Skip to content

GitLab

Commit 11950a44 authored by Robert Ricci's avatar Robert Ricci
Browse files
Options

DB-ification major reformatting and commenting pass. 

Now back to its old behavior, where frisbee reloads are handled no differently
than other reloads, because this difference is now taken care of in the
reload_daemon.

Note that the '-w' flag has been removed. This is due to an error that perl
reports on line 123, which, as far as I can tell, is not real. The code works
fine, anyway.
parent 230b1727
Hide whitespace changes
Inline Side-by-side
Showing with 229 additions and 225 deletions
db/nfree.in
View file @ 11950a44
#!/usr/local/bin/perl -w
use Mysql;
#!/usr/bin/perl
#
# nfree - Takes pysical node names, and frees them from the experiment they
# are allocated to. If nodes are ommited, frees up all nodes in the given
# experiment. Looks in the scheduled_reloads and next_reserve tables to see
# if this node should be re-reserved into another experiment and/or reloaded,
# rather than being put back into the pool of free nodes
#
#
# Configure variables
#
my $TB = "@prefix@";
my $DBNAME = "@TBDBNAME@";
use lib '@prefix@/lib';
use libdb;
my $dbh = Mysql->connect("localhost",$DBNAME,"script","none");
if ($#ARGV < 1) {
die("Usage: nfree <pid> <eid> [<node> <node> <...>]\n".
"Releases all nodes in the specified experiment. If nodes are listed,\n".
"nfree releases only the listed nodes.\n");
if (@ARGV < 2) {
die("Usage: nfree <pid> <eid> [<node> <node> <...>]\n".
"Releases all nodes in the specified experiment. If nodes are listed,\n".
"nfree releases only the listed nodes.\n");
}
my $error = 0;
......@@ -22,254 +28,247 @@ my $os_load = "$TB/bin/os_load -r ";
my $reloadpid="emulab-ops";
my $reload_pendingeid="reloadpending";
my $reloadeid="reloading";
my $pid = shift;
my $eid = shift;
my @node_names=();
my @freed_nodes=();
foreach my $n (@ARGV) {
if ($n =~ /(sh\d+)/ ) {
# its a shark - do the whole shelf if its not done already.
my $shelf = $1;
if ( ! (join(",",@node_names) =~ /,$shelf-\d,/)) {
# Shelf hasn't been done yet...
foreach my $n ( 1 .. 8 ) {
push(@node_names,"$shelf-$n");
}
}
} else {
# its not a shark - just add it in...
push(@node_names,"$n");
}
my $pid = shift;
my $eid = shift;
use strict;
# Make sure that the experiment actually exists
# NOTE: project permissions checking is done later, on an individual
# node basis.
if (!ExpState($pid,$eid)) {
die("There is no experiment '$eid' in project '$pid'.\n");
}
######################################################################
# Step 1 - Free nodes
#
# Find nodes that can be freed at this time, and do so. Nodes which
# are awaiting reloads and which have been scheduled to be reserved
# to another experiment, are put into lists so that they can be
# handled later
######################################################################
my $cmd = "";
my $sth = "";
# Make a list of nodes given on the command line
foreach my $n (@ARGV) {
# Shark hack
if ($n =~ /(sh\d+)/ ) {
#
# It's a shark - do the whole shelf if its not done already.
#
my $shelf = $1;
if ( ! (join(",",@node_names) =~ /,$shelf-\d,/)) {
# Shelf hasn't been done yet...
foreach my $n ( 1 .. 8 ) {
push(@node_names,"$shelf-$n");
}
}
# End shark hack
$cmd = "select * from experiments where eid='$eid' and pid='$pid'";
$sth = $dbh->query($cmd);
if ($sth->numrows < 1) {
die("There is no experiment '$eid' in project '$pid'.\n");
} else {
# its not a shark - just add it in...
push(@node_names,"$n");
}
}
$cmd = "lock tables nodes write, node_types read, scheduled_reloads read, ".
"interfaces write, delays write, reserved write, last_reservation write,".
"current_reloads write, next_reserve read";
$sth = $dbh->query($cmd)
|| die("Locking error:\n$cmd\nError string is:".$dbh->errstr."\n");
# If list is empty, put in all the nodes
if ($#node_names == -1) {
print "Releasing all nodes from experiment '$eid' in project '$pid'...\n";
$sth = $dbh->
query("select node_id from reserved where pid='$pid' and eid='$eid'");
while (@row = $sth->fetchrow_array()) {
push(@node_names, $row[0]);
}
#
# Lock all of the tables we'll be reading, so that we get a consistent
# view of the current state
#
DBQueryFatal("lock tables nodes write, node_types read, " .
"scheduled_reloads read, interfaces write, delays write, " .
"reserved write, last_reservation write, current_reloads write, " .
"next_reserve read");
#
# If no nodes were given on the command line, fill the list with all nodes
# in the experiment.
#
if (@node_names == 0) {
print "Releasing all nodes from experiment '$eid' in project '$pid'...\n";
push @node_names,ExpNodes($pid,$eid);
}
my @reloads = ();
# Frisbee reloads get handled differently than 'normal' reloads
my @frisbee_reloads = ();
my %reserves= ();
foreach my $n (@node_names) {
$sth = $dbh->query("select * from reserved where node_id='$n' ".
"and eid='$eid' and pid='$pid'");
if ($sth->numrows == 0) {
print "Node '$n' is not reserved by your experiment.\n";
$error++;
next;
} else {
push(@freed_nodes,$n);
}
# This little sillyness is for disk reloading. Kill the last reservation.
$cmd = "delete from last_reservation where node_id='$n'";
$dbh->query($cmd);
# If the node has a reloads entry, change the reservation and start it
$cmd = "select node_id,image_id,reload_type from scheduled_reloads ".
"where node_id='$n'";
$sth = $dbh->query($cmd)
|| (print "Failed Command:\n$cmd\nError string is:".$dbh->errstr."\n"
&& $error++);
if ( ($sth->num_rows()) > 0) {
if ($sth->num_rows() != 1) {
print "Warning: multiple reloads scheduled for $n - using the first\n";
}
my @row = $sth->fetchrow();
#
# Check to make sure they have acutally reserved the nodes.
#
my $result = DBQueryFatal("select * from reserved where node_id='$n' ".
"and eid='$eid' and pid='$pid'");
if ($result->numrows == 0) {
print "Node '$n' is not reserved by your experiment.\n";
$error++;
next;
} else {
push(@freed_nodes,$n);
}
#
# This little sillyness is for disk reloading. Kill the last reservation.
#
DBQueryWarn("delete from last_reservation where node_id='$n'") || $error++;
#
# If the node has a reloads entry, change the reservation so that the
# reload_daemon will pick it up.
#
$result = DBQueryFatal("select node_id,image_id from scheduled_reloads " .
"where node_id='$n'");
if ( $result->num_rows() > 0 ) {
my @row = $result->fetchrow();
my $image_id = $row[1];
my $reload_type = $row[2];
# Check for Frisbee-type reloads
if ($reload_type eq "frisbee") {
print "Adding scheduled reload for $n of $image_id to the list using Frisbee.\n";
# We'll need to know both the node and image_id to run os_load
push(@frisbee_reloads,[$n,$image_id]);
print "Adding scheduled reload for $n to the list.\n";
push(@reloads,$n);
} else {
#
# If the node has a next_reserve entry, change the reservation
#
my $result = DBQueryFatal("select node_id,pid,eid from next_reserve ".
"where node_id='$n'");
if ( $result->num_rows() > 0 ) {
#
# Add the reservation to a list to be taken care of later
#
my ($node, $next_pid, $next_eid) = $result->fetchrow_array();
$reserves{$node} = "$next_pid:$next_eid";
} else {
print "Adding scheduled reload for $n to the list.\n";
push(@reloads,$n);
#
# No reloads or reservation changes, so really free the node
#
# This little sillyness is for disk reloading. Remember the last
# project a node was reserved into.
#
DBQuery("insert into last_reservation values ('$n', '$pid')");
print "Releasing node '$n'...";
if (DBQueryWarn("delete from reserved " .
"where node_id='$n' and eid='$eid'")) {
"Succeeded.\n";
} else {
$error++;
}
}
} else {
# If the node has a next_reserve entry, change the reservation and start it
$cmd = "select node_id,pid,eid from next_reserve where node_id='$n'";
$sth = $dbh->query($cmd)
|| (print "Failed Command:\n$cmd\nError string is:".$dbh->errstr."\n"
&& $error++);
if ( ($sth->num_rows()) > 0) {
my @reserve;
print "Adding next reservation for $n to the list.\n";
my $i = 0;
my $max = $sth->num_rows();
while ( $i < $max ) {
$i++;
@reserve = $sth->fetchrow_array();
my ($node, $next_pid, $next_eid) = @reserve;
$reserves{$node} = "$next_pid:$next_eid";
}
}
#
# Find the control net interface for this node type, as well as some
# of the default values for its node type
#
$result = DBQueryFatal("select control_net,osid,node_types.pxe_boot_path " .
"from node_types " .
"left join nodes on nodes.type=node_types.type " .
"where node_id='$n'");
my ($control, $osid, $pxe_boot_path) = $result->fetchrow_array();
if (! ($n =~ /sh\d+/)) { # If its not a shark
# Clean out all IPs except the control net
DBQueryWarn("update interfaces set IP='' " .
"where node_id='$n' and card!='$control'") || $error++;
} else {
# No reloads or reservation changes, so really free the node
# This little sillyness is for disk reloading. Remember the last
# project a node was reserved into.
$cmd = "insert into last_reservation values ('$n', '$pid')";
$dbh->query($cmd);
print "Releasing node '$n'...";
$cmd = "delete from reserved where node_id='$n' and eid='$eid'";
$sth = $dbh->query($cmd) && print "Succeeded.\n"
|| (print "Failed Command:\n$cmd\nError string is:".$dbh->errstr."\n"
&& $error++);
# Shark hack
# it is a shark, so clear out the alias(es)
DBQueryWarn("update interfaces set IPalias='' ".
"where node_id='$n'") || $error++;
# End shark hack
}
}
# Find the control net interface for this node type
$sth =
$dbh->query("select control_net,osid,node_types.pxe_boot_path " .
"from node_types " .
"left join nodes on nodes.type=node_types.type " .
"where node_id='$n'");
my @row= $sth->fetchrow_array();
my $control= $row[0];
my $osid= $row[1];
my $pxe_boot_path = $row[2];
if (! ($n =~ /sh\d+/)) { #If its not a shark
# Clean out all IPs except the control net
$cmd =
"update interfaces set IP='' where node_id='$n' and card!='$control'";
$sth = $dbh->query($cmd)
|| (print "Failed Command:\n$cmd\nError string is:".$dbh->errstr."\n"
&& $error++);
} else {
# it is a shark, so clear out the alias(es)
$cmd = "update interfaces set IPalias='' where node_id='$n'";
$sth = $dbh->query($cmd)
|| (print "Failed Command:\n$cmd\nError string is:".$dbh->errstr."\n"
&& $error++);
}
# Clean out all delays
$cmd =
"delete from delays where node_id='$n'";
$sth = $dbh->query($cmd)
|| (print "Failed Command:\n$cmd\nError string is:".$dbh->errstr."\n"
&& $error++);
# And clean out various tidbits from the nodes table.
$cmd = "update nodes set def_boot_osid='$osid', def_boot_cmd_line='',".
"def_boot_path='',startupcmd='',rpms='',deltas='',tarballs='',".
"pxe_boot_path='$pxe_boot_path', next_pxe_boot_path='' ".
"where node_id='$n'";
$sth = $dbh->query($cmd)
|| (print "Failed Command:\n$cmd\nError string is:".$dbh->errstr."\n"
&& $error++);
# Clean out the current_reloads table
$cmd = "delete from current_reloads where node_id='$n'";
$sth = $dbh->query($cmd)
|| (print "Failed Command:\n$cmd\nError string is:".$dbh->errstr."\n"
&& $error++);
#
# Clean out all delays
#
DBQueryWarn("delete from delays where node_id='$n'") || $error++;
#
# And clean out various tidbits from the nodes table.
#
DBQueryWarn("update nodes set def_boot_osid='$osid', def_boot_cmd_line='',".
"def_boot_path='',startupcmd='',rpms='',deltas='',tarballs='',".
"pxe_boot_path='$pxe_boot_path', next_pxe_boot_path='' ".
"where node_id='$n'") || $error++;
#
# Clean out the current_reloads table
#
DBQueryWarn("delete from current_reloads where node_id='$n'") || $error++;
}
$cmd = "unlock tables";
$sth = $dbh->query($cmd)
|| die("Locking error:\n$cmd\nError string is:".$dbh->errstr."\n");
DBQueryFatal("unlock tables");
if ( (@reloads > 0) || (@frisbee_reloads > 0) || (keys %reserves > 0) ) {
print "Locking tables.\n";
$cmd = "lock tables nodes read, node_types read, scheduled_reloads read, ".
"interfaces write, reserved write, next_reserve write";
$sth = $dbh->query($cmd)
|| die("Locking error:\n$cmd\nError string is:".$dbh->errstr."\n");
######################################################################
# Step 2 - Set up reserves and reloads
#
# If any nodes were awaiting reloads, put them in the proper
# experiment, and issue an os_load, if appropriate. If any nodes were
# set up for simple reservations, we take care of that in this stage
# too.
######################################################################
if ((@reloads > 0) || (keys %reserves > 0)) {
#
# Lock tables to maintain consistency
#
print "Locking tables.\n";
DBQueryFatal("lock tables nodes read, node_types read, scheduled_reloads read, ".
"interfaces write, reserved write, next_reserve write");
foreach $n ( @reloads ) {
# Change reservation (don't delete or we'll get races)
print "Changing reservation for $n to $reloadpid/$reload_pendingeid...\n";
$cmd = "update reserved set ".
"pid='$reloadpid',eid='$reload_pendingeid',vname=NULL ".
"where node_id='$n'";
$sth = $dbh->query($cmd)
|| (print "Failed Command:\n$cmd\nError string is:".$dbh->errstr."\n"
&& $error++ && next);
}
#
# Take care of reloads by putting them into a special experiment,
# which is processed by the batch_daemon
#
foreach my $n ( @reloads ) {
#
# Change reservation (don't delete or we'll get races)
#
print "Changing reservation for $n to $reloadpid/$reload_pendingeid...\n";
DBQueryWarn("update reserved set ".
"pid='$reloadpid',eid='$reload_pendingeid',vname=NULL ".
"where node_id='$n'") || $error++;
}
foreach $aref ( @frisbee_reloads ) {
my $n = $$aref[0];
my $image = $$aref[1];
# Change reservation (don't delete or we'll get races)
print "Changing reservation for $n to $reloadpid/$reloadeid...\n";
$cmd = "update reserved set pid='$reloadpid',eid='$reloadeid',vname=NULL ".
"where node_id='$n'";
$sth = $dbh->query($cmd)
|| (print "Failed Command:\n$cmd\nError string is:".$dbh->errstr."\n"
&& $error++ && next);
# NOTE: os_load for frisbee nodes gets run later, because the tables
# need to be unlocked for os_load to run
}
foreach $n ( keys %reserves ) {
my ($next_pid,$next_eid) = split (":",$reserves{$n});
# Change reservation (don't delete or we'll get races)
print "Changing reservation for $n to $next_pid/$next_eid...\n";
$cmd = "update reserved set pid='$next_pid',eid='$next_eid',vname=NULL ".
"where node_id='$n'";
$sth = $dbh->query($cmd)
|| (print "Failed Command:\n$cmd\nError string is:".$dbh->errstr."\n"
&& $error++ && next);
$cmd = "delete from next_reserve where node_id='$n'";
$sth = $dbh->query($cmd)
|| (print "Failed Command:\n$cmd\nError string is:".$dbh->errstr."\n"
&& $error++ && next);
}
print "Unlocking tables.\n";
$cmd = "unlock tables";
$sth = $dbh->query($cmd)
|| die("Locking error:\n$cmd\nError string is:".$dbh->errstr."\n");
# Continuation of Frisbee code... now that the tables are unlocked, we can
# call os_load
foreach $aref ( @frisbee_reloads ) {
my $n = $$aref[0];
my $image = $$aref[1];
# Call os_load for the machine, since the reload daemon doesn't handle
# frisbee reloads
print "Running os_load for $n\n";
system "$os_load -i $image $n"
|| (print "Failed to run '$os_load $image $n': $!\n"
&& $error++ && next);
}
#
# Handle scheduled reservations
#
foreach my $n ( keys %reserves ) {
my ($next_pid,$next_eid) = split (":",$reserves{$n});
#
# Change reservation (don't delete or we'll get races)
#
print "Changing reservation for $n to $next_pid/$next_eid...\n";
DBQueryWarn("update reserved set pid='$next_pid',eid='$next_eid'," .
"vname=NULL where node_id='$n'") || $error++;
DBQueryWarn("delete from next_reserve where node_id='$n'") || $error++;
}
#
# Done, so we can now unlock tables
#
print "Unlocking tables.\n";
DBQueryFatal("unlock tables");
}
######################################################################
# Step 3 - Set up console for freed nodes.
#
# Using a list of freed nodes build eariler, run consetup to reset
# their serial consoles.
######################################################################
if (@freed_nodes) {
my @conlist=();
my @sharks=();
foreach $n ( @freed_nodes ) {
foreach my $n ( @freed_nodes ) {
# Shark hack
if ($n =~ /(sh\d+)/) {
# Its a shark - do the shelf if it hasn't been done yet
my $shelf = $1;
......@@ -278,12 +277,17 @@ if (@freed_nodes) {
push(@conlist,$shelf);
}
}
# End shark hack
else {
push(@conlist,$n);
}
}
#
# Finally, run the actual command
#
system("$consetup @conlist") == 0 or
print STDERR "WARNING: $consetup @conlist failed!\n";
print STDERR "WARNING: $consetup @conlist failed!\n";
}
exit($error);
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment