28a84cd465
git-svn-id: https://svn.code.sf.net/p/xcat/code/xcat-core/trunk@2093 8638fb3e-16cb-4fca-ae20-7b5d299a9bcd
1106 lines
38 KiB
Perl
1106 lines
38 KiB
Perl
# IBM(c) 2007 EPL license http://www.eclipse.org/legal/epl-v10.html
|
|
#####################################################
|
|
#
|
|
# xCAT plugin package to handle various commands that work with the
|
|
# xCAT tables
|
|
#
|
|
#####################################################
|
|
package xCAT_plugin::tabutils;
|
|
use strict;
|
|
use warnings;
|
|
use xCAT::Table;
|
|
use xCAT::Schema;
|
|
use Data::Dumper;
|
|
use xCAT::NodeRange;
|
|
use xCAT::Schema;
|
|
use xCAT::Utils;
|
|
use Getopt::Long;
|
|
|
|
1;
|
|
|
|
#some quick aliases to table/value
|
|
my %shortnames = (
|
|
groups => [qw(nodelist groups)],
|
|
tags => [qw(nodelist groups)],
|
|
mgt => [qw(nodehm mgt)],
|
|
#switch => [qw(switch switch)],
|
|
);
|
|
|
|
#####################################################
|
|
# Return list of commands handled by this plugin
|
|
#####################################################
|
|
sub handled_commands
|
|
{
|
|
return {
|
|
gettab => "tabutils",
|
|
tabdump => "tabutils",
|
|
tabrestore => "tabutils",
|
|
tabch => "tabutils", # not implemented yet
|
|
nodech => "tabutils",
|
|
nodeadd => "tabutils",
|
|
noderm => "tabutils",
|
|
tabls => "tabutils", # not implemented yet
|
|
nodels => "tabutils",
|
|
getnodecfg => "tabutils", # not implemented yet (?? this doesn't seem much different from gettab)
|
|
addattr => "tabutils", # not implemented yet
|
|
delattr => "tabutils", # not implemented yet
|
|
chtype => "tabutils", # not implemented yet
|
|
nr => "tabutils", # not implemented yet
|
|
tabgrep => "tabutils"
|
|
};
|
|
}
|
|
|
|
# Each cmd now returns its own usage inside its function
|
|
#my %usage = (
|
|
#nodech => "Usage: nodech <noderange> [table.column=value] [table.column=value] ...",
|
|
#nodeadd => "Usage: nodeadd <noderange> [table.column=value] [table.column=value] ...",
|
|
#noderm => "Usage: noderm <noderange>",
|
|
# the usage for tabdump is in the tabdump function
|
|
#tabdump => "Usage: tabdump <tablename>\n where <tablename> is one of the following:\n " . join("\n ", keys %xCAT::Schema::tabspec),
|
|
# the usage for tabrestore is in the tabrestore client cmd
|
|
#tabrestore => "Usage: tabrestore <tablename>.csv",
|
|
#);
|
|
|
|
#####################################################
|
|
# Process the command
|
|
#####################################################
|
|
sub process_request
|
|
{
|
|
#use Getopt::Long;
|
|
Getopt::Long::Configure("bundling");
|
|
#Getopt::Long::Configure("pass_through");
|
|
Getopt::Long::Configure("no_pass_through");
|
|
|
|
my $request = shift;
|
|
my $callback = shift;
|
|
my $nodes = $request->{node};
|
|
my $command = $request->{command}->[0];
|
|
my $args = $request->{arg};
|
|
#unless ($args or $nodes or $request->{data})
|
|
#{
|
|
#if ($usage{$command})
|
|
#{
|
|
#$callback->({data => [$usage{$command}]});
|
|
#return;
|
|
#}
|
|
#}
|
|
|
|
if ($command eq "nodels")
|
|
{
|
|
return nodels($nodes, $args, $callback, $request->{emptynoderange}->[0]);
|
|
}
|
|
elsif ($command eq "noderm" or $command eq "rmnode")
|
|
{
|
|
return noderm($nodes, $args, $callback);
|
|
}
|
|
elsif ($command eq "nodeadd" or $command eq "addnode")
|
|
{
|
|
return nodech($nodes, $args, $callback, 1);
|
|
}
|
|
elsif ($command eq "nodech" or $command eq "nodech")
|
|
{
|
|
return nodech($nodes, $args, $callback, 0);
|
|
}
|
|
elsif ($command eq "tabrestore")
|
|
{
|
|
return tabrestore($request, $callback);
|
|
}
|
|
elsif ($command eq "tabdump")
|
|
{
|
|
return tabdump($args, $callback);
|
|
}
|
|
elsif ($command eq "gettab")
|
|
{
|
|
return gettab($request, $callback);
|
|
}
|
|
elsif ($command eq "tabgrep")
|
|
{
|
|
return tabgrep($nodes, $callback);
|
|
}
|
|
else
|
|
{
|
|
print "$command not implemented yet\n";
|
|
return (1, "$command not written yet");
|
|
}
|
|
|
|
}
|
|
|
|
# Display particular attributes, using query strings.
|
|
sub gettab
|
|
{
|
|
my $req = shift;
|
|
my $callback = shift;
|
|
my $HELP;
|
|
my $NOTERSE;
|
|
|
|
my $gettab_usage = sub {
|
|
my $exitcode = shift @_;
|
|
my %rsp;
|
|
push @{$rsp{data}}, "Usage: gettab [-H|--with-fieldname] key=value,... table.attribute ...";
|
|
push @{$rsp{data}}, " gettab [-?|-h|--help]";
|
|
if ($exitcode) { $rsp{errorcode} = $exitcode; }
|
|
$callback->(\%rsp);
|
|
};
|
|
|
|
# Process arguments
|
|
@ARGV = @{$req->{arg}};
|
|
if (!GetOptions('h|?|help' => \$HELP,'H|with-fieldname' => \$NOTERSE)) { $gettab_usage->(1); return; }
|
|
|
|
if ($HELP) { $gettab_usage->(0); return; }
|
|
if (scalar(@ARGV)<2) { $gettab_usage->(1); return; }
|
|
|
|
# Get all the key/value pairs into a hash
|
|
my $keyspec = shift @ARGV;
|
|
my @keypairs = split /,/, $keyspec;
|
|
my %keyhash;
|
|
foreach (@keypairs)
|
|
{
|
|
(my $key, my $value) = split /=/, $_;
|
|
unless (defined $key) {
|
|
$gettab_usage->(1);
|
|
return;
|
|
}
|
|
$keyhash{$key} = $value;
|
|
}
|
|
|
|
# Group the columns asked for by table (so we can do 1 query per table)
|
|
my %tabhash;
|
|
my $terse = 2;
|
|
if ($NOTERSE) {
|
|
$terse = 0;
|
|
}
|
|
foreach my $tabvalue (@ARGV)
|
|
{
|
|
$terse--;
|
|
(my $table, my $column) = split /\./, $tabvalue;
|
|
$tabhash{$table}->{$column} = 1;
|
|
}
|
|
|
|
#Sanity check the key against all tables in question
|
|
foreach my $tabn (keys %tabhash) {
|
|
foreach my $kcheck (keys %keyhash) {
|
|
unless (grep /^$kcheck$/, @{$xCAT::Schema::tabspec{$tabn}->{cols}}) {
|
|
$callback->({error => ["Unkown key $kcheck to $tabn"],errorcode=>[1]});
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
# Get the requested columns from each table
|
|
foreach my $tabn (keys %tabhash)
|
|
{
|
|
my $tab = xCAT::Table->new($tabn);
|
|
(my $ent) = $tab->getAttribs(\%keyhash, keys %{$tabhash{$tabn}});
|
|
foreach my $coln (keys %{$tabhash{$tabn}})
|
|
{
|
|
if ($terse > 0) {
|
|
$callback->({data => ["" . $ent->{$coln}]});
|
|
} else {
|
|
$callback->({data => ["$tabn.$coln: " . $ent->{$coln}]});
|
|
}
|
|
}
|
|
$tab->close;
|
|
}
|
|
}
|
|
|
|
sub noderm
|
|
{
|
|
my $nodes = shift;
|
|
my $args = shift;
|
|
my $cb = shift;
|
|
my $VERSION;
|
|
my $HELP;
|
|
|
|
my $noderm_usage = sub {
|
|
my $exitcode = shift @_;
|
|
my %rsp;
|
|
push @{$rsp{data}}, "Usage:";
|
|
push @{$rsp{data}}, " noderm noderange";
|
|
push @{$rsp{data}}, " noderm {-v|--version}";
|
|
push @{$rsp{data}}, " noderm [-?|-h|--help]";
|
|
if ($exitcode) { $rsp{errorcode} = $exitcode; }
|
|
$cb->(\%rsp);
|
|
};
|
|
|
|
if ($args) {
|
|
@ARGV = @{$args};
|
|
}
|
|
if (!GetOptions('h|?|help' => \$HELP, 'v|version' => \$VERSION) ) { $noderm_usage->(1); return; }
|
|
|
|
if ($HELP) { $noderm_usage->(0); return; }
|
|
|
|
if ($VERSION) {
|
|
my %rsp;
|
|
my $version = xCAT::Utils->Version();
|
|
$rsp{data}->[0] = "$version";
|
|
$cb->(\%rsp);
|
|
return;
|
|
}
|
|
|
|
if (!$nodes) { $noderm_usage->(1); return; }
|
|
|
|
# Build the argument list for using the -d option of nodech to do our work for us
|
|
my @tablist = ("-d");
|
|
foreach (keys %{xCAT::Schema::tabspec})
|
|
{
|
|
if (grep /^node$/, @{$xCAT::Schema::tabspec{$_}->{cols}})
|
|
{
|
|
push @tablist, $_;
|
|
}
|
|
}
|
|
nodech($nodes, \@tablist, $cb, 0);
|
|
}
|
|
|
|
sub tabrestore
|
|
{
|
|
# the usage for tabrestore is in the tabrestore client cmd
|
|
|
|
#request->{data} is an array of CSV formatted lines
|
|
my $request = shift;
|
|
my $cb = shift;
|
|
my $table = $request->{table}->[0];
|
|
my $linenumber = 1;
|
|
my $tab = xCAT::Table->new($table, -create => 1, -autocommit => 0);
|
|
unless ($tab) {
|
|
$cb->({error => "Unable to open $table",errorcode=>4});
|
|
return;
|
|
}
|
|
$tab->delEntries(); #Yes, delete *all* entries
|
|
my $header = shift @{$request->{data}};
|
|
$header =~ s/"//g; #Strip " from overzealous CSV apps
|
|
$header =~ s/^#//;
|
|
$header =~ s/\s+$//;
|
|
my @colns = split(/,/, $header);
|
|
my $line;
|
|
my $rollback = 0;
|
|
LINE: foreach $line (@{$request->{data}})
|
|
{
|
|
$linenumber++;
|
|
$line =~ s/\s+$//;
|
|
my $origline = $line; #save for error reporting
|
|
my %record;
|
|
my $col;
|
|
foreach $col (@colns)
|
|
{
|
|
if ($line =~ /^,/ or $line eq "")
|
|
{ #Match empty, or end of line that is empty
|
|
#TODO: should we detect when there weren't enough CSV fields on a line to match colums?
|
|
$record{$col} = undef;
|
|
$line =~ s/^,//;
|
|
}
|
|
elsif ($line =~ /^[^,]*"/)
|
|
{ # We have stuff in quotes... pain...
|
|
#I don't know what I'm doing, so I'll do it a hard way....
|
|
if ($line !~ /^"/)
|
|
{
|
|
$rollback = 1;
|
|
$cb->(
|
|
{
|
|
error =>
|
|
"CSV missing opening \" for record with \" characters on line $linenumber, character "
|
|
. index($origline, $line) . ": $origline", errorcode=>4
|
|
}
|
|
);
|
|
next LINE;
|
|
}
|
|
my $offset = 1;
|
|
my $nextchar;
|
|
my $ent;
|
|
while (not defined $ent)
|
|
{
|
|
$offset = index($line, '"', $offset);
|
|
$offset++;
|
|
if ($offset <= 0)
|
|
{
|
|
|
|
#MALFORMED CSV, request rollback, report an error
|
|
$rollback = 1;
|
|
$cb->(
|
|
{
|
|
error =>
|
|
"CSV unmatched \" in record on line $linenumber, character "
|
|
. index($origline, $line) . ": $origline", errorcode=>4
|
|
}
|
|
);
|
|
next LINE;
|
|
}
|
|
$nextchar = substr($line, $offset, 1);
|
|
if ($nextchar eq '"')
|
|
{
|
|
$offset++;
|
|
}
|
|
elsif ($offset eq length($line) or $nextchar eq ',')
|
|
{
|
|
$ent = substr($line, 0, $offset, '');
|
|
$line =~ s/^,//;
|
|
chop $ent;
|
|
$ent = substr($ent, 1);
|
|
$ent =~ s/""/"/g;
|
|
$record{$col} = $ent;
|
|
}
|
|
else
|
|
{
|
|
$cb->(
|
|
{
|
|
error =>
|
|
"CSV unescaped \" in record on line $linenumber, character "
|
|
. index($origline, $line) . ": $origline", errorcode=>4
|
|
}
|
|
);
|
|
$rollback = 1;
|
|
next LINE;
|
|
}
|
|
}
|
|
}
|
|
elsif ($line =~ /^([^,]+)/)
|
|
{ #easiest case, no Text::Balanced needed..
|
|
$record{$col} = $1;
|
|
$line =~ s/^([^,]+)(,|$)//;
|
|
}
|
|
}
|
|
if ($line)
|
|
{
|
|
$rollback = 1;
|
|
$cb->({error => "Too many fields on line $linenumber: $origline | $line", errorcode=>4});
|
|
next LINE;
|
|
}
|
|
|
|
#TODO: check for error from DB and rollback
|
|
my @rc = $tab->setAttribs(\%record, \%record);
|
|
if (not defined($rc[0]))
|
|
{
|
|
$rollback = 1;
|
|
$cb->({error => "DB error " . $rc[1] . " with line $linenumber: " . $origline, errorcode=>4});
|
|
}
|
|
}
|
|
if ($rollback)
|
|
{
|
|
$tab->rollback();
|
|
$tab->close;
|
|
undef $tab;
|
|
return;
|
|
}
|
|
else
|
|
{
|
|
$tab->commit; #Made it all the way here, commit
|
|
}
|
|
}
|
|
|
|
# Display a list of tables, or a specific table in CSV format
|
|
sub tabdump
|
|
{
|
|
my $args = shift;
|
|
my $cb = shift;
|
|
my $table = "";
|
|
my $HELP;
|
|
my $DESC;
|
|
|
|
my $tabdump_usage = sub {
|
|
my $exitcode = shift @_;
|
|
my %rsp;
|
|
push @{$rsp{data}}, "Usage: tabdump [-d] [table]";
|
|
push @{$rsp{data}}, " tabdump [-?|-h|--help]";
|
|
if ($exitcode) { $rsp{errorcode} = $exitcode; }
|
|
$cb->(\%rsp);
|
|
};
|
|
|
|
# Process arguments
|
|
if ($args) {
|
|
@ARGV = @{$args};
|
|
}
|
|
if (!GetOptions('h|?|help' => \$HELP, 'd' => \$DESC)) { $tabdump_usage->(1); return; }
|
|
|
|
if ($HELP) { $tabdump_usage->(0); return; }
|
|
if (scalar(@ARGV)>1) { $tabdump_usage->(1); return; }
|
|
|
|
my %rsp;
|
|
# If no arguments given, we display a list of the tables
|
|
if (!scalar(@ARGV)) {
|
|
if ($DESC) { # display the description of each table
|
|
my $tab = xCAT::Table->getDescriptions();
|
|
foreach my $key (keys %$tab) {
|
|
my $space = (length($key)<7 ? "\t\t" : "\t");
|
|
push @{$rsp{data}}, "$key:$space".$tab->{$key}."\n";
|
|
}
|
|
}
|
|
else { push @{$rsp{data}}, xCAT::Table->getTableList(); } # if no descriptions, just display the list of table names
|
|
@{$rsp{data}} = sort @{$rsp{data}};
|
|
if ($DESC && scalar(@{$rsp{data}})) { chop($rsp{data}->[scalar(@{$rsp{data}})-1]); } # remove the final newline
|
|
$cb->(\%rsp);
|
|
return;
|
|
}
|
|
|
|
$table = $ARGV[0];
|
|
if ($DESC) { # only show the attribute descriptions, not the values
|
|
my $schema = xCAT::Table->getTableSchema($table);
|
|
if (!$schema) { $cb->({error => "table $table does not exist.",errorcode=>1}); return; }
|
|
my $desc = $schema->{descriptions};
|
|
foreach my $c (@{$schema->{cols}}) {
|
|
my $space = (length($c)<7 ? "\t\t" : "\t");
|
|
push @{$rsp{data}}, "$c:$space".$desc->{$c}."\n";
|
|
}
|
|
if (scalar(@{$rsp{data}})) { chop($rsp{data}->[scalar(@{$rsp{data}})-1]); } # remove the final newline
|
|
$cb->(\%rsp);
|
|
return;
|
|
}
|
|
|
|
|
|
my $tabh = xCAT::Table->new($table);
|
|
|
|
my $tabdump_header = sub {
|
|
my $header = "#" . join(",", @_);
|
|
push @{$rsp{data}}, $header;
|
|
};
|
|
|
|
# If the table does not exist yet (because its never been written to),
|
|
# at least show the header (the column names)
|
|
unless ($tabh)
|
|
{
|
|
if (defined($xCAT::Schema::tabspec{$table}))
|
|
{
|
|
$tabdump_header->(@{$xCAT::Schema::tabspec{$table}->{cols}});
|
|
$cb->(\%rsp);
|
|
return;
|
|
}
|
|
$cb->({error => "No such table: $table",errorcode=>1});
|
|
return 1;
|
|
}
|
|
|
|
my $recs = $tabh->getAllEntries();
|
|
my $rec;
|
|
unless (@$recs) # table exists, but is empty. Show header.
|
|
{
|
|
if (defined($xCAT::Schema::tabspec{$table}))
|
|
{
|
|
$tabdump_header->(@{$xCAT::Schema::tabspec{$table}->{cols}});
|
|
$cb->(\%rsp);
|
|
return;
|
|
}
|
|
}
|
|
|
|
# Display all the rows of the table in the order of the columns in the schema
|
|
$tabdump_header->(@{$tabh->{colnames}});
|
|
foreach $rec (@$recs)
|
|
{
|
|
my $line = '';
|
|
foreach (@{$tabh->{colnames}})
|
|
{
|
|
if (defined $rec->{$_})
|
|
{
|
|
$rec->{$_} =~ s/"/""/g;
|
|
$line = $line . '"' . $rec->{$_} . '",';
|
|
}
|
|
else
|
|
{
|
|
$line .= ',';
|
|
}
|
|
}
|
|
$line =~ s/,$//; # remove the extra comma at the end
|
|
push @{$rsp{data}}, $line;
|
|
}
|
|
$cb->(\%rsp);
|
|
}
|
|
|
|
sub getTableColumn {
|
|
my $string = shift;
|
|
if ($shortnames{$string}) {
|
|
return @{$shortnames{$string}};
|
|
}
|
|
unless ($string =~ /\./) {
|
|
return undef;
|
|
}
|
|
return split /\./,$string,2;
|
|
}
|
|
|
|
sub nodech
|
|
{
|
|
my $nodes = shift;
|
|
my $args = shift;
|
|
my $callback = shift;
|
|
my $addmode = shift;
|
|
my $VERSION;
|
|
my $HELP;
|
|
my $deletemode;
|
|
|
|
my $nodech_usage = sub
|
|
{
|
|
my $exitcode = shift @_;
|
|
my $addmode = shift @_;
|
|
my $cmdname = $addmode ? 'nodeadd' : 'nodech';
|
|
my %rsp;
|
|
if ($addmode) {
|
|
push @{$rsp{data}}, "Usage: $cmdname <noderange> groups=<groupnames> [table.column=value] [...]";
|
|
} else {
|
|
push @{$rsp{data}}, "Usage: $cmdname <noderange> table.column=value [...]";
|
|
push @{$rsp{data}}, " $cmdname {-d | --delete} <noderange> <table> [...]";
|
|
}
|
|
push @{$rsp{data}}, " $cmdname {-v | --version}";
|
|
push @{$rsp{data}}, " $cmdname [-? | -h | --help]";
|
|
if ($exitcode) { $rsp{errorcode} = $exitcode; }
|
|
$callback->(\%rsp);
|
|
};
|
|
|
|
@ARGV = @{$args};
|
|
my %options = ('h|?|help' => \$HELP, 'v|version' => \$VERSION);
|
|
if (!$addmode) { $options{'d|delete'} = \$deletemode; }
|
|
if (!GetOptions(%options)) {
|
|
$nodech_usage->(1, $addmode);
|
|
return;
|
|
}
|
|
|
|
# Help
|
|
if ($HELP) {
|
|
$nodech_usage->(0, $addmode);
|
|
return;
|
|
}
|
|
|
|
# Version
|
|
if ($VERSION) {
|
|
my %rsp;
|
|
my $version = xCAT::Utils->Version();
|
|
$rsp{data}->[0] = "$version";
|
|
$callback->(\%rsp);
|
|
return;
|
|
}
|
|
|
|
# Note: the noderange comes through in $arg (and therefore @ARGV) for nodeadd,
|
|
# because it is linked to xcatclientnnr, since the nodes specified in the noderange
|
|
# do not exist yet. The nodech cmd is linked to xcatclient, so its noderange is
|
|
# put in $nodes instead of $args.
|
|
if (scalar(@ARGV) < (1+$addmode)) { $nodech_usage->(1, $addmode); return; }
|
|
|
|
if ($addmode)
|
|
{
|
|
my $nr = shift @ARGV;
|
|
$nodes = [noderange($nr, 0)];
|
|
unless ($nodes) {
|
|
$callback->({error => "No noderange to add.\n",errorcode=>1});
|
|
return;
|
|
}
|
|
}
|
|
my $column;
|
|
my $value;
|
|
my $temp;
|
|
my %tables;
|
|
my %criteria=();
|
|
my $tab;
|
|
|
|
#print Dumper($deletemode);
|
|
foreach (@ARGV)
|
|
{
|
|
if ($deletemode)
|
|
{
|
|
if (m/[=\.]/) # in delete mode they can only specify tables names
|
|
{
|
|
$callback->({error => [". and = not valid in delete mode."],errorcode=>1});
|
|
next;
|
|
}
|
|
$tables{$_} = 1;
|
|
next;
|
|
}
|
|
unless (m/=/ or m/!~/)
|
|
{
|
|
$callback->({error => ["Malformed argument $_ ignored."],errorcode=>1});
|
|
next;
|
|
}
|
|
my $stable;
|
|
my $scolumn;
|
|
#Check for selection criteria
|
|
if (m/^[^=]*==/) {
|
|
($temp,$value)=split /==/,$_,2;
|
|
($stable,$scolumn)=getTableColumn($temp);
|
|
$criteria{$stable}->{$scolumn}=[$value,'match'];
|
|
|
|
next; #Is a selection criteria, not an assignment specification
|
|
} elsif (m/^[^=]*!=/) {
|
|
($temp,$value)=split /!=/,$_,2;
|
|
($stable,$scolumn)=getTableColumn($temp);
|
|
$criteria{$stable}->{$scolumn}=[$value,'natch'];
|
|
next; #Is a selection criteria, not an assignment specification
|
|
} elsif (m/^[^=]*=~/) {
|
|
($temp,$value)=split /=~/,$_,2;
|
|
($stable,$scolumn)=getTableColumn($temp);
|
|
$value =~ s/^\///;
|
|
$value =~ s/\/$//;
|
|
$criteria{$stable}->{$scolumn}=[$value,'regex'];
|
|
next; #Is a selection criteria, not an assignment specification
|
|
} elsif (m/^[^=]*!~/) {
|
|
($temp,$value)=split /!~/,$_,2;
|
|
($stable,$scolumn)=getTableColumn($temp);
|
|
$value =~ s/^\///;
|
|
$value =~ s/\/$//;
|
|
$criteria{$stable}->{$scolumn}=[$value,'negex'];
|
|
next; #Is a selection criteria, not an assignment specification
|
|
}
|
|
#Now definitely an assignment
|
|
|
|
($temp, $value) = split('=', $_, 2);
|
|
$value =~ s/^@//; #Allow the =@ operator to exist for an unambiguous assignmenet operator
|
|
#So before, table.column==value meant set to =value, now it would be matching value
|
|
#the new way would be table.column=@=value to be unambiguous
|
|
#now a value like '@hi' would be set with table.column=@@hi
|
|
my $op = '=';
|
|
if ($temp =~ /,$/)
|
|
{
|
|
$op = ',=';
|
|
chop($temp);
|
|
}
|
|
elsif ($temp =~ /\^$/)
|
|
{
|
|
$op = '^=';
|
|
chop($temp);
|
|
}
|
|
|
|
my $table;
|
|
if ($shortnames{$temp})
|
|
{
|
|
($table, $column) = @{$shortnames{$temp}};
|
|
}
|
|
else
|
|
{
|
|
($table, $column) = split('\.', $temp, 2);
|
|
}
|
|
unless (grep /$column/,@{$xCAT::Schema::tabspec{$table}->{cols}}) {
|
|
$callback->({error=>"$table.$column not a valid table.column description",errorcode=>[1]});
|
|
return;
|
|
}
|
|
|
|
# Keep a list of the value/op pairs, in case there is more than 1 per table.column
|
|
#$tables{$table}->{$column} = [$value, $op];
|
|
push @{$tables{$table}->{$column}}, ($value, $op);
|
|
}
|
|
my %nodehash;
|
|
if (keys %criteria) {
|
|
foreach (@$nodes) {
|
|
$nodehash{$_}=1;
|
|
}
|
|
}
|
|
foreach $tab (keys %criteria) {
|
|
my $tabhdl = xCAT::Table->new($tab, -create => 1, -autocommit => 0);
|
|
my @columns=keys %{$criteria{$tab}};
|
|
my $tabhash = $tabhdl->getNodesAttribs($nodes,\@columns);
|
|
my $node;
|
|
my $col;
|
|
my $rec;
|
|
foreach $node (@$nodes) {
|
|
foreach $rec (@{$tabhash->{$node}}) {
|
|
foreach $col (@columns) {
|
|
my $value=$criteria{$tab}->{$col}->[0];
|
|
unless (defined $value) {
|
|
$value = "";
|
|
}
|
|
my $matchtype=$criteria{$tab}->{$col}->[1];
|
|
if ($matchtype eq 'match' and not ($rec->{$col} eq $value) or
|
|
$matchtype eq 'natch' and ($rec->{$col} eq $value) or
|
|
$matchtype eq 'regex' and ($rec->{$col} !~ /$value/) or
|
|
$matchtype eq 'negex' and ($rec->{$col} =~ /$value/)) {
|
|
delete $nodehash{$node};
|
|
}
|
|
}
|
|
}
|
|
}
|
|
$nodes = [keys %nodehash];
|
|
}
|
|
foreach $tab (keys %tables)
|
|
{
|
|
my $tabhdl = xCAT::Table->new($tab, -create => 1, -autocommit => 0);
|
|
if ($tabhdl)
|
|
{
|
|
foreach (@$nodes)
|
|
{
|
|
if ($deletemode)
|
|
{
|
|
$tabhdl->delEntries({'node' => $_});
|
|
}
|
|
else
|
|
{
|
|
|
|
#$tabhdl->setNodeAttribs($_,$tables{$tab});
|
|
my %uhsh;
|
|
my $node = $_;
|
|
foreach (keys %{$tables{$tab}}) # for each column specified for this table
|
|
{
|
|
#my $op = $tables{$tab}->{$_}->[1];
|
|
#my $val = $tables{$tab}->{$_}->[0];
|
|
my @valoppairs = @{$tables{$tab}->{$_}}; #Deep copy
|
|
while (scalar(@valoppairs)) { # alternating list of value and op for this table.column
|
|
my $val = shift @valoppairs;
|
|
my $op = shift @valoppairs;
|
|
my $key = $_;
|
|
if ($op eq '=') {
|
|
$uhsh{$key} = $val;
|
|
}
|
|
elsif ($op eq ',=') { #splice assignment
|
|
my $curval = $uhsh{$key}; # in case it was already set
|
|
if (!defined($curval)) {
|
|
my $cent = $tabhdl->getNodeAttribs($node, [$key]);
|
|
if ($cent) { $curval = $cent->{$key}; }
|
|
}
|
|
if ($curval) {
|
|
my @vals = split(/,/, $curval);
|
|
unless (grep /^$val$/, @vals) {
|
|
@vals = (@vals, $val);
|
|
my $newval = join(',', @vals);
|
|
$uhsh{$key} = $newval;
|
|
}
|
|
} else {
|
|
$uhsh{$key} = $val;
|
|
}
|
|
}
|
|
elsif ($op eq '^=') {
|
|
my $curval = $uhsh{$key}; # in case it was already set
|
|
if (!defined($curval)) {
|
|
my $cent = $tabhdl->getNodeAttribs($node, [$key]);
|
|
if ($cent) { $curval = $cent->{$key}; }
|
|
}
|
|
if ($curval) {
|
|
my @vals = split(/,/, $curval);
|
|
if (grep /^$val$/, @vals) { #only bother if there
|
|
@vals = grep(!/^$val$/, @vals);
|
|
my $newval = join(',', @vals);
|
|
$uhsh{$key} = $newval;
|
|
}
|
|
} #else, what they asked for is the case alredy
|
|
}
|
|
} # end of while @valoppairs
|
|
} # end of foreach column specified for this table
|
|
|
|
if (keys %uhsh)
|
|
{
|
|
my @rc = $tabhdl->setNodeAttribs($node, \%uhsh);
|
|
if (not defined($rc[0]))
|
|
{
|
|
$callback->({error => "DB error " . $rc[1],errorcode=>1});
|
|
}
|
|
}
|
|
}
|
|
}
|
|
$tabhdl->commit;
|
|
}
|
|
else
|
|
{
|
|
$callback->(
|
|
{error => ["ERROR: Unable to open table $tab in configuration"],errorcode=>1}
|
|
);
|
|
}
|
|
}
|
|
}
|
|
|
|
sub tabgrep
|
|
{
|
|
my $node = shift;
|
|
my @tablist;
|
|
my $callback = shift;
|
|
|
|
foreach (keys %{xCAT::Schema::tabspec})
|
|
{
|
|
if (grep /^node$/, @{$xCAT::Schema::tabspec{$_}->{cols}})
|
|
{
|
|
push @tablist, $_;
|
|
}
|
|
}
|
|
foreach (@tablist)
|
|
{
|
|
my $tab = xCAT::Table->new($_);
|
|
unless ($tab) { next; }
|
|
if ($tab and $tab->getNodeAttribs($node->[0], ["node"]))
|
|
{
|
|
$callback->({data => [$_]});
|
|
}
|
|
$tab->close;
|
|
}
|
|
|
|
}
|
|
|
|
#####################################################
|
|
# nodels command
|
|
#####################################################
|
|
sub nodels
|
|
{
|
|
my $nodes = shift;
|
|
my $args = shift;
|
|
my $callback = shift;
|
|
my $noderange = shift;
|
|
unless ($nodes) {
|
|
$nodes=[];
|
|
}
|
|
|
|
my $VERSION;
|
|
my $HELP;
|
|
|
|
my $nodels_usage = sub
|
|
{
|
|
my $exitcode = shift @_;
|
|
my %rsp;
|
|
push @{$rsp{data}}, "Usage:";
|
|
push @{$rsp{data}}, " nodels [noderange] [-H|--with-fieldname] [table.attribute | shortname] [...]";
|
|
push @{$rsp{data}}, " nodels {-v|--version}";
|
|
push @{$rsp{data}}, " nodels [-?|-h|--help]";
|
|
##### xcat 1.2 nodels usage:
|
|
# $rsp->{data}->[1]= " nodels [noderange] [group|pos|type|rg|install|hm|all]";
|
|
# $rsp->{data}->[2]= " ";
|
|
# $rsp->{data}->[3]= " nodels [noderange] hm.{power|reset|cad|vitals|inv|cons}";
|
|
# $rsp->{data}->[4]= " hm.{bioscons|eventlogs|getmacs|netboot}";
|
|
# $rsp->{data}->[5]= " hm.{eth0|gcons|serialbios|beacon}";
|
|
# $rsp->{data}->[6]= " hm.{bootseq|serialbps|all}";
|
|
# $rsp->{data}->[7]= " ";
|
|
# $rsp->{data}->[8]= " nodels [noderange] rg.{tftp|nfs_install|install_dir|serial}";
|
|
# $rsp->{data}->[9]= " rg.{usenis|install_roll|acct|gm|pbs}";
|
|
# $rsp->{data}->[10]=" rg.{access|gpfs|netdevice|prinic|all}";
|
|
if ($exitcode) { $rsp{errorcode} = $exitcode; }
|
|
$callback->(\%rsp);
|
|
};
|
|
|
|
if ($args) {
|
|
@ARGV = @{$args};
|
|
} else {
|
|
@ARGV=();
|
|
}
|
|
my $NOTERSE;
|
|
if (!GetOptions('h|?|help' => \$HELP, 'H|with-fieldname' => \$NOTERSE, 'v|version' => \$VERSION,) ) { nodels_usage->(1); return; }
|
|
|
|
# Help
|
|
if ($HELP) { nodels_usage->(0); return; }
|
|
|
|
# Version
|
|
if ($VERSION)
|
|
{
|
|
my %rsp;
|
|
my $version = xCAT::Utils->Version();
|
|
$rsp{data}->[0] = "$version";
|
|
$callback->(\%rsp);
|
|
return;
|
|
}
|
|
|
|
# TODO -- Parse command arguments
|
|
# my $opt;
|
|
# my %attrs;
|
|
# foreach $opt (@ARGV) {
|
|
# if ($opt =~ /^group/) {
|
|
# }
|
|
# }
|
|
my $argc = @ARGV;
|
|
my $terse = 2;
|
|
if ($NOTERSE) {
|
|
$terse = 0;
|
|
}
|
|
|
|
if (@$nodes > 0 or $noderange)
|
|
{ #Make sure that there are zero nodes *and* that a noderange wasn't requested
|
|
# TODO - gather data for each node
|
|
# for now just return the flattened list of nodes)
|
|
my $rsp; #build up fewer requests, be less chatty
|
|
if ($argc)
|
|
{
|
|
my %tables;
|
|
foreach (@ARGV)
|
|
{
|
|
my $table;
|
|
my $column;
|
|
my $value;
|
|
my $matchtype;
|
|
my $temp = $_;
|
|
if ($temp =~ /^[^=]*\!=/) {
|
|
($temp,$value) = split /!=/,$temp,2;
|
|
$matchtype='natch';
|
|
}
|
|
elsif ($temp =~ /^[^=]*=~/) {
|
|
($temp,$value) = split /=~/,$temp,2;
|
|
$value =~ s/^\///;
|
|
$value =~ s/\/$//;
|
|
$matchtype='regex';
|
|
}
|
|
elsif ($temp =~ /[^=]*==/) {
|
|
($temp,$value) = split /==/,$temp,2;
|
|
$matchtype='match';
|
|
}
|
|
elsif ($temp =~ /[^=]*!~/) {
|
|
($temp,$value) = split /!~/,$temp,2;
|
|
$value =~ s/^\///;
|
|
$value =~ s/\/$//;
|
|
$matchtype='negex';
|
|
}
|
|
if ($shortnames{$temp})
|
|
{
|
|
($table, $column) = @{$shortnames{$temp}};
|
|
$terse--;
|
|
} elsif ($temp =~ /\./) {
|
|
($table, $column) = split('\.', $temp, 2);
|
|
$terse--;
|
|
} elsif ($xCAT::Schema::tabspec{$temp}) {
|
|
$terse=0;
|
|
$table = $temp;
|
|
foreach my $column (@{$xCAT::Schema::tabspec{$table}->{cols}}) {
|
|
unless (grep /^$column$/, @{$tables{$table}}) {
|
|
push @{$tables{$table}},[$column,"$temp.$column"];
|
|
}
|
|
}
|
|
next;
|
|
} else {
|
|
$callback->({error=>"$temp not a valid table.column description",errorcode=>[1]});
|
|
next;
|
|
}
|
|
|
|
|
|
unless (grep /$column/,@{$xCAT::Schema::tabspec{$table}->{cols}}) {
|
|
$callback->({error=>"$table.$column not a valid table.column description",errorcode=>[1]});
|
|
next;
|
|
}
|
|
unless (grep /^$column$/, @{$tables{$table}})
|
|
{
|
|
push @{$tables{$table}},
|
|
[$column, $temp,$value,$matchtype]; #Mark this as something to get
|
|
}
|
|
}
|
|
my $tab;
|
|
my %noderecs;
|
|
my %filterednodes=();
|
|
my %mustdisplaynodes=();
|
|
my %forcedisplaykeys=();
|
|
foreach $tab (keys %tables)
|
|
{
|
|
my $tabh = xCAT::Table->new($tab);
|
|
unless ($tabh) { next; }
|
|
|
|
#print Dumper($tables{$tab});
|
|
my $node;
|
|
my %labels;
|
|
my %values;
|
|
my %matchtypes;
|
|
my @cols=();
|
|
foreach (@{$tables{$tab}})
|
|
{
|
|
push @cols, $_->[0];
|
|
$labels{$_->[0]} = $_->[1]; #Remember user supplied discreptions and use them
|
|
if (not defined $values{$_->[0]}) { #If selection criteria not previously specified
|
|
$values{$_->[0]} = $_->[2]; #assign selection criteria
|
|
} elsif (not defined $_->[2]) { #we already have selection criteria, but this field isn't that
|
|
$forcedisplaykeys{$_->[0]}=1; #allow switch.switch=~switch switch.switch, for example
|
|
} else { #User attempted multiple selection criteria on the same field, bail
|
|
$callback->({error=>["Multiple selection critera for ".$labels{$_->[0]}]});
|
|
return;
|
|
}
|
|
if (not defined $matchtypes{$_->[0]}) {
|
|
$matchtypes{$_->[0]} = $_->[3];
|
|
}
|
|
}
|
|
my $removenodecol=1;
|
|
if (grep /^node$/,@cols) {
|
|
$removenodecol=0;
|
|
}
|
|
my $rechash=$tabh->getNodesAttribs($nodes,\@cols);
|
|
foreach $node (@$nodes)
|
|
{
|
|
my @cols;
|
|
my $recs = $rechash->{$node}; #$tabh->getNodeAttribs($node, \@cols);
|
|
my %satisfiedreqs=();
|
|
foreach my $rec (@$recs) {
|
|
|
|
foreach (keys %$rec)
|
|
{
|
|
if ($_ eq "node" and $removenodecol) { next; }
|
|
$satisfiedreqs{$_}=1;
|
|
my %datseg=();
|
|
if (defined $values{$_}) {
|
|
my $criteria=$values{$_}; #At least vim highlighting makes me worry about syntax in regex
|
|
if ($matchtypes{$_} eq 'match' and not ($rec->{$_} eq $criteria) or
|
|
$matchtypes{$_} eq 'natch' and ($rec->{$_} eq $criteria) or
|
|
$matchtypes{$_} eq 'regex' and ($rec->{$_} !~ /$criteria/) or
|
|
$matchtypes{$_} eq 'negex' and ($rec->{$_} =~ /$criteria/)) {
|
|
#unless ($rec->{$_} eq $values{$_}) {
|
|
$filterednodes{$node}=1;
|
|
next;
|
|
}
|
|
$mustdisplaynodes{$node}=1;
|
|
unless ($forcedisplaykeys{$_}) { next; } #skip if only specified once on command line
|
|
}
|
|
unless ($terse > 0) {
|
|
$datseg{data}->[0]->{desc} = [$labels{$_}];
|
|
}
|
|
$datseg{data}->[0]->{contents} = [$rec->{$_}];
|
|
$datseg{name} = [$node]; #{}->{contents} = [$rec->{$_}];
|
|
push @{$noderecs{$node}}, \%datseg;
|
|
}
|
|
}
|
|
foreach (keys %labels) {
|
|
unless (defined $satisfiedreqs{$_}) {
|
|
my %dataseg;
|
|
if (defined $values{$_}) {
|
|
my $criteria = $values{$_};
|
|
if ($matchtypes{$_} eq 'match' and not ("" eq $criteria) or
|
|
$matchtypes{$_} eq 'natch' and ("" eq $criteria) or
|
|
$matchtypes{$_} eq 'regex' and ("" !~ /$criteria/) or
|
|
$matchtypes{$_} eq 'negex' and ("" =~ /$criteria/)) {
|
|
#unless ("" eq $values{$_}) {
|
|
$filterednodes{$node}=1;
|
|
next;
|
|
}
|
|
$mustdisplaynodes{$node}=1;
|
|
unless ($forcedisplaykeys{$_}) { next; }
|
|
}
|
|
$dataseg{name} = [ $node ];
|
|
unless ($terse > 0) {
|
|
$dataseg{data}->[0]->{desc} = [$labels{$_}];
|
|
}
|
|
$dataseg{data}->[0]->{contents} = [""];
|
|
push @{$noderecs{$node}}, \%dataseg;
|
|
}
|
|
}
|
|
}
|
|
|
|
#$rsp->{node}->[0]->{data}->[0]->{desc}->[0] = $_;
|
|
#$rsp->{node}->[0]->{data}->[0]->{contents}->[0] = $_;
|
|
$tabh->close();
|
|
undef $tabh;
|
|
}
|
|
foreach (keys %mustdisplaynodes) {
|
|
if ($filterednodes{$_} or defined $noderecs{$_}) {
|
|
next;
|
|
}
|
|
$noderecs{$_}=[{name=>[$_]}];
|
|
}
|
|
foreach (keys %filterednodes) {
|
|
delete $noderecs{$_};
|
|
}
|
|
foreach (sort (keys %noderecs))
|
|
{
|
|
push @{$rsp->{"node"}}, @{$noderecs{$_}};
|
|
}
|
|
}
|
|
else
|
|
{
|
|
foreach (@$nodes)
|
|
{
|
|
my $noderec;
|
|
$noderec->{name}->[0] = ($_);
|
|
push @{$rsp->{node}}, $noderec;
|
|
}
|
|
}
|
|
$callback->($rsp);
|
|
}
|
|
else
|
|
{
|
|
|
|
# no noderange specified on command line, return list of all nodes
|
|
my $nodelisttab;
|
|
if ($nodelisttab = xCAT::Table->new("nodelist"))
|
|
{
|
|
my @attribs = ("node");
|
|
my @ents = $nodelisttab->getAllAttribs(@attribs);
|
|
foreach (@ents)
|
|
{
|
|
my $rsp;
|
|
if ($_->{node})
|
|
{
|
|
$rsp->{node}->[0]->{name}->[0] = ($_->{node});
|
|
|
|
# $rsp->{node}->[0]->{data}->[0]->{contents}->[0]="$_->{node} node contents";
|
|
# $rsp->{node}->[0]->{data}->[0]->{desc}->[0]="$_->{node} node desc";
|
|
$callback->($rsp);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
return 0;
|
|
}
|