mirror of
https://github.com/xcat2/xcat-core.git
synced 2025-10-31 11:22:27 +00:00
added node status monitoring using RMC
git-svn-id: https://svn.code.sf.net/p/xcat/code/xcat-core/trunk@1731 8638fb3e-16cb-4fca-ae20-7b5d299a9bcd
This commit is contained in:
linggao
@@ -12,6 +12,7 @@ use Socket;
|
||||
use xCAT::Utils;
|
||||
use xCAT::GlobalDef;
|
||||
use xCAT_monitoring::monitorctrl;
|
||||
use xCAT_monitoring::xcatmon;
|
||||
use xCAT::MsgUtils;
|
||||
|
||||
#print "xCAT_monitoring::rmcmon loaded\n";
|
||||
@@ -60,12 +61,13 @@ sub start {
|
||||
if (! -e "/usr/bin/lsrsrc") {
|
||||
return (1, "RSCT is not installed.\n");
|
||||
}
|
||||
my $result=`/usr/bin/lssrc -s ctrmc`; #TODO: change this
|
||||
if ($result !~ /active/) {
|
||||
|
||||
chomp(my $pid= `/bin/ps -ef | /bin/grep rmcd | /bin/grep -v grep | /bin/awk '{print \$2}'`);
|
||||
unless($pid){
|
||||
#restart rmc daemon
|
||||
$result=`startsrc -s ctrmc`;
|
||||
if ($?) {
|
||||
return (1, "rmc deamon cannot be started\n");
|
||||
return (1, "RMC deamon cannot be started\n");
|
||||
}
|
||||
}
|
||||
|
||||
@@ -255,10 +257,135 @@ sub supportNodeStatusMon {
|
||||
#--------------------------------------------------------------------------------
|
||||
sub startNodeStatusMon {
|
||||
#print "rmcmon::startNodeStatusMon called\n";
|
||||
return (0, "started");
|
||||
my $retcode=0;
|
||||
my $retmsg="started";
|
||||
my $isSV=xCAT::Utils->isServiceNode();
|
||||
if ($isSV) { return ($retcode, $retmsg); }
|
||||
|
||||
#get all the nodes status from IBM.MngNode class of local host and
|
||||
#the identification of this node
|
||||
my $noderef=xCAT_monitoring::monitorctrl->getMonHierarchy();
|
||||
my @hostinfo=xCAT::Utils->determinehostname();
|
||||
%iphash=();
|
||||
foreach(@hostinfo) {$iphash{$_}=1;}
|
||||
if (!$isSV) { $iphash{'noservicenode'}=1;}
|
||||
|
||||
my @servicenodes=();
|
||||
my %status_hash=();
|
||||
foreach my $key (keys (%$noderef)) {
|
||||
my @key_a=split(',', $key);
|
||||
if (! $iphash{$key_a[0]}) { push @servicenodes, $key_a[0]; }
|
||||
my $mon_nodes=$monservers->{$key};
|
||||
foreach(@$mon_nodes) {
|
||||
my $node_info=$_;
|
||||
$status_hash{$node_info->[0]}=$node_info->[2];
|
||||
}
|
||||
}
|
||||
|
||||
#get nodestatus from RMC and update the xCAT DB
|
||||
($retcode, $retmsg) = saveRMCNodeStatusToxCAT(\%status_hash);
|
||||
if ($retcode != 0) {
|
||||
$retmsg="Error occurred while updating xCAT node status from RMC data.:$retmsg";
|
||||
xCAT::MsgUtils->message('SI', "[mon]: $retmsg\n");
|
||||
}
|
||||
foreach (@servicenodes) {
|
||||
($retcode, $retmsg) = saveRMCNodeStatusToxCAT(\%status_hash, $_);
|
||||
if ($retcode != 0) {
|
||||
$retmsg="Error occurred while updating xCAT node status from RMC data from $_.:$retmsg";
|
||||
xCAT::MsgUtils->message('SI', "[mon]: $retmsg\n");
|
||||
}
|
||||
}
|
||||
|
||||
#start monitoring the status of mn's immediate children
|
||||
my $result=`startcondresp NodeReachability UpdatexCATNodeStatus 2>&1`;
|
||||
if ($?) {
|
||||
$retcode=$?;
|
||||
$retmsg="Error start node status monitoring: $result";
|
||||
xCAT::MsgUtils->message('SI', "[mon]: $retmsg\n");
|
||||
}
|
||||
|
||||
#start monitoring the status of mn's grandchildren via their service nodes
|
||||
$result=`startcondresp NodeReachability_H UpdatexCATNodeStatus 2>&1`;
|
||||
if ($?) {
|
||||
$retcode=$?;
|
||||
$retmsg="Error start node status monitoring: $result";
|
||||
xCAT::MsgUtils->message('SI', "[mon]: $retmsg\n");
|
||||
}
|
||||
|
||||
return ($retcode, $retmsg);
|
||||
}
|
||||
|
||||
|
||||
#--------------------------------------------------------------------------------
|
||||
=head3 saveRMCNodeStatusToxCAT
|
||||
This function gets RMC node status and save them to xCAT database
|
||||
|
||||
Arguments:
|
||||
$oldstatus a pointer to a hash table that has the current node status
|
||||
$node the name of the service node to run RMC command from. If null, get from local host.
|
||||
Returns:
|
||||
(return code, message)
|
||||
=cut
|
||||
#--------------------------------------------------------------------------------
|
||||
sub saveRMCNodeStatusToxCAT {
|
||||
#print "rmcmon::saveRMCNodeStatusToxCAT called\n";
|
||||
my $retcode=0;
|
||||
my $retmsg="started";
|
||||
my $statusref=shift;
|
||||
if ($statusref =~ /xCAT_monitoring::rmcmon/) {
|
||||
$statusref=shift;
|
||||
}
|
||||
my $node=shift;
|
||||
|
||||
%status_hash=%$statusref;
|
||||
|
||||
#get all the node status from mn's children
|
||||
my $result;
|
||||
if ($node) {
|
||||
$result=`lsrsrc-api -s IBM.MngNode::::Name::Status 2>&1`;
|
||||
} else {
|
||||
$result=`lsrsrc-api -s IBM.MngNode::::Name::Status 2>&1`;
|
||||
}
|
||||
if ($?) {
|
||||
$retcode=$?;
|
||||
$retmsg=$result;
|
||||
xCAT::MsgUtils->message('SI', "[mon]: Error getting node status from RMC: $result\n");
|
||||
return ($retcode, $retmsg);
|
||||
} else {
|
||||
my @active_nodes=();
|
||||
my @inactive_nodes=();
|
||||
if ($result) {
|
||||
my @lines=split('\n', $result);
|
||||
#only save the ones that needs to change
|
||||
foreach (@lines) {
|
||||
@pairs=split('::', $_);
|
||||
if ($pairs[1]==1) {
|
||||
if ($status_hash{$pairs[0]} ne $::STATUS_ACTIVE) { push @active_nodes,$pairs[0];}
|
||||
}
|
||||
else {
|
||||
if ($status_hash{$pairs[0]} ne $::STATUS_INACTIVE) { push @inactive_nodes, $pairs[0];}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
my %new_node_status=();
|
||||
if (@active_nodes>0) {
|
||||
$new_node_status{$::STATUS_ACTIVE}=\@active_nodes;
|
||||
}
|
||||
if (@inactive_nodes>0) {
|
||||
$new_node_status{$::STATUS_INACTIVE}=\@inactive_nodes;
|
||||
}
|
||||
#only set the node status for the changed ones
|
||||
if (keys(%new_node_status) > 0) {
|
||||
xCAT_monitoring::xcatmon::processNodeStatusChanges(\%new_node_status);
|
||||
}
|
||||
return ($retcode, $retmsg);
|
||||
}
|
||||
|
||||
|
||||
|
||||
|
||||
#--------------------------------------------------------------------------------
|
||||
=head3 stopNodeStatusMon
|
||||
This function is called by the monitorctrl module to tell
|
||||
@@ -273,7 +400,27 @@ sub startNodeStatusMon {
|
||||
#--------------------------------------------------------------------------------
|
||||
sub stopNodeStatusMon {
|
||||
#print "rmcmon::stopNodeStatusMon called\n";
|
||||
return (0, "stopped");
|
||||
my $retcode=0;
|
||||
my $retmsg="stopped";
|
||||
my $isSV=xCAT::Utils->isServiceNode();
|
||||
if ($isSV) { return ($retcode, $retmsg); }
|
||||
|
||||
#stop monitoring the status of mn's immediate children
|
||||
my $result=`stopcondresp NodeReachability UpdatexCATNodeStatus 2>&1`;
|
||||
if ($?) {
|
||||
$retcode=$?;
|
||||
$retmsg="Error stop node status monitoring: $result";
|
||||
xCAT::MsgUtils->message('SI', "[mon]: $retmsg\n");
|
||||
}
|
||||
|
||||
#stop monitoring the status of mn's grandchildren via their service nodes
|
||||
$result=`stopcondresp NodeReachability_H UpdatexCATNodeStatus 2>&1`;
|
||||
if ($?) {
|
||||
$retcode=$?;
|
||||
$retmsg="Error stop node status monitoring: $result";
|
||||
xCAT::MsgUtils->message('SI', "[mon]: $retmsg\n");
|
||||
}
|
||||
return ($retcode, $retmsg);
|
||||
}
|
||||
|
||||
|
||||
|
||||
@@ -0,0 +1,8 @@
|
||||
#!/usr/bin/perl
|
||||
|
||||
$RES::EventResponse{'UpdatexCATNodeStatus'} = {
|
||||
Name => q(UpdatexCATNodeStatus),
|
||||
Locked => q(0),
|
||||
Actions => q({[updatexCAT,{127},{0},{86400},/opt/xcat/sbin/rmcmon/updatexcatnodestatus,3,0,0,0,{},0]}),
|
||||
};
|
||||
1;
|
||||
67
xCAT-rmc/scripts/updatexcatnodestatus
Executable file
67
xCAT-rmc/scripts/updatexcatnodestatus
Executable file
@@ -0,0 +1,67 @@
|
||||
#!/usr/bin/env perl
|
||||
# IBM(c) 2007 EPL license http://www.eclipse.org/legal/epl-v10.html
|
||||
|
||||
# This script is used by UpdatexCATNodeStatus event response to handle the node
|
||||
# status changes from the condition NodeReachability and NodeReachability_H
|
||||
|
||||
|
||||
use strict;
|
||||
use Getopt::Std;
|
||||
use POSIX qw(strftime);
|
||||
|
||||
my $cond_name=$ENV{ERRM_COND_NAME};
|
||||
my $node;
|
||||
my $status;
|
||||
|
||||
if ($cond_name eq "NodeReachability") {
|
||||
$node=$ENV{ERRM_RSRC_NAME};
|
||||
$status=$ENV{ERRM_VALUE};
|
||||
} elsif ($cond_name eq "NodeReachability_H") {
|
||||
# Parse the ERRM_VALUE attribute, which will contain the
|
||||
# LastEvent structured data variable from the Condition class
|
||||
# The fields in this structured data variable are documented below where we parse them out.
|
||||
my $event = $ENV{ERRM_VALUE};
|
||||
$event =~ s/^\[(.*)\]$/$1/; # SD variables have square brackets around them
|
||||
|
||||
# This parse the LastEvent
|
||||
my ( # split the SD into the following fields:
|
||||
$Occurred, # One if the condition has been triggered
|
||||
$ErrNum, # Non-zero if there was in error in the event registration
|
||||
$ErrMsg, # The string msg related to ErrNum
|
||||
$EventFlags, # Bit mask giving some additional info about the event
|
||||
$EventTime, # Time of event expressed in seconds since 1/1/1970
|
||||
$EventTimeMicros, # Number of microseconds past EventTime
|
||||
$ResourceHandle, # Binary address of the RMC resource that caused the condition to be triggered
|
||||
$NodeName, # The node on which the event occurred. For conditions that use the management domain scope (4),
|
||||
# this will be the leaf node. For conditions that use the local scope (e.g. NodeReachability),
|
||||
# this will be the FMS.
|
||||
$NumAttrs, # Number of attr values from the resource returned in this event
|
||||
$NumAttrsInExpr, # How many of the above were attributes in the event expression
|
||||
$IndexForAttrs, # The starting index of the array of values. Until new fixed fields are added
|
||||
# to LastEvent, this will be the element right after this one.
|
||||
$AttrArray # This list of attribute names, types, and values
|
||||
) = split(/,/, $event, 12);
|
||||
|
||||
my @attrArray = split(/,/, $AttrArray); # Note: parsing this way does not support SDs or SD Arrays that may be in this list
|
||||
|
||||
my $j = 0; # index into attrArray
|
||||
for (my $i=0; $i<$NumAttrs; $i++) {
|
||||
my $attrName = $attrArray[$j++];
|
||||
my $attrType = $attrArray[$j++]; # Types <= 8 are "simple" types. Types > 8 are SDs and arrays.
|
||||
my $attrValue = $attrArray[$j++];
|
||||
if ($attrName eq '"Name"') { $node = $attrValue; }
|
||||
if ($attrName eq '"Status"') { $status = $attrValue; }
|
||||
}
|
||||
} else {
|
||||
print "this script does not handle condition $cond_name\n";
|
||||
exit 1;
|
||||
}
|
||||
|
||||
|
||||
open(CMD, "| wall") || die "Error: can not start wall command.\n";
|
||||
print CMD "node=$node status=$status\n";
|
||||
close(CMD);
|
||||
|
||||
exit 0
|
||||
|
||||
|
||||
Reference in New Issue
Block a user