Test pt-stalk. Fix disk space+margin check. Fix checking --iterations. Add EXIT_REASON. Fix grepping status var.

bin/pt-stalk

@@ -646,6 +646,7 @@ lock_waits() {
 # ###########################################################################
 # Global variables
 # ###########################################################################
+EXIT_REASON=""
 TOOL=`basename $0`
 OKTORUN=1
 ITER=1
@@ -709,7 +710,7 @@ set_trg_func() {
 
 trg_status() {
    local var=$1
-   mysqladmin $EXT_ARGV extended-status | grep "$OPT_VARIABLE" | awk '{print $4}'
+   mysqladmin $EXT_ARGV extended-status | grep "$OPT_VARIABLE " | awk '{print $4}'
 }
 
 trg_processlist() {
@@ -729,10 +730,12 @@ trg_magic() {
 
 oktorun() {
    if [ $OKTORUN -eq 0 ]; then
+      EXIT_REASON="OKTORUN is false, exiting"
       return 1  # stop running
    fi
 
-   if [ -n "$OPT_ITERATIONS" ] && [ $ITER -ge $OPT_ITERATIONS ]; then
+   if [ -n "$OPT_ITERATIONS" ] && [ $ITER -gt $OPT_ITERATIONS ]; then
+      EXIT_REASON="No more iterations, exiting"      
       return 1  # stop running
    fi
 
@@ -808,15 +811,14 @@ stalk() {
          # is also checked every interval while collecting.
          local margin="20"  # default 20M margin, unless:
          if [ -n "$last_prefix" ]; then
-            margin=$(du -mc $d/$last_prefix-* | tail -n 1 | awk '{print $1'})
+            margin=$(du -mc $OPT_DEST/$last_prefix-* | tail -n 1 | awk '{print $1'})
          fi 
-         disk_space $d > $d/$p-disk-space  # Get real disk space.
+         disk_space $OPT_DEST > $OPT_DEST/$prefix-disk-space 
-         check_disk_space          \       # Then check it plus...
+         check_disk_space                \
-            $d/$p-disk-space       \
+            $OPT_DEST/$prefix-disk-space \
-            "$OPT_DISK_BYTE_LIMIT" \        
+            "$OPT_DISK_BYTE_LIMIT"       \
-            "$OPT_DISK_PCT_LIMIT"  \
+            "$OPT_DISK_PCT_LIMIT"        \
-            "$margin"              \       # ... the margin.
+            "$margin"  # real used MB + margin MB
-
          if [ $? -eq 0 ]; then
             # There should be enough disk space, so collect.
             log "$msg" >> "$OPT_DEST/$prefix-trigger"
@@ -893,6 +895,7 @@ main() {
    rm_tmpdir
    remove_pid_file "$OPT_PID"
 
+   log "$EXIT_REASON"
    log "$0 exit status $EXIT_STATUS"
    exit $EXIT_STATUS
 }

t/pt-stalk/pt-stalk.t

@@ -9,16 +9,162 @@ BEGIN {
 use strict;
 use warnings FATAL => 'all';
 use English qw(-no_match_vars);
-use Test::More tests => 1;
+use Test::More;
 
 use PerconaTest;
+use DSNParser;
+use Sandbox;
 
-TODO: {
+my $dp = new DSNParser(opts=>$dsn_opts);
-   local $TODO = "Test pt-stalk";
+my $sb = new Sandbox(basedir => '/tmp', DSNParser => $dp);
-   ok(1, 'ok');
+my $dbh = $sb->get_dbh_for('master');
-};
+
+if ( !$dbh ) {
+   plan skip_all => 'Cannot connect to sandbox master';
+}
+else {
+   plan tests => 14;
+}
+
+my $cnf      = "/tmp/12345/my.sandbox.cnf";
+my $pid_file = "/tmp/pt-stalk.pid.$PID";
+my $log_file = "/tmp/pt-stalk.log.$PID";
+my $dest     = "/tmp/pt-stalk.collect.$PID";
+my $pid;
+
+diag(`rm $pid_file 2>/dev/null`);
+diag(`rm $log_file 2>/dev/null`);
+diag(`rm -rf $dest 2>/dev/null`);
+
+# ###########################################################################
+# Test that it won't run if can't connect to MySQL.
+# ###########################################################################
+
+my $retval = system("$trunk/bin/pt-stalk >$log_file 2>&1");
+my $output = `cat $log_file`;
+
+like(
+   $output,
+   qr/Cannot connect to MySQL/,
+   "Cannot connect to MySQL"
+);
+
+is(
+   $retval >> 8,
+   1,
+   "Exit 1"
+);
+
+# ###########################################################################
+# Test that it runs and dies normally.
+# ###########################################################################
+diag(`rm $pid_file 2>/dev/null`);
+diag(`rm $log_file 2>/dev/null`);
+diag(`rm -rf $dest 2>/dev/null`);
+
+$retval = system("$trunk/bin/pt-stalk --pid $pid_file --log $log_file --dest $dest -- --defaults-file=$cnf");
+
+is(
+   $retval >> 8,
+   0,
+   "Parent exit 0"
+);
+
+PerconaTest::wait_for_files($pid_file, $log_file);
+ok(
+   -f $pid_file,
+   "Creates PID file"
+);
+
+ok(
+   -f $log_file,
+   "Creates log file"
+);
+
+sleep 1;
+
+ok(
+   -d $dest,
+   "Creates --dest (collect) dir"
+);
+
+chomp($pid = `cat $pid_file`);
+$retval = system("kill -0 $pid");
+is(
+   $retval >> 0,
+   0,
+   "pt-stalk is running ($pid)"
+);
+
+$output = `cat $log_file`;
+like(
+   $output,
+   qr/Check results: Threads_running=\d+, matched=no, cycles_true=0/,
+   "Check results logged"
+);
+
+$retval = system("kill $pid 2>/dev/null");
+is(
+   $retval >> 0,
+   0,
+   "Killed pt-stalk"
+);
+
+sleep 1;
+
+ok(
+   ! -f $pid_file,
+   "Removes PID file"
+);
+
+$output = `cat $log_file`;
+like(
+   $output,
+   qr/Caught signal, exiting/,
+   "Caught signal logged"
+);
+
+# ###########################################################################
+# Test collect.
+# ###########################################################################
+diag(`rm $pid_file 2>/dev/null`);
+diag(`rm $log_file 2>/dev/null`);
+diag(`rm $dest/*   2>/dev/null`);
+
+# We'll have to watch Uptime since it's the only status var that's going
+# to be predictable.
+my (undef, $uptime) = $dbh->selectrow_array("SHOW STATUS LIKE 'Uptime'");
+my $threshold = $uptime + 2;
+
+$retval = system("$trunk/bin/pt-stalk --no-daemonize --iterations 1 --dest $dest  --variable Uptime --threshold $threshold --cycles 2 --run-time 2 -- --defaults-file=$cnf >$log_file 2>&1");
+
+sleep 3;
+
+$output = `cat $dest/*-trigger`;
+like(
+   $output,
+   qr/Check results: Uptime=\d+, matched=yes, cycles_true=2/,
+   "Collect triggered"
+);
+
+chomp($output = `cat $dest/*-df | grep -c '^TS'`);
+is(
+   $output,
+   2,
+   "Collect ran for --run-time"
+);
+
+$output = `ps x | grep -v grep | grep 'pt-stalk pt-stalk --no-daemonize --iterations 1 --dest $dest'`;
+is(
+   $output,
+   "",
+   "pt-stalk is not running"
+);
 
 # #############################################################################
 # Done.
 # #############################################################################
+diag(`rm $pid_file 2>/dev/null`);
+diag(`rm $log_file 2>/dev/null`);
+diag(`rm -rf $dest 2>/dev/null`);
 exit;