summaryrefslogtreecommitdiffstats
path: root/contrib/slapd-tools/statslog
diff options
context:
space:
mode:
Diffstat (limited to 'contrib/slapd-tools/statslog')
-rwxr-xr-xcontrib/slapd-tools/statslog171
1 files changed, 171 insertions, 0 deletions
diff --git a/contrib/slapd-tools/statslog b/contrib/slapd-tools/statslog
new file mode 100755
index 0000000..d809c58
--- /dev/null
+++ b/contrib/slapd-tools/statslog
@@ -0,0 +1,171 @@
+#!/usr/bin/perl -w
+# statslog - Rearrange and output selected parts of slapd's statslog output.
+# $OpenLDAP$
+# This work is part of OpenLDAP Software <http://www.openldap.org/>.
+#
+# Copyright 1998-2018 The OpenLDAP Foundation.
+# Portions Copyright 2004 Hallvard B. Furuseth.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted only as authorized by the OpenLDAP
+# Public License.
+#
+# A copy of this license is available in the file LICENSE in the
+# top-level directory of the distribution or, alternatively, at
+# <http://www.OpenLDAP.org/license.html>.
+
+sub usage {
+ die join("", @_, <<'EOM');
+Usage: statslog [options] [logfiles; may be .gz or .bz2 files]
+
+ Output selected parts of slapd's statslog output (LDAP request/response
+ log to syslog or stderr; loglevel 256), grouping log lines by LDAP
+ connection. Lines with no connection are excluded by default.
+
+Options:
+ --brief -b Brief output (omit time, host/process name/ID).
+ --exclude=RE -e RE Exclude connections whose output matches REgexp.
+ --include=RE -i RE Only include connections matching REgexp.
+ --EXCLUDE=RE -E RE Case-sensitive '--exclude'.
+ --INCLUDE=RE -I RE Case-sensitive '--include'.
+ --loose -l Include "loose" lines (lines with no connection).
+ --no-loose -L RE Only exclude the "loose" lines that match RE.
+ --join -j Join the inputs as if they were one big log file.
+ Each file must start where the previous left off.
+ --no-join -J Do not --join. (Can be useful with --sort.)
+ --sort -s Sort input files by age. Implies --join.
+ --trace -t Print file names when read. Implies --no-join.
+All --exclude/include options are applied. Note: --exclude/include are
+unreliable without --join/sort for connections spanning several log files.
+EOM
+}
+
+########################################################################
+
+use bytes;
+use strict;
+use Getopt::Long;
+
+# Globals
+my %conns; # Hash (connection number -> output)
+my @loose; # Collected output with no connection number
+
+# Command line options
+my($brief, @filters, @conditions, $no_loose);
+my($join_files, $sort_files, $trace, $getopt_ok);
+
+# Handle --include/INCLUDE/exclude/EXCLUDE options
+sub filter_opt {
+ my($opt, $regexp) = @_;
+ push(@conditions, sprintf('$lines %s /$filters[%d]/om%s',
+ (lc($opt) eq 'include' ? "=~" : "!~"),
+ scalar(@filters),
+ ($opt eq lc($opt) ? "i" : "")));
+ push(@filters, $regexp);
+}
+
+# Parse options at compile time so some can become constants to optimize away
+BEGIN {
+ &Getopt::Long::Configure(qw(bundling no_ignore_case));
+ $getopt_ok = GetOptions("brief|b" => \$brief,
+ "include|i=s" => \&filter_opt,
+ "exclude|e=s" => \&filter_opt,
+ "INCLUDE|I=s" => \&filter_opt,
+ "EXCLUDE|E=s" => \&filter_opt,
+ "join|j" => \$join_files,
+ "no-join|J" => sub { $join_files = 0; },
+ "sort|s" => \$sort_files,
+ "loose|l" => sub { $no_loose = ".^"; },
+ "no-loose|L=s" => \$no_loose,
+ "trace|t" => \$trace);
+}
+usage() unless $getopt_ok;
+usage("--trace is incompatible with --join.\n") if $trace && $join_files;
+
+$join_files = 1 if !defined($join_files) && $sort_files && !$trace;
+use constant BRIEF => !!$brief;
+use constant LOOSE => defined($no_loose) && ($no_loose eq ".^" ? 2 : 1);
+
+# Build sub out(header, connection number) to output one connection's data
+my $out_body = (LOOSE
+ ? ' if (@loose) { print "\n", @loose; @loose = (); } '
+ : '');
+$out_body .= ' print "\n", $_[0], $lines; ';
+$out_body = " if (" . join("\n && ", @conditions) . ") {\n$out_body\n}"
+ if @conditions;
+eval <<EOM;
+sub out {
+ my \$lines = delete(\$conns{\$_[1]});
+ $out_body
+}
+1;
+EOM
+die $@ if $@;
+
+# Read and output log lines from one file
+sub do_file {
+ local(@ARGV) = @_;
+ my($conn, $line, $act);
+ while (<>) {
+ if (BRIEF
+ ? (($conn, $line, $act) = /\bconn=(\d+) (\S+ (\S+).*\n)/)
+ : (($conn, $act) = /\bconn=(\d+) \S+ (\S+)/ )) {
+ $conns{$conn} .= (BRIEF ? $line : $_);
+ out("", $conn) if $act eq 'closed';
+ } elsif (LOOSE && (LOOSE > 1 || !/$no_loose/omi)) {
+ s/^\w{3} [ \d]+:\d\d:\d\d [^:]*: // if BRIEF;
+ push(@loose, $_);
+ }
+ }
+ final() unless $join_files;
+}
+
+# Output log lines for unfinished connections
+sub final {
+ if (%conns) {
+ for my $conn (sort keys %conns) {
+ out("UNFINISHED:\n", $conn);
+ }
+ die if %conns;
+ }
+ if (LOOSE && @loose) { print "\n", @loose; @loose = (); }
+}
+
+# Main program
+if (!@ARGV) {
+ # Read from stdin
+ do_file();
+} else {
+ if ($sort_files && @ARGV > 1) {
+ # Sort files by last modified time; oldest first
+ my @fileinfo;
+ for my $file (@ARGV) {
+ my $age = -M $file;
+ if (defined($age)) {
+ push(@fileinfo, [$age, $file]);
+ } else {
+ print STDERR "File not found: $file\n";
+ }
+ }
+ exit(1) unless @fileinfo;
+ @ARGV = map { $_->[1] } sort { $b->[0] <=> $a->[0] } @fileinfo;
+ }
+
+ # Prepare to pipe .gz, .bz2 and .bz files through gunzip or bunzip2
+ my %type2prog = ("gz" => "gunzip", "bz2" => "bunzip2", "bz" => "bunzip2");
+ for (@ARGV) {
+ if (/\.(gz|bz2?)$/) {
+ my $type = $1;
+ die "Bad filename: $_\n" if /^[+-]|[^\w\/.,:%=+-]|^$/;
+ $_ = "$type2prog{$type} -c $_ |";
+ }
+ }
+
+ # Process the files
+ for my $file (@ARGV) {
+ print "\n$file:\n" if $trace;
+ do_file($file);
+ }
+}
+final();