LU-13705 utils: improve llstat/llobdstat usability

[fs/lustre-release.git] / lustre / utils / llobdstat

diff --git a/lustre/utils/llobdstat b/lustre/utils/llobdstat
index 733cb4b..7a3a4ba 100755 (executable)
--- a/lustre/utils/llobdstat
+++ b/lustre/utils/llobdstat
@@ -1,95 +1,131 @@
 #!/usr/bin/perl
 #!/usr/bin/perl

-# llobdstat is a utility that parses obdfilter statistics files
-# found at proc/fs/lustre/<ostname>/stats.
-# It is mainly useful to watch the statistics change over time.
+# llobdstat is a utility that parses OST statistics files
+# found at e.g. obdfilter.<ostname>.stats or osc.<ostname>*.stats.
+# It is mainly useful to watch the bandwidth usage over time.

 
 my $pname = $0;
 
 my $pname = $0;

-
-my $defaultpath = "/proc/fs/lustre";
+my $obddev = "";

 my $obdstats = "stats";
 my $obdstats = "stats";

+my $statspath = "None";
+my $statsname = $obdstats;
+my $interval = 0;
+my $counter = 999999999;
+my $debug = 0;
+my $have_readkey = 1;
+my $width = 120;
+my $height = 25;

 
 sub usage()
 {
 
 sub usage()
 {

-    print STDERR "Usage: $pname <ost_name> [<interval> [<count>}]\n";
-    print STDERR "where  ost_name  : ost name under $defaultpath/obdfilter\n";
-    print STDERR "       interval  : sample interaval in seconds\n";
-    print STDERR "example: $pname lustre-OST0000 2\n";
-    print STDERR "Use CTRL + C to stop statistics printing\n";
-    exit 1;
+       print STDERR "Monitor read/write bandwidth of an OST device\n";
+       print STDERR "Usage: $pname [-h] [-i <interval>] [-v] <ost_name> [<interval> [<count>}]\n";
+       print STDERR "       stats_file : Lustre 'stats' file to watch\n";
+       print STDERR "       -d         : debug mode\n";
+       print STDERR "       -h         : help, display this information\n";
+       print STDERR "       -i interval: polling period in seconds\n";
+       print STDERR "       -n count   : number of samples printed\n";
+       print STDERR "example: $pname -i 5 lustre-OST0000\n";
+       print STDERR "Use CTRL + C to stop statistics printing\n";
+       exit 1;

 }
 
 }
 

-my $statspath = "None";
-my $interval = 0;
-my $counter = undef;
+# Command line parameter parsing
+use Getopt::Std;
+getopts('dhi:n:') or usage();
+usage() if $opt_h;
+$debug = $opt_d if $opt_d;
+$interval = $opt_i if $opt_i;
+$counter = $opt_n if $opt_n;
+
+my $i = 0;
+foreach (@ARGV) {
+       if ($i == 0) {
+               $obddev = $_;
+               $obddev =~ s/\./\//g;
+       } elsif ($i == 1) {
+               $interval = $_;
+       } elsif ($i == 2) {
+               $counter = $_;
+       } else {
+               print "ERROR: extra argument $_\n";
+               usage();
+       }
+       $i++;
+}
+if ( !$obddev ) {
+       print "ERROR: Need to specify stats_file\n";
+       usage();
+}

 
 

-if (($#ARGV < 0) || ($#ARGV > 2)) {
-    usage();
-} else {
-    if ( $ARGV[0] =~ /help$/ ) {
-        usage();
-    }
-    if ( -f $ARGV[0] ) {
-        $statspath = $ARGV[0];
-    } elsif ( -f "$ARGV[0]/$obdstats" ) {
-        $statspath = "$ARGV[0]/$obdstats";
-    } else {
-       my $st = "$defaultpath/obdfilter/$ARGV[0]/$obdstats";
+# Process arguments
+my $procpath = "/proc/fs/lustre";
+foreach my $param ( "$obddev", "$obddev*/$obdstats", "$obddev*/*/$obdstats",
+                   "*/$obddev*/$obdstats", "*/*/$obddev*/$obdstats" ) {
+       if ($debug) {
+               printf "trying $procpath/$param\n";
+       }
+       my $st = glob("$procpath/$param");

        if ( -f "$st" ) {
        if ( -f "$st" ) {

-           $statspath = $st;
+               $statspath = $st;
+               $statsname = `lctl list_param $param | head -n 1`;
+               if ($debug) {
+                       print "using $statspath\n"
+               }
+               last;
+       }
+}
+if ($statspath =~ /^None$/) {
+       # server stats are currently in /proc/fs/lustre, but may move eventually
+       $procpath = "/sys/kernel/debug/lustre";
+
+       foreach my $param ( "$obddev", "$obddev*/$obdstats", "$obddev*/*/$obdstats",
+                           "*/$obddev*/$obdstats", "*/*/$obddev*/$obdstats" ) {
+               if ($debug) {
+                       print "trying $procpath/$param\n";
+               }
+               $st = glob("$procpath/$param");
+               if ($debug) {
+                       print "glob $procpath/$param = $st\n";
+               }
+               if (-f "$st") {
+                       $statspath = $st;
+                       $statsname = `lctl list_param $param | head -n 1`;
+                       if ($debug) {
+                               print "using $statspath\n"
+                       }
+                       last;
+               }
+       }
+       if ($statspath =~ /^None$/) {
+               die "Cannot locate stat file for: $obddev\n";

        }
        }

-    }
-    if ( $statspath =~ /^None$/ ) {
-        die "Cannot locate stat file for: $ARGV[0]\n";
-    }
-    if ($#ARGV > 0) {
-        $interval = $ARGV[1];
-    }
-    if ($#ARGV > 1) {
-        $counter = $ARGV[2];
-    }

 }
 
 }
 

-print "$pname on $statspath\n";
+# check if Term::ReadKey is installed for efficient tty size, but OK if missing
+eval "require Term::ReadKey" or $have_readkey = 0;
+if ($debug) {
+       print "have_readkey=$have_readkey\n";
+}
+if ($have_readkey) {
+       eval "use Term::ReadKey";
+}
+
+print "$pname on $statsname\n";

 
 my %cur;
 my %last;
 
 my %cur;
 my %last;

-my $mhz = 0;

 
 

-# Removed some statstics like open, close that obdfilter doesn't contain.
+# Removed some statstics like open, close that the OST doesn't contain.

 # To add statistics parameters one needs to specify parameter names in the
 # below declarations in the same sequence.
 # To add statistics parameters one needs to specify parameter names in the
 # below declarations in the same sequence.

-my ($read_bytes, $write_bytes, $create, $destroy, $statfs, $punch,
-    $snapshot_time) = ("read_bytes", "write_bytes", "create", "destroy",
-                       "statfs", "punch", "snapshot_time");
+my ($read_bytes, $write_bytes, $create, $destroy, $statfs, $punch, $timestamp) =
+       ("read_bytes", "write_bytes", "create", "destroy", "statfs", "punch",
+        "snapshot_time");

 
 my @extinfo = ($create, $destroy, $statfs, $punch);
 my %shortname = ($create => "cx", $destroy => "dx", $statfs => "st", $punch => "pu");
 
 
 my @extinfo = ($create, $destroy, $statfs, $punch);
 my %shortname = ($create => "cx", $destroy => "dx", $statfs => "st", $punch => "pu");
 

-sub get_cpumhz()
-{
-    my $cpu_freq;
-    my $itc_freq; # On Itanium systems use this
-    if (open(CPUINFO, "/proc/cpuinfo")==0) {
-        return;
-    }
-    while (<CPUINFO>) {
-        if (/^cpu MHz\s+:\s*([\d\.]+)/) { $cpu_freq=$1; }
-        elsif (/^itc MHz\s+:\s*([\d\.]+)/) { $itc_freq=$1; }
-    }
-    if (defined($itc_freq)) {
-        $mhz = $itc_freq;
-    } elsif (defined($cpu_freq)) {
-        $mhz = $cpu_freq;
-    } else {
-        $mhz = 1;
-    }
-    close CPUINFO;
-}
-
-get_cpumhz();
-print "Processor counters run at $mhz MHz\n";
-
-# read statistics from obdfilter stats file.
+# read statistics from the stats file.

 # This subroutine gets called after every interval specified by user.
 sub readstat()
 {
 # This subroutine gets called after every interval specified by user.
 sub readstat()
 {


@@ -99,7 +135,6 @@ sub readstat()
     seek STATS, 0, 0;
     while (<STATS>) {
         chop;
     seek STATS, 0, 0;
     while (<STATS>) {
         chop;

-#        ($name, $cumulcount, $samples, $unit, $min, $max, $sum, $sumsquare)

         @iodata = split(/\s+/, $_);
         my $name = $iodata[0];
 
         @iodata = split(/\s+/, $_);
         my $name = $iodata[0];
 


@@ -107,46 +142,61 @@ sub readstat()
         if (defined($prevcount)) {
             $last{$name} = $prevcount;
         }
         if (defined($prevcount)) {
             $last{$name} = $prevcount;
         }

-        if ($name =~ /^read_bytes$/ || $name =~ /^write_bytes$/) {
-            $cur{$name} = $iodata[6];
-        }
-        elsif ($name =~ /^snapshot_time$/) {
-#            $cumulcount =~ /(\d+)/;
+        if ($name =~ /^$timestamp$/) {

             $cur{$name} = $iodata[1];
             $cur{$name} = $iodata[1];

+        } elsif ($name =~ /^$read_bytes$/) {
+           if (defined($cur{"read_ops"})) {
+               $last{"read_ops"} = $cur{"read_ops"};
+           }
+            $cur{"read_ops"} = $iodata[1];
+            $cur{$name} = $iodata[6];
+       } elsif ($name =~ /^$write_bytes$/) {
+           if (defined($cur{"write_ops"})) {
+               $last{"write_ops"} = $cur{"write_ops"};
+           }
+            $cur{"write_ops"} = $iodata[1];
+            $cur{$name} = $iodata[6];

         } else {
             $cur{$name} = $iodata[1];
         }
     }
 }
 
         } else {
             $cur{$name} = $iodata[1];
         }
     }
 }
 

-# process stats information read from obdfilter stats file.
+# process stats information read from the stats file.

 # This subroutine gets called after every interval specified by user.
 sub process_stats()
 {
     my $delta;
     my $data;
 # This subroutine gets called after every interval specified by user.
 sub process_stats()
 {
     my $delta;
     my $data;

-    my $last_time = $last{$snapshot_time};
+    my $last_time = $last{$timestamp};

     if (!defined($last_time)) {
     if (!defined($last_time)) {

-        printf "Read: %-g, Write: %-g, create/destroy: %-g/%-g, stat: %-g, punch: %-g\n",
-        $cur{$read_bytes}, $cur{$write_bytes},
-        $cur{$create}, $cur{$destroy},
-        $cur{$statfs}, $cur{$punch};
-        if ($interval) {
-            print "[NOTE: cx: create, dx: destroy, st: statfs, pu: punch ]\n\n";
-            print "Timestamp   Read-delta  ReadRate  Write-delta  WriteRate\n";
-            print "--------------------------------------------------------\n";
-        }
+        printf "Read: %.1f GiB, Write: %.1f GiB, cr: %lu dx: %lu, st: %lu, pu: %lu\n",
+            $cur{$read_bytes} / (1 << 30), $cur{$write_bytes} / (1 << 30),
+            $cur{$create}, $cur{$destroy}, $cur{$statfs}, $cur{$punch};
+       print "[NOTE: cx: create, dx: destroy, st: statfs, pu: punch]\n\n";

     } else {
     } else {

-        my $timespan = $cur{$snapshot_time} - $last{$snapshot_time};
-        my $rdelta = $cur{$read_bytes} - $last{$read_bytes};
-        my $rrate = ($rdelta) / ($timespan * ( 1 << 20 ));
-        my $wdelta = $cur{$write_bytes} - $last{$write_bytes};
-        my $wrate = ($wdelta) / ($timespan * ( 1 << 20 ));
-        $rdelta = ($rdelta) / (1024 * 1024);
-        $wdelta = ($wdelta) / (1024 * 1024);
+        my $timespan = $cur{$timestamp} - $last{$timestamp};
+        my $rtot = ($cur{$read_bytes} - $last{$read_bytes}) / (1 << 20);
+        my $riops = ($cur{"read_ops"} - $last{"read_ops"}) / $timespan;
+        my $rrate = $rtot / $timespan;
+        my $wtot = ($cur{$write_bytes} - $last{$write_bytes}) / (1 << 20);
+        my $wiops = ($cur{"write_ops"} - $last{"write_ops"}) / $timespan;
+        my $wrate = $wtot / $timespan;
+
+       # this is printed once per screen, like vmstat/iostat
+        if ($count++ % ($height - 2) == 0) {
+            print "Timestamp  Read-MiB RdMiB/s WriteMiB WrMiB/s RdIOPS WrIOPS\n";
+            print "---------- -------- ------- -------- ------- ------ ------\n";
+           if ($have_readkey) {
+               ($width, $height, $wpixels, $hpixels) = GetTerminalSize();
+           } else {
+               ($height, $width) = split / /, `stty size 2> /dev/null`;
+               #$width = 120 if ! $width
+           }
+        }

         # This print repeats after every interval.
         # This print repeats after every interval.

-        printf "%10lu  %6.2fMB  %6.2fMB/s   %6.2fMB  %6.2fMB/s",
-               $cur{$snapshot_time}, $rdelta, $rrate, $wdelta, $wrate;
+        printf "%10lu %8.1f %7.1f %8.1f %7.1f %6lu %6lu",
+               $cur{$timestamp}, $rtot, $rrate, $wtot, $wrate, $riops, $wiops;

 
         $delta = $cur{$getattr} - $last{$getattr};
         if ( $delta != 0 ) {
 
         $delta = $cur{$getattr} - $last{$getattr};
         if ( $delta != 0 ) {


@@ -165,7 +215,7 @@ sub process_stats()
     }
 }
 
     }
 }
 

-#Open the obdfilter stat file with STATS
+#Open the stat file with STATS

 open(STATS, $statspath) || die "Cannot open $statspath: $!\n";
 do {
     # read the statistics from stat file.
 open(STATS, $statspath) || die "Cannot open $statspath: $!\n";
 do {
     # read the statistics from stat file.


@@ -177,6 +227,5 @@ do {
     }
     # Repeat the statistics printing after every "interval" specified in
     # command line, up to counter times, if specified
     }
     # Repeat the statistics printing after every "interval" specified in
     # command line, up to counter times, if specified

-} while ($interval && (defined($counter) ? $counter-- > 0 : 1));
+} while ($interval && $counter-- > 0);

 close STATS;
 close STATS;

-# llobdfilter.pl ends here.