From d1066678b367942a74ab8c93e65f7bd29b7e3a92 Mon Sep 17 00:00:00 2001
From: Simon Rettberg
Date: Fri, 10 Jan 2014 15:39:54 +0100
Subject: [smartctl] Also parse health status reported by drive

---
 .../smartctl/data/opt/openslx/scripts/systemd-smartctl   | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

(limited to 'remote/modules/smartctl/data/opt/openslx/scripts/systemd-smartctl')

diff --git a/remote/modules/smartctl/data/opt/openslx/scripts/systemd-smartctl b/remote/modules/smartctl/data/opt/openslx/scripts/systemd-smartctl
index 6025c962..6a798563 100755
--- a/remote/modules/smartctl/data/opt/openslx/scripts/systemd-smartctl
+++ b/remote/modules/smartctl/data/opt/openslx/scripts/systemd-smartctl
@@ -2,18 +2,30 @@
 
 # Check Reallocated_Sector_Ct and Spin_Retry_Count of local disk(s)
 
+. /opt/openslx/config
+
+[ -z "$SLX_SMARTCTL_MIN_REALLOC" ] && SLX_SMARTCTL_MIN_REALLOC=0
+
 FILE=$(mktemp)
 for dev in /dev/sd?; do
-	smartctl -A "$dev" -f "brief" > "$FILE" || continue # should we report devices where smartctl doesn't work?
+	smartctl -H -A -f "brief" "$dev" > "$FILE" || continue # should we report devices where smartctl doesn't work?
+	# parse
+	OVERALL=$(grep -o "test result: .*$" "$FILE" | cut -c 14-)
+	[ "x$OVERALL" = "xPASSED" ] && OVERALL=""
 	REALLOC=$(grep "^ *5 " "$FILE" | awk '{print $8}')
 	SPINRETRY_VAL=$(grep "^ *10 " "$FILE" | awk '{print $4}')
 	SPINRETRY_THR=$(grep "^ *10 " "$FILE" | awk '{print $6}')
-	if [ -n "$REALLOC" ] && [ "$REALLOC" != "0" ]; then
+	# report if applicable
+	if [ -n "$OVERALL" ]; then
+		slxlog "smartctl-fail" "Failed HDD: $dev reports health as $OVERALL" "$FILE"
+	fi
+	if [ -n "$REALLOC" ] && [ "$REALLOC" -gt "$SLX_SMARTCTL_MIN_REALLOC" ]; then
 		slxlog "smartctl-realloc" "Failing HDD: $dev has $REALLOC reallocated sectors!" "$FILE"
 	fi
 	if [ -n "$SPINRETRY_VAL" ] && [ "$SPINRETRY_VAL" -le "$SPINRETRY_THR" ]; then
 		slxlog "smartctl-spinretry" "Failing HDD: $dev has bad spin retry count! ($SPINRETRY_VAL/$SPINRETRY_THR)" "$FILE"
 	fi
 done
+sleep 2 # give slxlog a little time, as it's running async
 rm -f -- "$FILE"
 
-- 
cgit v1.2.3-55-g7522