From: David Galloway <dgallowa@redhat.com>
Date: Wed, 25 Nov 2020 15:23:00 +0000 (-0500)
Subject: common: smart.sh - print serial numbers of bad drives
X-Git-Url: http://git.apps.os.sepia.ceph.com/?a=commitdiff_plain;h=72e424645df4453a2dcb7f8f34174c67a86ba56b;p=ceph-cm-ansible.git

common: smart.sh - print serial numbers of bad drives

Signed-off-by: David Galloway <dgallowa@redhat.com>
---

diff --git a/roles/common/files/libexec/smart.sh b/roles/common/files/libexec/smart.sh
index 2287a1d..09a3275 100755
--- a/roles/common/files/libexec/smart.sh
+++ b/roles/common/files/libexec/smart.sh
@@ -264,7 +264,7 @@ normal_smart ()
         fi
       # The SSDs in the bruuni just straight up say failed with no additional detail
       elif sudo smartctl -a /dev/$l | grep -q "FAILED!"; then
-        messages+=("Drive $l has completely failed")
+        messages+=("Drive $l ($(get_serial $l)) has completely failed")
         failed=true
         rc=2
       else
@@ -277,7 +277,7 @@ normal_smart ()
     if echo "$output" | grep -q '^0x05'; then
       reallocated=$(echo "$output" | grep '^0x05' | awk '{print $NF}')
       if [ "$reallocated" != "0" ] && [ $is_ssd = false ]; then
-        messages+=("Drive $l has $reallocated reallocated sectors")
+        messages+=("Drive $l ($(get_serial $l)) has $reallocated reallocated sectors")
         failed=true
         # A small number of reallocated sectors is OK
 	# Don't set rc to WARN if we were already CRIT from previous drive
@@ -293,7 +293,7 @@ normal_smart ()
     if echo "$output" | grep -q '^0xbb'; then
       uncorrect=$(echo "$output" | grep '^0xbb' | awk '{print $NF}')
       if [ "$uncorrect" != "0" ]; then
-        messages+=("Drive $l has $uncorrect reported uncorrect sectors")
+        messages+=("Drive $l ($(get_serial $l)) had $uncorrect reported uncorrect sectors")
         failed=true
         rc=2
       fi
@@ -302,7 +302,7 @@ normal_smart ()
     if echo "$output" | grep -q '^0xc4'; then
       reallocatedevents=$(echo "$output" | grep '^0xc4' | awk '{print $NF}')
       if [ "$reallocatedevents" != "0" ]; then
-        messages+=("Drive $l has $reallocatedevents reallocated events")
+        messages+=("Drive $l ($(get_serial $l)) has $reallocatedevents reallocated events")
         failed=true
         rc=2
       fi
@@ -311,7 +311,7 @@ normal_smart ()
     if echo "$output" | grep -q '^0xc5'; then
       pending=$(echo "$output" | grep '^0xc5' | awk '{print $NF}')
       if [ "$pending" != "0" ]; then
-        messages+=("Drive $l has $pending pending sectors")
+        messages+=("Drive $l ($(get_serial $l)) has $pending pending sectors")
         failed=true
         rc=2
       fi
@@ -320,7 +320,7 @@ normal_smart ()
     if echo "$output" | grep -q '^0xc6'; then
       uncorrect=$(echo "$output" | grep '^0xc6' | awk '{print $NF}')
       if [ "$uncorrect" != "0" ]; then
-        messages+=("Drive $l has $uncorrect uncorrect sectors")
+        messages+=("Drive $l ($(get_serial $l)) has $uncorrect uncorrect sectors")
         failed=true
         rc=2
       fi
@@ -329,7 +329,7 @@ normal_smart ()
     if echo -e "$output" | grep -q '^0xe9'; then
       wearout=$(echo "$output" | grep '^0xe9' | awk '{print $NF}')
       if [ "$wearout" == "1" ]; then
-        messages+=("Drive $l has exhausted its Media_Wearout_Indicator")
+        messages+=("Drive $l ($(get_serial $l)) has exhausted its Media_Wearout_Indicator")
         failed=true
 	# Don't set rc to WARN if we were already CRIT from previous drive
         if [ "$rc" != 2 ]
@@ -420,5 +420,14 @@ nvme_smart ()
   done
 }
 
+get_serial() {
+  serial=$(sudo smartctl -i /dev/$1 | grep "Serial Number:" | awk '{ print $3 }')
+  if [ "$serial" == "" ]; then
+    echo "S/N unknown"
+  else
+    echo $serial
+  fi
+}
+
 ## Call main() function
 main