power-management: add failed service checks

After running tests that cause the device to reboot, check for any failed services. Could be pertinent to the watchdog test where the shutdown is not manageed. 

5 files changed, 44 insertions, 0 deletions

diff --git a/bin/failed_service_check.sh b/bin/failed_service_check.sh
new file mode 100755
index 00000000..172f20a8
--- /dev/null
+++ b/bin/failed_service_check.sh
@@ -0,0 +1,11 @@
+#!/bin/bash
+
+COUNT=$(systemctl --system --no-ask-password --no-pager --no-legend list-units --state=failed | wc -l)
+printf "Found %s failed units\n" "$COUNT"
+if [ "$COUNT" -eq 0 ]; then
+ exit 0
+else
+ printf "\nFailed units:\n"
+ systemctl --system --no-ask-password --no-pager list-units --state=failed
+fi
+exit 1
diff --git a/units/power-management/jobs.pxu b/units/power-management/jobs.pxu
index 36e31c2a..a84f0045 100644
--- a/units/power-management/jobs.pxu
+++ b/units/power-management/jobs.pxu
@@ -290,6 +290,16 @@ user: root
 flags: preserve-locale noreturn autorestart
 estimated_duration: 180.0
 
+id: power-management/post-warm-reboot
+after: power-management/warm-reboot
+category_id: com.canonical.plainbox::power-management
+_summary: Post warm reboot service check
+_description: Check there are no failed services after the warm reboot
+unit: job
+plugin: shell
+command: failed_service_check.sh
+estimated_duration: 1.0
+
 id: power-management/cold-reboot
 category_id: com.canonical.plainbox::power-management
 _summary: Cold reboot
@@ -307,6 +317,16 @@ user: root
 flags: preserve-locale noreturn autorestart
 estimated_duration: 300
 
+id: power-management/post-cold-reboot
+after: power-management/cold-reboot
+category_id: com.canonical.plainbox::power-management
+_summary: Post cold reboot service check
+_description: Check there are no failed services after the cold reboot
+unit: job
+plugin: shell
+command: failed_service_check.sh
+estimated_duration: 1.0
+
 unit: template
 template-resource: model_assertion
 template-unit: job
diff --git a/units/power-management/test-plan.pxu b/units/power-management/test-plan.pxu
index bb782c10..8391bca5 100644
--- a/units/power-management/test-plan.pxu
+++ b/units/power-management/test-plan.pxu
@@ -75,7 +75,9 @@ _name: Automated power tests
 _description: Automated power tests for Snappy Ubuntu Core devices
 include:
 power-management/warm-reboot
+ power-management/post-warm-reboot
 power-management/cold-reboot
+ power-management/post-cold-reboot
 
 id: power-manual
 unit: test plan
diff --git a/units/watchdog/jobs.pxu b/units/watchdog/jobs.pxu
index 2b5b236c..3247941e 100644
--- a/units/watchdog/jobs.pxu
+++ b/units/watchdog/jobs.pxu
@@ -65,3 +65,13 @@ user: root
 plugin: shell
 category_id: com.canonical.plainbox::power-management
 estimated_duration: 60
+
+id: watchdog/post-trigger-system-reset-auto
+after: watchdog/trigger-system-reset-auto
+category_id: com.canonical.plainbox::power-management
+_summary: Post watchdog reset service check
+_description: Check there are no failed services after the watchdog triggered
+unit: job
+plugin: shell
+command: failed_service_check.sh
+estimated_duration: 1.0
diff --git a/units/watchdog/test-plan.pxu b/units/watchdog/test-plan.pxu
index 407a2734..6de8e710 100644
--- a/units/watchdog/test-plan.pxu
+++ b/units/watchdog/test-plan.pxu
@@ -25,3 +25,4 @@ estimated_duration: 1s
 include:
 watchdog/systemd-config
 watchdog/trigger-system-reset-auto
+ watchdog/post-trigger-system-reset-auto