package Common; # # Some common functions # # Revision: $Header: trunk/BNC/scripts/Common.pm 10099 2023-06-17 20:56:07Z wiese $ # use strict; use warnings; use Exporter; use Data::Dumper; use Log::Log4perl qw(:easy); use File::Temp; use File::Basename; use File::Spec::Functions; #$| = 1; # Buffer off (print) sub trim { my ($text) = @_; $text =~ s/^[\s]*//g; $text =~ s/[\s]*$//g; return $text; } # run a command in the shell. sub runCmd { my ($cmd) = shift; DEBUG("run cmd: \"$cmd\""); my $tmpfh = File::Temp->new( UNLINK => 1, SUFFIX => '.dat' ); my $tmpfile = $tmpfh->filename(); $cmd = "$cmd 2>$tmpfile |"; my $PIPE; unless ( open $PIPE, $cmd ) { ERROR("open $cmd | failed ($!)"); return ( -1, "", "" ); } my $stdout = join '', <$PIPE>; close $PIPE; my $exit_code = $?; my $stderr = getContent($tmpfile); return ( $exit_code, $stdout, $stderr ); } sub mkDir { my $dir = shift; return 1 if ( -d $dir ); eval { mkpath($dir) }; if ($@) { warn ("Could not create path [$dir]: $@"); return 0; } return 1; } #** @function amInteractiv () # @brief Checks if a program runs interactivly, or e.g. was invoked by cron. # @see Perl Cookbook Kap. 15.2 # @return true if interactive, else false. #* sub amInteractiv { use POSIX qw(getpgrp tcgetpgrp); my $tty; open ( $tty, "<", "/dev/tty" ) || LOGDIE "Could not open /dev/tty: $!"; my $tpgrp = tcgetpgrp( fileno ($tty) ); # get terminal foreground process group my $pgrp = getpgrp (); # get the current process group close $tty; return ( $tpgrp == $pgrp ); } #** @function printData ($data) # @brief Prints perl data structures. Uses Data::Dumper. # @param $data [required] Daten # @return - #* sub printData { my $data = shift; my $d = Data::Dumper->new($data); $d->Indent(3); # pretty print with array indices print $d->Dump; } sub getContent { my ($fileName) = @_; my $errmsg = ""; if ( -f $fileName ) { my $readMode = $/; undef $/; open ( FILE, "<", $fileName ) or ERROR("Kann Datei nicht zum lesen oeffnen"); $errmsg = ; close (FILE); $/ = $readMode; } return $errmsg; } #** @function filePosition ($file, $filePos) # @brief Save position of a file in an idxfile. # # Especially for logfiles. For performance reasons, logfiles should not parsed always from the # beginning, the next read access should start at the pos of the last read. # Saves the position of the last read in a help-file ".filePos" in # the directory of the logfile. If the param $filePos is set, then this position will be saved (Write). # Otherwise the current position will be returned (Read). # @param $file [required] File of interest # @param $filePos [optional] if given, this position ()in Bytes) will be saved, else return current position # @return file position in Bytes from beginning of the file #* sub filePosition { my ( $file, $filePos ) = @_; my ( $filName, $filDir ) = fileparse($file); my $idxFil = catfile( $filDir, ".filePos" ); if ($filePos) { # Save this position in idxfile my $newPos = "$filName $filePos"; if ( -e $idxFil ) { unless ( replace( $idxFil, "^${filName} \\d+", $newPos ) ) { # could not replace -> append blurt( "$newPos\n", $idxFil, { append => 1 } ); # my $INP; # if ( !open ( $INP, ">>", $idxFil ) ) { # ERROR("Could not append file [$idxFil]: $!"); # return 0; # } # print $INP "$newPos\n"; # close ($INP); } } else { # initial write blurt( "$newPos\n", $idxFil ); #writeTextInFile( $idxFil, "$newPos\n" ); } } else { # Read current position from idxfile # ---------------------------------- if ( !-e $idxFil ) { return 0; } my $line = ""; my $idx_fh; if ( !open ( $idx_fh, "<", $idxFil ) ) { ERROR("Could not open for read $idxFil: $!"); return 0; } while (<$idx_fh>) { if ( $_ =~ /^$filName (\d+)/ ) { # RegEx problematic with getContentFiltered() $line .= $_; $filePos = $1; last; } } close ($idx_fh); if ( $line && !$filePos ) { ERROR("Could not read file position in idxfile $idxFil. Weired line \"$line\""); $filePos = 0; } if ($filePos) { # If the file was meanwhile newly created, e.g. after restart from BNC my $fileSize = -s $file; if ( $filePos > $fileSize ) { WARN "file $file seams to be newly created"; DEBUG("size:$fileSize pos:$filePos"); $filePos = 0; } } else { WARN("No position entry found in idxfile for \"$filName\""); $filePos = 0; } } return $filePos; } #** @function isAlreadyRunning () # @brief Check if this program is already running. # @details Usage: isAlreadyRunning() && die "Same job is already running!\n";@n # Purpose: e.g. to avoid a program running twice # @see bernese gpsutil::check_job() # @return pid of already running job else false #* sub isAlreadyRunning { my ($scriptName) = fileparse($0); use Proc::ProcessTable; my $pt = new Proc::ProcessTable; my $pid_found = 0; # from already running job foreach my $p ( @{ $pt->table } ) { my $cmd = $p->cmndline; if ( $cmd =~ /$scriptName/ && $p->pid != $$ ) { next if ( $p->pid == getppid () ); # when started from cron! DEBUG( "Found: " . $p->pid . ", myself: $$" ); $pid_found = $p->pid; DEBUG "Process $cmd is running"; last; } } return $pid_found; } # toLineProtocol($rcd) # @brief converts $rcd to InfluxDB line protocol # @see https://docs.influxdata.com/influxdb/v2.1/reference/syntax/line-protocol/ sub toLineProtocol { my ($rcd) = @_; my $err = 0; my $str = $rcd->{'measurement'} . ','; # Tags (sort by key (in lexicographic order) for performance reasons) foreach ( sort { lc ($a) cmp lc ($b) } keys %{ $rcd->{'tags'} } ) { if ( !exists $rcd->{'tags'}->{$_} ) { $err++; #next; # ? } $str .= $_ . "=" . $rcd->{'tags'}->{$_} . ","; } chop $str; $str .= " "; # Fields foreach ( sort { lc ($a) cmp lc ($b) } keys %{ $rcd->{'fields'} } ) { if ( !exists $rcd->{'fields'}->{$_} ) { $err++; #next; # ? } $str .= $_ . "=" . $rcd->{'fields'}->{$_} . ","; } chop $str; $str .= " " . $rcd->{'time'}; # prec=s if ($err) { return 0; } return $str; } #** @function rms ($numbers) # @brief Compute the Root mean square. # @param $numbers [required] List with your numbers # @return RMS value # @note use PDL stats() if installed! # @see http://rosettacode.org/wiki/Averages/Root_mean_square#Perl #* sub rms { my $r = 0; $r += $_**2 for @_; return sqrt ( $r / @_ ); } #** @function avg ($numbers) # @brief Compute the arithmetic mean value. # @param $numbers [required] List with your numbers # @note use PDL stats() if installed! # @return mean value # @see http://rosettacode.org/wiki/Average/Arithmetic_mean#Perl #* sub avg { @_ or return 0; my $sum = 0; $sum += $_ foreach @_; return $sum / @_; } #** @function sigma ($numbers) # @brief Compute the STDDEV of the arithmetic mean. # @param $numbers [required] List with your numbers # @return Sigma value # @note use use PDL::Primitive qw(stats) if installed! #* sub sigma { my $n = scalar (@_); if ( $n == 1 ) { WARN "Could not compute sigma for only one sample"; return; } my $e = 0; # arithm. Mittel $e += $_ for @_; $e = $e / $n; my @v = map { $e - $_ } @_; my $vv = 0; $vv += $_**2 for @v; return sqrt ( $vv / ( $n * ( $n - 1 ) ) ); } #** @function median (@numbers) # @brief Compute the median of a list of numbers. # @param @numbers [required] List with your numbers # @see https://en.wikipedia.org/wiki/Median # @note use PDL stats() if installed! # @return median #* sub median { my $mid = int ( @_ / 2 ); my @sorted = sort { $a <=> $b } (@_); if ( @sorted % 2 ) { return $sorted[$mid]; } else { return ( $sorted[ $mid - 1 ] + $sorted[$mid] ) / 2; } return; # undef } 1; # End of Common