,4 @@ exit_status=$((exit_status | $?))
 /tmp/$PORT/use < sakila-db/sakila-data.sql
 exit_status=$((exit_status | $?))
 
-$PERCONA_TOOLKIT_BRANCH/sandbox/test-env reset
-exit_status=$((exit_status | $?))
-
 exit $exit_status

sandbox/test-env

@@ -291,6 +291,7 @@ case $opt in
       exit_status=$?
       if [ $exit_status -eq 0 ]; then
          ./start-sandbox slave 12346 12345
+         ./start-sandbox slave 12347 12346
          exit_status=$?
          if [ "$version" != "4.0" ] && [ "$version" != "4.1" ]; then
             if [ $? -eq 0 ]; then
@@ -353,11 +354,18 @@ case $opt in
       # don't replicate to new sandbox servers.  This makes creating new
       # sandbox servers a lot faster.  There's no check if this works or
       # not, so... yeah.
-      /tmp/12345/use -e "RESET MASTER"
-      /tmp/12346/use -e "RESET MASTER"
+      /tmp/12347/use -e "STOP SLAVE"
       /tmp/12346/use -e "STOP SLAVE"
+
+      /tmp/12346/use -e "RESET MASTER"
+      /tmp/12345/use -e "RESET MASTER"
+
+      /tmp/12347/use -e "change master to master_host='127.0.0.1', master_user='msandbox', master_password='msandbox', master_port=12346, master_log_file='mysql-bin.000001', master_log_pos=0"
       /tmp/12346/use -e "change master to master_host='127.0.0.1', master_user='msandbox', master_password='msandbox', master_port=12345, master_log_file='mysql-bin.000001', master_log_pos=0"
+
+      /tmp/12347/use -e "START SLAVE"
       /tmp/12346/use -e "START SLAVE"
+
       exit_status=0
       ;;
    version)

t/pt-table-checksum/ignore_columns.t

@@ -42,21 +42,9 @@ $sb->load_file('master', 't/pt-table-checksum/samples/issue_94.sql');
 PerconaTest::wait_for_table($slave_dbh, 'test.issue_94', 'a=11');
 $slave_dbh->do("update test.issue_94 set c=''");
 
-sub get_diffs {
-   my ($output) = @_;
-   my (@diffs) = $output =~ m/
-      ^\S+\s+  # TS
-      \d+\s+   # ERRORS
-      (\d+)    # DIFFS
-   /gmx;
-   my $total_diffs = 0;
-   map { $total_diffs += $_ } @diffs;
-   return $total_diffs;
-}
-
 $output = output(
    sub { pt_table_checksum::main(@args, qw(-d test -t issue_94)) },
-   trf => \&get_diffs,
+   trf => sub { return PerconaTest::count_checksum_results(@_, 'DIFFS') },
 );
 is(
    $output,
@@ -67,7 +55,7 @@ is(
 $output = output(
    sub { pt_table_checksum::main(@args, qw(-d test -t issue_94),
       qw(--ignore-columns c)) },
-   trf => \&get_diffs,
+   trf => sub { return PerconaTest::count_checksum_results(@_, 'DIFFS') },
 );
 is(
    $output,

t/pt-table-checksum/option_sanity.t

@@ -9,26 +9,132 @@ BEGIN {
 use strict;
 use warnings FATAL => 'all';
 use English qw(-no_match_vars);
-use Test::More tests => 2;
+use Test::More tests => 16;
 
 use PerconaTest;
 require "$trunk/bin/pt-table-checksum";
 
 my $output;
 
-# Test DSN value inheritance
-$output = `$trunk/bin/pt-table-checksum h=127.1 --replicate table`;
+# Calling the tool pt_table_checksum::main() doesn't work when we're
+# dealing with cmd line option error and --help because OptionParser
+# exits on such things which would cause this test to exit too.
+
+# ############################################################################
+# Check default values for some options to ensure something wasn't
+# changed accidentally.  A lot of tests rely on these defaults, so
+# if they change, it can have weird side-effects.
+# ############################################################################
+$output = `$trunk/bin/pt-table-checksum h=127.1 --help`;
+
+like(
+   $output,
+   qr/^  --check-replication-filters\s+TRUE$/m,
+   "Default --check-replication-filters=TRUE"
+);
+
+like(
+   $output,
+   qr/^  --create-replicate-table\s+TRUE$/m,
+   "Default --create-replicate-table=TRUE"
+);
+
+like(
+   $output,
+   qr/^  --empty-replicate-table\s+TRUE$/m,
+   "Default --empty-replicate-table=TRUE"
+);
+
+like(
+   $output,
+   qr/^  --explain\s+0$/m,
+   "Default --explain=0"
+);
+
+like(
+   $output,
+   qr/^  --host\s+localhost$/m,
+   "Default --host=localhost"
+);
+
+like(
+   $output,
+   qr/^  --lock-wait-timeout\s+1$/m,
+   "Default --lock-wait-timeout=1"
+);
+
+like(
+   $output,
+   qr/^  --max-lag\s+1$/m,
+   "Default --max-lag=1"
+);
+
+like(
+   $output,
+   qr/^  --quiet\s+0$/m,
+   "Default --quiet=0"
+);
+
+like(
+   $output,
+   qr/^  --recheck\s+TRUE$/m,
+   "Default --recheck=TRUE"
+);
+
+like(
+   $output,
+   qr/^  --replicate\s+percona\.checksums$/m,
+   "Default --replicate=percona.checksums"
+);
+
+like(
+   $output,
+   qr/^  --replicate-check\s+TRUE$/m,
+   "Default --replicate-check=TRUE"
+);
+
+# ############################################################################
+# Check opts that disable other opts.
+# ############################################################################
+$output = `$trunk/bin/pt-table-checksum h=127.1 --help --explain`;
+like(
+   $output,
+   qr/^  --empty-replicate-table\s+FALSE$/m,
+   "--explain disables --empty-replicate-table"
+);
+
+$output = `$trunk/bin/pt-table-checksum h=127.1 --help --resume`;
+like(
+   $output,
+   qr/^  --empty-replicate-table\s+FALSE$/m,
+   "--resume disables --empty-replicate-table"
+);
+
+$output = `$trunk/bin/pt-table-checksum h=127.1 --help --quiet`;
+like(
+   $output,
+   qr/^  --progress\s+\(No value\)$/m,
+   "--quiet disables --progress"
+);
+
+# ############################################################################
+# Only 1 DSN should be allowed on the command line; no extra args.
+# ############################################################################
+$output = `$trunk/bin/pt-table-checksum h=127.1 h=host1 h=host2`;
+like(
+   $output,
+   qr/More than one host specified; only one allowed/,
+   "Only one DSN allowed on the command line"
+);
+
+# ############################################################################
+# --replicate table must be db-qualified.
+# ############################################################################
+$output = `$trunk/bin/pt-table-checksum h=127.1 --replicate checksums`;
 like(
    $output,
    qr/--replicate table must be database-qualified/,
-   "--replicate table must be db-qualified"
-);
-
-$output = `$trunk/bin/pt-table-checksum h=127.1 --replicate test.checksum --throttle-method foo`;
-like(
-   $output,
-   qr/Invalid --throttle-method: foo/,
-   "Invalid --throttle-method"
+   "--replicate table must be database-qualified"
 );
 
 # #############################################################################

t/pt-table-checksum/progress.t

@@ -15,10 +15,6 @@ use PerconaTest;
 use Sandbox;
 require "$trunk/bin/pt-table-checksum";
 
-diag(`$trunk/sandbox/test-env reset`);
-diag(`$trunk/sandbox/stop-sandbox 12347 >/dev/null`);
-diag(`$trunk/sandbox/start-sandbox slave 12347 12346 >/dev/null`);
-
 my $dp = new DSNParser(opts=>$dsn_opts);
 my $sb = new Sandbox(basedir => '/tmp', DSNParser => $dp);
 my $master_dbh = $sb->get_dbh_for('master');
@@ -38,17 +34,21 @@ else {
    plan tests => 3;
 }
 
+# Must have empty checksums table for these tests.
+$master_dbh->do('drop table if exists percona.checksums');
+
+# The sandbox servers run with lock_wait_timeout=3 and it's not dynamic
+# so we need to specify --lock-wait-timeout=3 else the tool will die.
+my $master_dsn = 'h=127.1,P=12345,u=msandbox,p=msandbox';
+my @args       = ($master_dsn, qw(--lock-wait-timeout 3),
+                  '--progress', 'time,1'); 
 my $output;
 my $row;
-my $cnf  ='/tmp/12345/my.sandbox.cnf';
-my @args = (qw(--replicate test.checksum  --empty-replicate-table -d test -t resume -q), '-F', $cnf, 'h=127.1', '--progress', 'time,1');
-
-$sb->create_dbs($master_dbh, [qw(test)]);
-$sb->load_file('master', 't/pt-table-checksum/samples/checksum_tbl.sql');
-# We're not using resume table for anything specific, we just need
-# any table to checksum.
-$sb->load_file('master', 't/pt-table-checksum/samples/resume.sql');
+my $scripts = "$trunk/t/pt-table-checksum/scripts/";
 
+# ############################################################################
+# Tool should check all slaves' lag, so slave2, not just slave1.
+# ############################################################################
 wait_until(  # slaves aren't lagging
    sub {
       $row = $slave1_dbh->selectrow_hashref('show slave status');
@@ -56,47 +56,40 @@ wait_until(  # slaves aren't lagging
       $row = $slave2_dbh->selectrow_hashref('show slave status');
       return 0 if $row->{Seconds_Behind_Master};
       return 1;
-   }, 0.5, 10);
+   }
+) or die "Slaves are still lagging";
 
-$slave1_dbh->do('stop slave sql_thread');
-$row = $slave1_dbh->selectrow_hashref('show slave status');
-is(
-   $row->{slave_sql_running},
-   'No',
-   'Stopped slave SQL thread on slave1'
-);
+# This big fancy command waits until it sees the checksum for sakila.city
+# in the repl table on the master, then it stops slave2 for 2 seconds,
+# then starts it again.
+system("$trunk/util/wait-to-exec '$scripts/wait-for-chunk.sh 12345 sakila city 1' '$scripts/exec-wait-exec.sh 12347 \"stop slave sql_thread\" 2 \"start slave sql_thread\"' 3 >/dev/null &");
 
-$slave2_dbh->do('stop slave sql_thread');
-$row = $slave2_dbh->selectrow_hashref('show slave status');
-is(
-   $row->{slave_sql_running},
-   'No',
-   'Stopped slave SQL thread on slave2'
-);
-
-system("sleep 2 && /tmp/12346/use -e 'start slave sql_thread' >/dev/null 2>/dev/null &");
-system("sleep 3 && /tmp/12347/use -e 'start slave sql_thread' >/dev/null 2>/dev/null &");
-
-# This time we do not need to capture STDERR because mk-table-checksum
-# should see slave2 come alive in 2 seconds then return before wait_for
-# dies.
 $output = output(
-   sub { pt_table_checksum::main(@args); },
+   sub { pt_table_checksum::main(@args, qw(-d sakila)); },
    stderr => 1,
 );
 
 like(
    $output,
-   qr/Waiting for slave.+?Still waiting/s,
-   "Progress reports while waiting for slaves"
+   qr/Replica h=127.0.0.1,P=12347 is stopped/,
+   "--progress for slave lag"
+);
+
+like(
+   $output,
+   qr/sakila.store$/m,
+   "Checksumming continues after waiting for slave lag"
+);
+
+is(
+   PerconaTest::count_checksum_results($output, 'errors'),
+   0,
+   "No errors after waiting for slave lag"
 );
 
 # #############################################################################
 # Done.
 # #############################################################################
-diag(`$trunk/sandbox/stop-sandbox 12347 >/dev/null`);
-diag(`/tmp/12346/stop >/dev/null`);  # Start/stop clears SHOW SLAVE HOSTS.
-diag(`/tmp/12346/start >/dev/null`);
 $sb->wipe_clean($master_dbh);
 diag(`$trunk/sandbox/test-env reset >/dev/null`);
 exit;

t/pt-table-checksum/samples/no-recheck.txt

@@ -3,3 +3,8 @@ TABLE CHUNK CNT_DIFF CRC_DIFF CHUNK_INDEX LOWER_BOUNDARY UPPER_BOUNDARY
 sakila.city 1 0 1 PRIMARY 1 100
 sakila.city 6 0 1 PRIMARY 501 600
 
+Differences on h=127.0.0.1,P=12347
+TABLE CHUNK CNT_DIFF CRC_DIFF CHUNK_INDEX LOWER_BOUNDARY UPPER_BOUNDARY
+sakila.city 1 0 1 PRIMARY 1 100
+sakila.city 6 0 1 PRIMARY 501 600
+

t/pt-table-checksum/scripts/exec-wait-exec.sh

@@ -0,0 +1,10 @@
+#!/bin/sh
+
+port=$1
+query1=$2
+t=$3
+query2=$4
+
+/tmp/$port/use -e "$query1"
+sleep $t
+/tmp/$port/use -e "$query2"

t/pt-table-checksum/scripts/wait-for-chunk.sh

@@ -0,0 +1,8 @@
+#!/bin/sh
+
+port=$1
+db=$2
+tbl=$3
+chunk=$4
+
+/tmp/$port/use -e "select 1 from percona.checksums where db='$db' and tbl='$tbl' and chunk=$chunk" 2>/dev/null | grep -q 1 2>/dev/null

t/pt-table-checksum/standard_options.t

@@ -27,7 +27,7 @@ elsif ( !$slave_dbh ) {
    plan skip_all => 'Cannot connect to sandbox slave1';
 }
 else {
-   plan tests => 4;
+   plan tests => 5;
 }
 
 # The sandbox servers run with lock_wait_timeout=3 and it's not dynamic
@@ -36,6 +36,37 @@ my @args     = (qw(--lock-wait-timeout 3 --explain --tables sakila.country));
 my $cnf      = "/tmp/12345/my.sandbox.cnf";
 my $pid_file = "/tmp/mk-table-checksum-test.pid";
 my $output;
+my $exit_status;
+
+# ############################################################################
+# Tool should connect to localhost without any options.
+# ############################################################################
+
+# This may not work because the sandbox servers aren't on localhost,
+# but if your box has MySQL running on localhost then maybe it will,
+# so we'll account for both of these possibilities.
+
+eval {
+   $exit_status = pt_table_checksum::main(@args);
+};
+if ( $EVAL_ERROR ) {
+   # It's ok that this fails.  It means that your box, like mine, doesn't
+   # have MySQL on localhost:3306:/tmp/mysql.socket/etc.
+   like(
+      $EVAL_ERROR,
+      qr/connect\(';host=localhost;/,
+      'Default DSN is h=localhost'
+   );
+}
+else {
+   # Apparently, your box is running MySQL on default ports.  That
+   # means the tool ran, so it should run without errors.
+   is(
+      $exit_status,
+      0,
+      'Default DSN is h=localhost'
+   );
+}
 
 # ############################################################################
 # DSN should inherit connection options (--port, etc.)

util/wait-to-exec

@@ -0,0 +1,40 @@
+#!/usr/bin/env perl
+
+use strict;
+use warnings FATAL => 'all';
+use Time::HiRes qw(sleep time);
+
+if ( !@ARGV || @ARGV < 2 ) {
+   print STDERR "Usage: wait-to-exec WAIT EXEC [RUNTIME [INTERVAL]]\n",
+                "When the WAIT script exits 0, execute the EXEC script.\n";
+   exit 1;
+}
+
+my ($wait, $exec, $runtime, $interval) = @ARGV;
+$runtime  ||= 1.0;
+$interval ||= 0.1;
+
+#warn "wait: $wait\n";
+#warn "exec: $exec\n";
+
+my $t_start = time;
+while ( time - $t_start < $runtime ) {
+   if ( exit_status($wait) ) {
+      sleep $interval;
+   }
+   else {
+      # exec does not return; this script becomes the exec process.
+      exec($exec);
+   }
+}
+
+warn "Never observed wait condition";
+exit 1;
+
+sub exit_status {
+   my ($wait) = @_;
+   system($wait);
+   # Like grep, exit status 0 means "match found", so stop waiting.
+   my $exit_status = $? >> 8;
+   return $exit_status;
+}