Fence na
NOTE: The comments in this file need to be update, please don't trust them.
This is the core fence agent that exists in /sbin/.
#!/usr/bin/perl
#
# Node Assassin - Fence Agent
# Digimer; digimer@alteeve.com
# Mar. 05, 2010.
# Version: 0.1.003
#
# Bugs;
# - None known, many expected
#
=pod
NAOS notes:
Assign two ports per node. Have states '0' and '1' act on port 1 (odd-numbered
ports) and have states '2' and '3' act on port 2 (even-numbered ports). For
example, have:
01:0 fence port 1
01:1 release fence on port 1
01:2 pulse port 2 for one second.
01:3 pulse port 2 for five seconds.
02:0 fence port 3
02:1 release fence on port 3
02:2 pulse port 4 for one second.
02:3 pulse port 4 for five seconds.
...
08:0 fence port 15
08:1 release fence on port 15
08:2 pulse port 16 for one second.
08:3 pulse port 16 for five seconds.
Fence Agent notes:
- 'fenced' issues the call to a fence agent.
- Commands come in from STDIN when called by 'fenced' or 'fence_node'.
- Commands are passed in as 'var=val' pairs with lines starting with '#' to
be ignored. The example they give is:
- argument=value
#this line is ignored
argument2=value2
- Each argument is on a new line.
- The argument and value may contain spaces or other characters invalid in
XML.
- No spaces are allowed on either side of the '='.
- Arguments not recognized by the fence agent should simply be ignored.
- Technically, the argument's value can have an '=' sign, so be sure to split
on the first equal sign only. However, this is highly discouraged.
Given the following:
<cluster name="an_san" config_version="1">
<clusternodes>
<clusternode name="an_san01.alteeve.com" nodeid="1">
<fence>
<method name="node_assassin">
<device name="ariel" port="01" action="off"/>
</method>
</fence>
</clusternode>
<clusternode name="an_san02.alteeve.com" nodeid="2">
<fence>
<method name="node_assassin">
<device name="ariel" port="02" action="off"/>
</method>
</fence>
</clusternode>
</clusternodes>
<fencedevices>
<fencedevice name="node_assassin" agent="fence_na" ipaddr="ariel.alteeve.com" name="ariel" passwd="gr0tt0"></fencedevice>
</fencedevices>
</cluster>
Questions:
- Is there a corelation between 'clusternode -> name', 'device -> name' and
'fencedevice -> name'? Which is used when sending 'name' to the fence agent?
'fencedevice'?
When 'fenced' decides to fence "an_san01.alteeve.com", it will:
- call '/sbin/fence_na' because of the 'fencedevices -> agent' value.
- It will pass the following arguments to the fence agent, one pair per line:
agent=fence_na # From 'fencedevices -> agent'
name=ariel # From 'fencedevices -> name'
ipaddr=ariel.alteeve.com # From 'fencedevices -> ipaddr'
passwd=gr0tt0 # From 'fencedevices -> passwd'
port=01 # From 'clusternode "an_san01.alteeve.com" -> port'
action=fence_na # From 'clusternode "an_san01.alteeve.com" -> option'
# This must be 'on', 'off', 'reboot', 'status'
# or 'monitor'. See below for how these terms
# are interpretted by this agent.
# NOTE: If 'option' is passed, it's value will
# be stored in 'action'. That is, 'action' and
# 'option' are synonymous.
#
- Node Assassin's implentation of options.
- 'off' This sets the node to state '0' on the reset port followed by
state '3' to the power port. State 0 is maintained to prevent
a reboot.
- 'on' This sets the node to state '1' on the reset port followed by
state '2' on the power port to boot the node.
- 'reboot' This sets the node to state '2' on the reset port to quickly
kill the node, then switches to state '3' on the power port,
checks the return value (later, will check the probe pin),
sets state '1' on the reset port, pauses 1 second, and then
sets state '2' on the power port to boot the node.
- 'status' This calls '00:0' and returns the state of the port. Later,
this will return the value from the voltage sensing pin.
- 'monitor' being a multi-port fence device, this should call 'list'.
MADI: Confirm that this is what is meant in "Issues" here:
http://sources.redhat.com/cluster/wiki/FenceAgentAPI
- 'list' No info on this
Command Line Arguments:
- Any command line arguments used by this fence agent are not dictated by the
Fence Agent API. By convention only, the following command line options are
used:
-a <ip> # Maps the value to 'ipaddr'.
-h # Print the help message and then exits.
-l <name> # Maps the value to 'name'.
-n <num> # Maps the value to 'port'.
-o <string> # Maps the value to 'action'.
-p <string> # Maps the value to 'passwd'.
-S <path> # Maps the value to 'passwd_script'. This is not used by Node
# Assassin yet and is simply ignored.
-q # Sets quiet mode. Only errors will be printed. Logging
# proceeds as normal
-V # Prints the 'fence_na' version and the version of any attached
# Node Assassin(s) and exits.
Note:
- For now, I will return '0' if the command succeeded, but will add a detection
line checks if there is voltage from the node's PSU later.
=cut
# Play safe!
use strict;
use warnings;
# Load my library.
require '/etc/na/fence_na.lib';
# This is how I talk.
use IO::Handle;
use Net::Telnet;
# This will be read in from a config file later.
my $conf={
'system' => {
max_valid_state => 3,
conf_file => "/etc/na/fence_na.conf",
quiet => "",
version => "",
list_state => "",
list => "",
monitor => "",
node_assassin_id=> 0,
},
node => {
ipaddr => "",
tcp_port => "",
port => "238",
login => "",
passwd => "",
port => "",
set_state => "",
passwd_script => "",
action => "",
agent => "", # This is only used by 'fenced'
handle => "",
set_state => [], # This anon array will store the states to set based on the action passed for the proper ports.
}
};
# This method can't pass in the '$log' handle, obviously, as it does not yet
# exist.
read_conf($conf);
# Log file for output.
my $log=IO::Handle->new();
open ($log, ">$conf->{'system'}{'log'}") || die "Failed to open: [$conf->{'system'}{'log'}] for writing; Error: $!\n";
# If this gets set in the next two function, the agent will exit.
my $bad=0;
# Read in arguments from the command line.
($bad)=read_cla($conf, $log, $bad);
# Now read in arguments from STDIN, which is how 'fenced' passes arguments.
($bad)=read_stdin($conf, $log, $bad);
# This makes sure the node ID is zero-padded or '00'.
$conf->{node}{port}=$conf->{node}{port} ? $conf->{node}{port}=sprintf("%02d", $conf->{node}{port}) : "00";
# Find the TCP port from the config file.
foreach my $i (1..$conf->{'system'}{nodes})
{
if ((lc($conf->{node}{$i}{ipaddr}) eq lc($conf->{node}{ipaddr})))
{
$conf->{'system'}{node_assassin_id}=$i;
$conf->{node}{tcp_port}=$conf->{node}{$i}{tcp_port};
last;
}
}
die "Exiting on errors.\n" if $bad;
record($conf, $log, "Node Assassin: [$conf->{node}{ipaddr}].\n");
record($conf, $log, "TCP Port: .... [$conf->{node}{tcp_port}].\n");
record($conf, $log, "Port: ........ [$conf->{node}{port}].\n");
record($conf, $log, "Login: ....... [$conf->{node}{login}].\n");
record($conf, $log, "Password: .... [$conf->{node}{passwd}].\n");
record($conf, $log, "Action: ...... [$conf->{node}{action}].\n");
record($conf, $log, "Done reading args.\n");
# Connect to the Node Assassin.
$conf->{node}{handle}=new Net::Telnet(
Timeout => 10,
Errmode => 'die',
Port => $conf->{node}{tcp_port},
Prompt => '/EOM$/',
);
# print "Handle: [$conf->{node}{handle}]\n";
$conf->{node}{handle}->open($conf->{node}{ipaddr});
# Validate credentials.
# NOTE: Checking before the telnet fails on the exit. Also, this will be moved
# into the Node Assassin soon anyway.
if (($conf->{node}{login} ne $conf->{'system'}{username}) or ($conf->{node}{passwd} ne $conf->{'system'}{password}))
{
record($conf, $log, "Username and/or password failed.\n");
do_exit($conf, $log, 8);
}
###############################################################################
# What do? #
###############################################################################
# If I've been asked to show the version information, do so and then exit.
record($conf, $log, "Version: ..... [$conf->{'system'}{version}].\n");
if ($conf->{'system'}{version})
{
version($conf, $log);
do_exit($conf, $log, 0);
}
# If I've been asked to show the info on the given node assassin, do so and
# then exit.
record($conf, $log, "List State: .. [$conf->{'system'}{list_state}].\n");
if ($conf->{'system'}{list_state})
{
show_state($conf, $log);
do_exit($conf, $log, 0);
}
# When asked to 'monitor' or 'list', do this... whatever 'this' is. All I know
# is that it should not generate output.
record($conf, $log, "Monitor: ..... [$conf->{'system'}{monitor}].\n");
record($conf, $log, "List: ........ [$conf->{'system'}{list}].\n");
if (($conf->{node}{monitor}) or ($conf->{node}{list}))
{
show_list($conf, $log);
do_exit($conf, $log, 0);
}
# If I made it this far, I am setting a state. Sort out what state from the
# values in my conf->{node} hash.
record ($conf, $log, "Setting node: [$conf->{node}{port}] to action: [$conf->{node}{action}] using the Node Assassin: [$conf->{node}{ipaddr}] using the login: [$conf->{node}{login}/$conf->{node}{passwd}]\n");
# Convert the action into Node Assassin protocol arguments.
process_action($conf, $log);
# Process the orders.
foreach my $order (split/,/, $conf->{'system'}{call_order})
{
if ($order=~/^sleep/)
{
my $time=$order=~/sleep (\d+)/ ? $1 : 1;
record ($conf, $log, "Sleeping for: [$time]...\n");
sleep $time;
next;
}
record ($conf, $log, "Calling: [$order]\n");
my @set_state=$conf->{node}{handle}->cmd("$order");
foreach my $line (@set_state)
{
record($conf, $log, $line);
}
record($conf, $log, "Call complete.\n");
}
record($conf, $log, "All calls complete, exiting.\n");
# Cleanup and exit.
do_exit($conf, $log, 0);