source: fedkit/splitter.pl @ c9f5490

#!/usr/bin/perl

use Getopt::Std;
use IO::File;
use IO::Dir;
use IO::Pipe;
use File::Copy;

@scripts = ("federate.sh", "smbmount.pl");
$local_script_dir = ".";

# Parse the config file.  The format is a colon-separated parameter name
# followed by the value of that parameter to the end of the line.  This parses
# that format and puts the parameters into the referenced hash.  Parameter
# names are mapped to lower case, parameter values are unchanged.  Returns 0 on
# failure (e.g. file open) and 1 on success.
sub parse_config {
    my($file, $href) = @_;
    my($fh) = new IO::File($file);
        
    unless ($fh) {
        warn "Can't open $file: $!\n";
        return 0;
    }

    while (<$fh>) {
        next if /^\s*#/ || /^\s*$/;     # Skip comments & blanks
        chomp;
        /^([^:]+):\s*(.*)/ && do {
            my($key) = $1; 

            $key =~ tr/A-Z/a-z/;
            $href->{$key} = $2;
            next;
        };
        warn "Unparasble line in $file: $_\n";
    }
    $fh->close();   # It will close when it goes out of scope, but...
    return 1;
}


# use scp to transfer a file, reporting true if successful and false otherwise.
# Parameters are the local file name, the ssh host destination (either hostname
# oe user@host), and an optional destination file name or directory.  If no
# destination is given, the file is transferred to the given user's home
# directory.  If only a machine is given in the ssh host destination, the
# current user is used.
sub scp_file {
    my($file, $where, $dest) = @_;

    # XXX system with a relative pathname is sort of gross
    system("scp $file $where:$dest");
    if ($?) {
        warn "scp failed $?\n";
        return 0;
    }
    else { return 1; }
}

# use ssh to execute the given command on the machine (and as the user) in
# $where.  Parameters are the ssh destination directive ($where) and the
# command to execute, and a prefix to be placed on a message generated if the
# command fails.   On failure print a warning if a warning prefix was given and
# return false.
sub ssh_cmd {
    my($user, $host, $cmd, $wname) = @_;

    # XXX system with a relative pathname is sort of gross
    system ("ssh $user\@$host $cmd");
    if ($?) {
        warn "$wname failed $?\n" if $wname;
        return 0;
    }
    else { return 1; }
}

# Ship local copies of the federation scripts out to the given host.  If any of
# the script transfers fails, return 0.  The scripts to transfer are from the
# global @scripts and are found locally in $local_script_dir (another global).
sub ship_scripts {
    my($host, $user, $dest_dir) = @_;       # Where, who, where remotely
    my($s);

    for $s (@scripts) {
        &scp_file("$local_script_dir/$s", "$user\@$host", $dest_dir) || 
            return 0;
    }
    return 1;
}

# Ship per-testbed configuration generated by this script to the remote /proj
# directories on the remote testbeds
sub ship_configs {
    my($host, $user, $src_dir, $dest_dir) = @_;     # Where, who, where remotely
    my($d, $f);

    $d = IO::Dir->new($src_dir) || return 0;

    while ( $f = $d->read()) {
        next if $f =~ /^\./;
        &scp_file("$src_dir/$f", "$user\@$host", $dest_dir) || return 0;
    }
    return 1;
}




# Start a sub section of the experiment on a given testbed.  The testbed and
# the user to start the experiment as are pulled from the global per-testbed
# hash, as is the project name on the remote testbed.  Parameters are the
# testbed and the experiment id.  Configuration files are scp-ed over to the
# target testbed from the global $tmpdir/$tb directory.  Then the current state
# of the experiment determined using expinfo.  From that state, the experiment
# is either created, modified or spapped in.  If everything succeeds, true is
# returned.
sub start_segment {
    my($tb, $eid) = @_;                     # testbed and experiment ID
    my($host) = "$host{$tb}$domain{$tb}";   # Host name of remote ops (FQDN)
    my($user) = $user{$tb};                 # user to pass to ssh
    my($pid) = $project{$tb};               # remote project to start the
                                            # experiment under
    my($tclfile) = "./$eid.$tb.tcl";        # Local tcl file with the
                                            # sub-experiment
    my($proj_dir) = "/proj/$pid/exp/$eid/tmp";  # Where to stash federation stuff
    my($to_hostname) = "$proj_dir/hosts";   # remote hostnames file
    my($status) = new IO::Pipe;             # The pipe to get status

    # Determine the status of the remote experiment
    $status->reader("ssh $user\@$host /usr/testbed/bin/expinfo $pid $eid") || 
        die "Can't ssh to $user\@$host:$!\n";

    # XXX: this is simple now.  Parsing may become more complex
    while (<$status>) {
        /State: (\w+)/ && ($state = $1);
        /No\s+such\s+experiment/ && ($state = "none");
    }
    $status->close();
    print "$tb: $state\n";

    # Copy the experiment definition data over (unless the host is local)
    &scp_file("$tmpdir/$tb/$tclfile", "$user\@$host") || return 0;
    # Remote experiment is active.  Modify it.
    if ($state eq "active") {
        # First copy new scripts and hostinfo into the remote /proj
        &scp_file("$tmpdir/$tb/hostnames", "$user\@$host", $to_hostname) ||
            return 0;
        &ship_scripts($host, $user, $proj_dir) || return 0;
        &ship_configs($host, $user, "$tmpdir/$tb", $proj_dir) || return 0;
        &ssh_cmd($user, $host, "/usr/testbed/bin/modexp -r -s -w $pid " . 
            "$eid $tclfile", "modexp") || return 0;
        return 1;
    }

    # Remote experiment is swapped out, modify it and swap it in.
    if ($state eq "swapped") {
        &ssh_cmd($user, $host, "/usr/testbed/bin/modexp -w $pid $eid $tclfile", 
            "modexp") || return 0;
        # First copy new scripts and hostinfo into the remote /proj
        &scp_file("./hostnames", "$user\@$host", $to_hostname) || return 0;
        &ship_scripts($host, $user, $proj_dir) || return 0;
        &ship_configs($host, $user, "$tmpdir/$tb", $proj_dir) || return 0;
        # Now start up
        &ssh_cmd($user, $host, "/usr/testbed/bin/swapexp -w $pid $eid in", 
            "swapexp") || return 0;
        return 1;
    }

    # No remote experiment.  Create one.  We do this in 2 steps so we can put
    # the configuration files and scripts into the new experiment directories.
    if ($state eq "none") {
        &ssh_cmd($user, $host, "/usr/testbed/bin/startexp -f -w -p " . 
            "$pid -e $eid $tclfile", "startexp") || return 0;
        # First copy new scripts and hostinfo into the remote /proj
        &scp_file("./hostnames", "$user\@$host", $to_hostname) || return 0;
        &ship_scripts($host, $user, $proj_dir) || return 0;
        &ship_configs($host, $user, "$tmpdir/$tb", $proj_dir) || return 0;
        # Now start up
        &ssh_cmd($user, $host, "/usr/testbed/bin/swapexp -w $pid $eid in", 
            "swapexp") || return 0;
        return 1;
    }

    # Every branch for a known state returns.  If execution gets here, the
    # state is unknown.
    warn "unknown state: $state\n";
    return 0;
}

# Swap out a sub-experiment - probably because another has failed.  Arguments
# are testbed and experiment.  Most of the control flow is similar to
# start_segment, though much simpler.
sub stop_segment {
    my($tb, $eid) = @_;
    my($user) = "$user{$tb}";
    my($host) = "$host{$tb}$domain{$tb}";
    my($pid) = $project{$tb};

    &ssh_cmd($user, $host, "/usr/testbed/bin/swapexp -w $pid $eid out", 
        "swapexp (out)") || return 0;
    return 1;
}


$pid = $gid = "dummy";              # Default project and group to pass to
                                    # $tcl_splitter above.  These are total
                                    # dummy arguments;  the splitter doesn't
                                    # use them at all, but we supply them to
                                    # keep our changes to the parser minimal.
# Argument processing.
getopts('c:f:nd', \%opts);
$splitter_config = $opts{'c'} || "./splitter.conf";
$debug = $opts{'d'};
&parse_config("$splitter_config", \%opts) || 
    die "Cannot read config file $splitter_conf: $!\n";


$startem = $opts{'n'} ? 0 : 1;          # If true, start the sub-experiments
$eid = $opts{'experiment'};             # Experiment ID
$tcl = $opts{'f'} || shift;             # The experiment description
$master = $opts{'master'};              # Master testbed
$tmpdir = $opts{'tmpdir'} || $opts{'tempdir'}|| "/tmp"; # tmp files
$tb_config = $opts{'testbeds'} || "./testbeds"; # testbed configurations
$local_script_dir = $opts{'scriptdir'}; # Local scripts
# For now specify these.  We may want to generate them later.
$gw_pubkey = $opts{'gatewaypubkey'};
($gw_pubkey_base = $gw_pubkey) =~ s#.*/##;
$gw_secretkey = $opts{'gatewaysecretkey'};
($gw_secretkey_base = $gw_secretkey) =~ s#.*/##;

# tcl program to split experiments (changed during devel)
$tcl_splitter = $opts{'tclparse'} || "/usr/testbed/lib/ns2ir/parse.tcl";
# tclsh to call directly (changed during devel)
$tclsh = $opts{'tclsh'} || "/usr/local/bin/otclsh";

# Prefix to avoid collisions
$tmpdir .= "/split$$";

# Create a workspace
unless (-d "$tmpdir") {
    mkdir("$tmpdir") || die "Can't create $tmpdir: $!";
}

# Validate scripts directory
for $s (@scripts) {
    die "$local_script_dir/$s not in local script directory. Try -d\n"
        unless -r "$local_script_dir/$s";
}

die "Must supply file, master and experiment" unless $master && $tcl && $eid;

# Read a hash of per-testbed parameters from the local configurations.
$conf = new IO::File($tb_config) || 
    die "can't read testbed configutions from $tb_config: $!\n";
while (<$conf>) {
    next if /^#/;
    chomp;
    ($tb, $h, $d, $u, $p, $es, $gs, $mes, $mgs, $t, $i) = split(":", $_);
    $host{$tb} = $h;
    $user{$tb} = $u;
    $domain{$tb} = $d;
    $project{$tb} = $p;
    $gwtype{$tb} = $t;
    $expstart{$tb} = $es;
    $gwstart{$tb} = $gs;
    $mexpstart{$tb} = $mes;
    $mgwstart{$tb} = $mgs;
    $gwimage{$tb} = $i;

    # Make sure the domain starts with a period
    $domain{$tb} = ".$domain{$tb}" unless $domain{$tb} =~ /^\./;
}
$conf->close();

# Open a pipe to the splitter program and start it parsing the experiments
$pipe = new IO::Pipe;
$pipe->reader("$tclsh $tcl_splitter -s -m $master -p $pid $gid $eid $tcl") || 
    die "Cannot execute $tclsh $tcl_splitter -s -p $pid $gid $eid $tcl:$!\n";

# Parse the splitter output.  This loop creates the sub experiments, gateway
# configurations and hostnames file
while (<$pipe>) {
    # Start of a sub-experiment
    /^#\s+Begin\s+Testbed\s+\((\w+)\)/ && do {
        $ctb = $1;

        # If we know the testbed, start collecting its sub experiment tcl
        # description.  If not, warn the caller and ignore the configuration of
        # this testbed.
        if ($host{$ctb}) {
            $allocated{$ctb}++; # Keep track of the testbeds allocated

            unless (-d "$tmpdir/$ctb") {
                mkdir("$tmpdir/$ctb") || die "Can't create $tmpdir/$ctb: $!";
            }
            $destfile = "$tmpdir/$ctb/$eid.$ctb.tcl";

            open(FILE, ">$destfile") || die "Cannot open $destfile:$!\n";
        }
        else { 
            warn "No such testbed $ctb\n";
            $destfile = "";
        }
        next;
    };

    # End of that experiment
    /^#\s+End\s+Testbed\s+\((\w+)\)/ && do {
        # Simple syntax check and close out this experiment's tcl description
        die "Mismatched testbed markers ($1, $ctb)\n" unless ($1 eq $ctb);
        close(FILE);
        $destfile = $ctb = "";
        next;
    };

    # Beginning of a gateway set
    /^#\s+Begin\s+gateways\s+\((\w+)\)/ && do {
        $gateways = $1;
        # If we've heard of this tb, create the config lines for it one at a
        # time.
        if ($allocated{$gateways}) {
            # Just in case.  This directory should already have been created
            # above.
            unless (-d "$tmpdir/$gateways") {
                mkdir("$tmpdir/$gateways") || 
                    die "Can't create $tmpdir/$gateways: $!";
            }
        }
        else {
            warn "Gateways given (and ignored) for testbed not in use: " .
                "$gateways\n";
            $gateways = 0;
        }
        next;
    };
    /^#\s+End\s+gateways\s+\((\w+)\)/ && do {
        die "Mismatched gateway markers ($1, $gateways)\n" 
            unless !$gateways || $gateways == $1;
        $gateways = 0;
        next;
    };
    # Beginning of the hostnames list.  Collection is always in the hostnames
    # file.
    /^#\s+Begin\s+hostnames/ && do {
        $destfile = "$tmpdir/hostnames";
        open(FILE, ">$destfile") || die "Can't open $destfile:$!\n";
        next;
    };
    # end of the hostnames list.
    /^#\s+End\s+hostnames/ && do {
        close(FILE);
        $destfile = "";
        next;
    };

    # Generate gateway configuration info, one file per line
    $gateways && do {
        chomp;
        my($dtb, $myname, $desthost, $type) = split(" ", $_);
        my($sdomain) = $domain{$gateways};      # domain for the source
        my($ddomain) = $domain{$dtb};           # domain for the destination
        my($sproject) = $project{$gateways};    # Project of the destination

        $sdomain = ".$eid.$project{$gateways}$sdomain";
        $ddomain = ".$eid.$project{$dtb}$ddomain";

        # If either end of this link is in the master side of the testbed, that
        # side is the active end. Otherwise the first testbed encountered in
        # the file will be the active end.  The $active_end variable keeps
        # track of those decisions
        if ( $dtb eq $master ) { $active = "false"; }
        elsif ($gateways eq $master ) { $active = "true"; }
        elsif ( $active_end{"$dtb-$gateways"} ) { $active="false"; }
        else { $active_end{"$gateways-$dtb"}++; $active = "true"; }

        # Write out the file
        open(GWCONFIG, ">$tmpdir/$gateways/$myname$sdomain.gw.conf") || 
            die "can't open $tmpdir/$gateways/$myname$sdomain.gw.conf: $!\n";
        print GWCONFIG "Active: $active\n";
        print GWCONFIG "Type: $type\n";
        print GWCONFIG "Peer: $desthost$ddomain\n";
        print GWCONFIG "Pubkeys: " . 
            "/proj/$sproject/exp/$eid/tmp/$gw_pubkey_base\n";
        print GWCONFIG "Privkeys: " .
            "/proj/$sproject/exp/$eid/tmp/$gw_secretkey_base\n";
        close(GWCONFIG);

        # This testbed has a gateway (most will) so make a copy of the keys it
        # needs in this testbed's subdirectory.  start_segment will transfer
        # them.
        unless (-r "$tmpdir/$gateways/$gw_pubkey_base" ) {
            copy($gw_pubkey, "$tmpdir/$gateways/$gw_pubkey_base") ||
                die "Can't copy pubkeys ($gw_pubkey to " . 
                    "$tmpdir/$gateways/$gw_pubkey_base): $!\n";
        }
        if ($active eq "true" ) {
            unless (-r "$tmpdir/$gateways/$gw_secretkey_base" ) {
                copy($gw_secretkey, "$tmpdir/$gateways/$gw_secretkey_base") ||
                    die "Can't copy secret keys ($gw_secretkey to " . 
                        "$tmpdir/$gateways/$gw_secretkey_base): $!\n";
            }
        }

        #done processing gateway entry, ready for next line
        next; 
    };


    next unless $destfile;  # Unidentified testbed, ignore config

    # Substitute variables
    s/GWTYPE/$gwtype{$ctb}/g;
    s/GWIMAGE/$gwimage{$ctb}/g;
    if ($ctb eq $master ) {
        s/GWSTART/$mgwstart{$ctb}/g;
        s/EXPSTART/$mexpstart{$ctb}/g;
    }
    else {
        s/GWSTART/$gwstart{$ctb}/g;
        s/EXPSTART/$expstart{$ctb}/g;
    }
    # XXX: oh is this bad
    s#GWCONF#FEDDIR\`hostname\`.gw.conf#g;
    s#FEDDIR#/proj/$project{$ctb}/exp/$eid/tmp/#g;
    print FILE;
}
$pipe->close();
die "No nodes in master testbed ($master)\n" unless $allocated{$master};

exit(0) unless $startem;

# Start up the slave sub-experiments first
TESTBED:
for $tb (keys %allocated) {
    if ($tb ne $master) {
        if (&start_segment($tb, $eid)) { $started{$tb}++; }
        else { last TESTBED; }
    }
}

# Now the master
if (&start_segment($master, $eid)) { 
    $started{$master}++;
}

# If any testbed failed, swap the rest out.
if ( scalar(keys %started) != scalar(keys %allocated)) {
    for $tb (keys %started) { &stop_segment($tb, $eid); }
    print "Error starting experiment\n";
    exit(1);
}
print "Experiment started\n";
system("rm -rf $tmpdir") unless $debug;
exit(0);    # set the exit value

=pod

=head1 NAME

B<splitter.pl>

=head1 SYNOPSIS

B<splitter.pl> [B<-nd>] [B<-c> F<config_file>] [B<-f> F<experiment_tcl>]
    [F<experiment_tcl>]

=head1 DESCRIPTION

B<splitter.pl> invokes the DETER experiment parser to split an annotated
experiment into multiple sub-experments and instantiates the sub-experiments on
their intended testbeds.  Annotation is accomplished using the
tb-set-node-testbed command, added to the parser.

The testbed labels are meaningful based on their presence in the testbeds file.
that file can be specified in the configuration file using the B<Testbeds>
directive, and defaults to F<./testbeds>.  The syntax is described below.

Most of the intermediate files are staged in a sub-directory of a temporary
files directory and deleted at the end of the script.  Specifying the B<-d>
flag on the command line avoids the deletion for debbugging.  By default the
temporary files directory is directory is F</tmp> and can be reset in the
configuration file using the B<Tmpdir> directive.  Intermediate files are
stored under a subdirectory formed by adding the process ID of the splitter
process.  For example, if the temporary files directory is F</tmp> and the
B<splitter.pl> process ID is 2323, the temporary files will be stored in
F</tmp/split2323/>.

The expreriment is split out into one experiment description per testbed in the
temporary directory named as F<experiment.testbed.tcl> where the experiment is
the experiment ID given in the configuration file, and the testbed is the
tb-set-node-testbed parameter for the nodes in the file.

If the B<-n> option is absent the sub-experiments are then instantiated on
their testbeds.  (Here B<-n> is analogous to its use in L<make(1)>).
Per-testbed parameters are set in the testbeds file.  Sub-experiments on
slave testbeds are instantiated in a random order, but the master testbed is
currently instantiated last.

Scripts to start federation (the federation kit) are copied into the local
experiment's tmp file - e.g., F</proj/DETER/exp/simple-split/tmp>.  These are
taken from the directory given by the B<ScriptDir> directive in the
configuration file.

If any sub-experiment fails to instantiate, the other sub-exeriments are
swapped out.

=head2 Configuration File

The configuration file is a simple attribute-value pair set of colon-separated
parameters and values.  A configuration file must be present, either specified
in the B<-c> flag or the default F<./splitter.conf>.  All the parameter names
are case insensitive, but should not include any whitespace.  Parameter values
may include whitespace, but no newlines.

Possible parameters are:

=over 5

=item Experiment

The name of the experiment on the various testbeds

=item Master

The master testbed label from the testbeds file, described below.

=item Testbeds

The testbeds file described below, giving per-testbed parameters.  If this
directive is absent the testbeds file defaults to F<./testbeds>

=item ScriptDir

Location of the default federation scripts, i.e. the federation kit.

=item GatewayPubkey

=item GatewaySecretKey

The names of the files containing secret and public keys to use in setting up
tunnels between testbeds.  These will eventually be automatically generated.

=item TmpDir

=item TempDir

The directory where temporary files are created.  These are synonyms, but
should both be specified, B<TmpDir> has priority.  If neither is specified,
F</tmp> is used.

=item Tclparse

The pathname to the experiment parsing program.  Only developers should set
this.

=item Tclsh

The pathname to the local oTcl shell.  Only developers should set
this.

=back

=head2 Testbeds file

The configuration file (F<./testbeds> unless overridden by B<-c>) is a
colon-separated set of parameters keyed by testbed name.  The fields, in order,
are:

=over 5

=item name

The testbed to which this line of parameters applies.

=item user

The user under which to make requests to this testbed.  The user running
B<splitter.pl> must be able to authenicate as this user under L<ssh(1)> to this
testbed.

=item host

The host name of the testbed's ops node.  The user calling B<splitter.pl> must
be able to execute commands on this host via L<ssh(1)>.

=item domain

The domain of nodes in this testbed (including the ops host).

=item project

The project under which to instantiate sub-experiments on this testbed.

=item gateway type

The node type for inter-testbed gateway nodes on this testbed.

=item experiment start (slave)

The start command to run on experimental nodes when this testbed is used as a
slave.  In all the start commands the string FEDDIR will be replaced by the
local experiment's federation scripts directory and the string GWCONF replaced
by the gatway configuration file.

=item gateway start (slave)

The start command to run on gateway nodes when this testbed is used as a slave.
The same string substitutions are made in this command as in experiment start.

=item experiment start (master)

The start command to run on experimental nodes when this testbed is used as a
master.  The same string substitutions are made in this command as in
experiment start.

=item gateway start (master)

The start command to run on gateway nodes when this testbed is used as a
master.  The same string substitutions are made in this command as in
experiment start.

=item gateway image

The disk image to be loaded on a gateway node on this testbed.

=back

The parsing of the testbeds is extremely simple.  Colons separate each
field and there is n provision for escaping them at this time.

=head1 ENVIRONMENT

B<splitter.pl> does not directly make use of environment variables, but calls
out to L<ssh(1)> and (indirectly) to L<sh(1)>, which may be influenced by the
environment.

=head1 SEE ALSO

L<sh(1)>, L<ssh(1)> 

=cut