From cd21084e3c409a3c2416b0cfab4563380e6dbf62 Mon Sep 17 00:00:00 2001
From: Elias Ohm <eohm@novomind.com>
Date: Mon, 20 Jul 2020 02:39:00 +0200
Subject: [PATCH] spare disks in shared mode, optional aggregte spare-count
 spares cluster- or node wide, fix filter to query only disks from home-node

* add handling of spare disks in "shared" mode (partitioned or formated disks)
  these are logically not assigned to a specific nodes spare disk capacity so are shown at a _SHARED_ node and not shown in single-node mode
* add a suboption "aggregate" that allows to sum up/aggregate all disks in the cluster (if run against a specific "node" the "shared" mode disks are accounted for the node)
* fix query-building to find only disks for the specific home-node (the error was not visible before because it was post-filtered in result function)
* fix broken branching conditions within these function (see also #86)
---
 check_netapp_ontap.pl | 39 +++++++++++++++++++++++++++++----------
 1 file changed, 29 insertions(+), 10 deletions(-)

diff --git a/check_netapp_ontap.pl b/check_netapp_ontap.pl
index 42defd6..ec05249 100644
--- a/check_netapp_ontap.pl
+++ b/check_netapp_ontap.pl
@@ -174,19 +174,26 @@ sub calc_disk_health {
 ##############################################
 
 sub get_spare_info {
-	my ($nahStorage, $strVHost, $strWarning, $strCritical) = @_;
+	my ($nahStorage, $strVHost, $strWarning, $strCritical, $strSuboption) = @_;
 	my $nahSpareIterator = NaElement->new("storage-disk-get-iter");
 	my $nahQuery = NaElement->new("query");
 	my $nahSpareInfo = NaElement->new("storage-disk-info");
 	my $nahSpareOwnerInfo = NaElement->new("disk-ownership-info");
 	my $strActiveTag = "";
 	my %hshSpareInfo;
+	my $aggregateNode = undef;
+
+	if (defined($strSuboption)) {
+		my @arySuboption = split(",",$strSuboption);
+		if ("aggregate" ~~ @arySuboption) { $aggregateNode = '_ALL_'; }
+	}
 
 	if (defined($strVHost)) {
 		$nahSpareIterator->child_add($nahQuery);
 		$nahQuery->child_add($nahSpareInfo);
 		$nahSpareInfo->child_add($nahSpareOwnerInfo);
-		$nahSpareOwnerInfo->child_add_string("home-node", $strVHost);
+		$nahSpareOwnerInfo->child_add_string("home-node-name", $strVHost);
+		if (defined($aggregateNode)) { $aggregateNode = $strVHost; }
 	}
 
 	while(defined($strActiveTag)) {
@@ -218,8 +225,15 @@ sub get_spare_info {
 				my $spareInfo = $raidInfo->child_get("disk-spare-info");
 				my $zeroed = $spareInfo->child_get_string('is-zeroed');
 
-				$hshSpareInfo{$nodeName}{$strSpareName}{'status'} = $containertype;
-				$hshSpareInfo{$nodeName}{$strSpareName}{'zeroed'} = $zeroed;
+				$hshSpareInfo{$aggregateNode||$nodeName}{$strSpareName}{'status'} = $containertype;
+				$hshSpareInfo{$aggregateNode||$nodeName}{$strSpareName}{'zeroed'} = $zeroed;
+			} elsif ($containertype eq "shared") {
+				my $sharedInfo = $raidInfo->child_get("disk-shared-info");
+				my $aggregateList = $sharedInfo->child_get("aggregate-list");
+				next SPARE if (defined($aggregateList));
+
+				$hshSpareInfo{$aggregateNode||'_SHARED_'}{$strSpareName}{'status'} = $containertype;
+				$hshSpareInfo{$aggregateNode||'_SHARED_'}{$strSpareName}{'zeroed'} = undef;
 			} else {
 				next SPARE;
 			}
@@ -242,7 +256,7 @@ sub calc_spare_health {
 			next NODE;
 		}
 
-		my ($spareCount, $unassignedCount, $unknownCount, $notZeroedCount) = (0, 0, 0, 0);
+		my ($spareCount, $unassigneSpareCount, $sharedSpareCount, $unassignedCount, $unknownCount, $notZeroedCount) = (0, 0, 0, 0, 0, 0);
 		my $strNewMessage;
 
 		foreach my $strSpare (keys %{$hrefSpareInfo->{$node}}) {
@@ -253,14 +267,17 @@ sub calc_spare_health {
 				$notZeroedCount++;
 			}
 			my $status = $hrefSpareInfo->{$node}->{$strSpare}->{'status'};
-			if (defined($status && $status eq "spare")) {
-				$spareCount++;
-			} elsif (defined($status && $status eq "unassigned")) {
+			if (defined($status) && $status eq "spare") {
+				$unassigneSpareCount++;
+			} elsif (defined($status) && $status eq "unassigned") {
 				$unassignedCount++;
+			} elsif (defined($status) && $status eq "shared") {
+				$sharedSpareCount++;
 			} else {
 				$unknownCount++;
 			}
 		}
+		$spareCount = $unassigneSpareCount + $sharedSpareCount;
 
 		if ($spareCount < $strCritical) {
 			$critStatus++;
@@ -272,7 +289,7 @@ sub calc_spare_health {
 			$unknownStatus++;
 		}
 
-		$strNewMessage = sprintf("%s: %d spare disks (%s not zeroed) and %s unassigned", $node, $spareCount, $notZeroedCount, $unassignedCount);
+		$strNewMessage = sprintf("%s: %d spare (%d shared, %s not zeroed) and %s unassigned", $node, $spareCount, $sharedSpareCount, $notZeroedCount, $unassignedCount);
 		$strOutput = get_nagios_description($strOutput, $strNewMessage);
 	}
 
@@ -1968,6 +1985,8 @@ sub help {
 	desc: Check the number of spare disks
 	thresh: Warning / critical required spare disks. Default thresholds are 2 / 1.
 	node: The node option restricts this check by cluster-node name.
+             (The spare disks already assigned to shared pools do not belong to a specific nodes aggregate, thus are not shown then.)
+	suboption: aggregate (sums up all spare disks in a cluster or on a node, includes also the shared disks on their home-node)
 
 * For keyword thresholds, if you want to ignore alerts for that particular keyword you set it at the same threshold that the alert defaults to.
 
@@ -2362,7 +2381,7 @@ sub filter_object {
 	$strWarning  //= 2;
 	$strCritical //= 1;
 
-	my $hrefSpareInfo = get_spare_info($nahStorage, $strVHost, $strWarning, $strCritical);
+	my $hrefSpareInfo = get_spare_info($nahStorage, $strVHost, $strWarning, $strCritical, $strSuboption);
 
 	if (defined($strModifier)) {
 		$hrefSpareInfo = filter_object($hrefSpareInfo, $strModifier);