Merging upstream version 1.16.0.

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

19 files changed, 87 insertions, 126 deletions

diff --git a/tests/health_mgmtapi/README.md b/tests/health_mgmtapi/README.md
index 278c72dc1..8473b35ea 100644
--- a/tests/health_mgmtapi/README.md
+++ b/tests/health_mgmtapi/README.md
@@ -4,9 +4,8 @@ The directory `tests/health_cmdapi` contains the test script `health-cmdapi-test
 
 The script can be executed with options to prepare the system for the tests, run them and restore the system to its previous state. 
 
-It depends on the management API being accessible and on the responses to the api/v1/alarms?all requests being functional.
-
-Run it with `tests/health_mgmtapi/health-cmdapi-test.sh -h` to see the options. 
+It depends on the management API being accessible on localhost:19999 and on the responses to the api/v1/alarms?all requests being functional.
+It also requires read access to the management API key that is usually under `/var/lib/netdata/netdata.api.key` (`@varlibdir_POST@/netdata.api.key`).
 
 [![analytics](https://www.google-analytics.com/collect?v=1&aip=1&t=pageview&_s=1&ds=github&dr=https%3A%2F%2Fgithub.com%2Fnetdata%2Fnetdata&dl=https%3A%2F%2Fmy-netdata.io%2Fgithub%2Ftests%2Fhealth_mgmtapi%2FREADME&_u=MAC~&cid=5792dfd7-8dc4-476b-af31-da2fdb9f93d2&tid=UA-64295674-3)]()
 
diff --git a/tests/health_mgmtapi/expected_list/ALARM_CPU_IOWAIT-list.json b/tests/health_mgmtapi/expected_list/ALARM_CPU_IOWAIT-list.json
new file mode 100644
index 000000000..9f05efe70
--- /dev/null
+++ b/tests/health_mgmtapi/expected_list/ALARM_CPU_IOWAIT-list.json
@@ -0,0 +1 @@
+{ "all": false, "type": "SILENCE", "silencers": [ { "alarm": "*10min_cpu_iowait" }, { "alarm": "*10min_cpu_usage *load_trigger" } ] }
diff --git a/tests/health_mgmtapi/expected_list/ALARM_CPU_USAGE-list.json b/tests/health_mgmtapi/expected_list/ALARM_CPU_USAGE-list.json
new file mode 100644
index 000000000..dbf879925
--- /dev/null
+++ b/tests/health_mgmtapi/expected_list/ALARM_CPU_USAGE-list.json
@@ -0,0 +1 @@
+{ "all": false, "type": "SILENCE", "silencers": [ { "alarm": "*10min_cpu_usage *load_trigger", "context": "system.cpu" }, { "alarm": "*10min_cpu_usage *load_trigger", "chart": "system.load" } ] }
diff --git a/tests/health_mgmtapi/expected_list/CONTEXT_SYSTEM_CPU-list.json b/tests/health_mgmtapi/expected_list/CONTEXT_SYSTEM_CPU-list.json
new file mode 100644
index 000000000..a267cfd6f
--- /dev/null
+++ b/tests/health_mgmtapi/expected_list/CONTEXT_SYSTEM_CPU-list.json
@@ -0,0 +1 @@
+{ "all": false, "type": "DISABLE", "silencers": [ { "context": "system.cpu" }, { "chart": "system.load" } ] }
diff --git a/tests/health_mgmtapi/expected_list/DISABLE-list.json b/tests/health_mgmtapi/expected_list/DISABLE-list.json
new file mode 100644
index 000000000..c2c778104
--- /dev/null
+++ b/tests/health_mgmtapi/expected_list/DISABLE-list.json
@@ -0,0 +1 @@
+{ "all": false, "type": "DISABLE", "silencers": [ { "alarm": "*10min_cpu_usage *load_trigger" } ] }
diff --git a/tests/health_mgmtapi/expected_list/DISABLE_ALL-list.json b/tests/health_mgmtapi/expected_list/DISABLE_ALL-list.json
new file mode 100644
index 000000000..bbc3f4f0c
--- /dev/null
+++ b/tests/health_mgmtapi/expected_list/DISABLE_ALL-list.json
@@ -0,0 +1 @@
+{ "all": true, "type": "DISABLE", "silencers": [] }
diff --git a/tests/health_mgmtapi/expected_list/DISABLE_ALL_ERROR-list.json b/tests/health_mgmtapi/expected_list/DISABLE_ALL_ERROR-list.json
new file mode 100644
index 000000000..e8aee1795
--- /dev/null
+++ b/tests/health_mgmtapi/expected_list/DISABLE_ALL_ERROR-list.json
@@ -0,0 +1 @@
+Auth Error
diff --git a/tests/health_mgmtapi/expected_list/DISABLE_SYSTEM_LOAD-list.json b/tests/health_mgmtapi/expected_list/DISABLE_SYSTEM_LOAD-list.json
new file mode 100644
index 000000000..a7fc1cb8a
--- /dev/null
+++ b/tests/health_mgmtapi/expected_list/DISABLE_SYSTEM_LOAD-list.json
@@ -0,0 +1 @@
+{ "all": false, "type": "DISABLE", "silencers": [ { "chart": "system.load" } ] }
diff --git a/tests/health_mgmtapi/expected_list/FAMILIES_LOAD-list.json b/tests/health_mgmtapi/expected_list/FAMILIES_LOAD-list.json
new file mode 100644
index 000000000..50119f79c
--- /dev/null
+++ b/tests/health_mgmtapi/expected_list/FAMILIES_LOAD-list.json
@@ -0,0 +1 @@
+{ "all": false, "type": "None", "silencers": [ { "families": "load" } ] }
diff --git a/tests/health_mgmtapi/expected_list/HOSTS-list.json b/tests/health_mgmtapi/expected_list/HOSTS-list.json
new file mode 100644
index 000000000..9db21b6c3
--- /dev/null
+++ b/tests/health_mgmtapi/expected_list/HOSTS-list.json
@@ -0,0 +1 @@
+{ "all": false, "type": "SILENCE", "silencers": [ { "hosts": "*" } ] }
diff --git a/tests/health_mgmtapi/expected_list/RESET-list.json b/tests/health_mgmtapi/expected_list/RESET-list.json
new file mode 100644
index 000000000..2d3f09d68
--- /dev/null
+++ b/tests/health_mgmtapi/expected_list/RESET-list.json
@@ -0,0 +1 @@
+{ "all": false, "type": "None", "silencers": [] }
diff --git a/tests/health_mgmtapi/expected_list/SILENCE-list.json b/tests/health_mgmtapi/expected_list/SILENCE-list.json
new file mode 100644
index 000000000..d157f2d3a
--- /dev/null
+++ b/tests/health_mgmtapi/expected_list/SILENCE-list.json
@@ -0,0 +1 @@
+{ "all": false, "type": "SILENCE", "silencers": [ { "alarm": "*10min_cpu_usage *load_trigger" } ] }
diff --git a/tests/health_mgmtapi/expected_list/SILENCE_2-list.json b/tests/health_mgmtapi/expected_list/SILENCE_2-list.json
new file mode 100644
index 000000000..d5e6fa2d1
--- /dev/null
+++ b/tests/health_mgmtapi/expected_list/SILENCE_2-list.json
@@ -0,0 +1 @@
+{ "all": false, "type": "SILENCE", "silencers": [ { "families": "load" } ] }
diff --git a/tests/health_mgmtapi/expected_list/SILENCE_3-list.json b/tests/health_mgmtapi/expected_list/SILENCE_3-list.json
new file mode 100644
index 000000000..69e98cc19
--- /dev/null
+++ b/tests/health_mgmtapi/expected_list/SILENCE_3-list.json
@@ -0,0 +1 @@
+{ "all": false, "type": "SILENCE", "silencers": [] } WARNING: SILENCE or DISABLE command is ineffective without defining any alarm selectors.
diff --git a/tests/health_mgmtapi/expected_list/SILENCE_ALARM_CPU_USAGE-list.json b/tests/health_mgmtapi/expected_list/SILENCE_ALARM_CPU_USAGE-list.json
new file mode 100644
index 000000000..dd789cd33
--- /dev/null
+++ b/tests/health_mgmtapi/expected_list/SILENCE_ALARM_CPU_USAGE-list.json
@@ -0,0 +1 @@
+{ "all": false, "type": "SILENCE", "silencers": [ { "alarm": "*10min_cpu_usage *load_trigger", "chart": "system.load" } ] }
diff --git a/tests/health_mgmtapi/expected_list/SILENCE_ALARM_CPU_USAGE_LOAD_TRIGGER-list.json b/tests/health_mgmtapi/expected_list/SILENCE_ALARM_CPU_USAGE_LOAD_TRIGGER-list.json
new file mode 100644
index 000000000..d157f2d3a
--- /dev/null
+++ b/tests/health_mgmtapi/expected_list/SILENCE_ALARM_CPU_USAGE_LOAD_TRIGGER-list.json
@@ -0,0 +1 @@
+{ "all": false, "type": "SILENCE", "silencers": [ { "alarm": "*10min_cpu_usage *load_trigger" } ] }
diff --git a/tests/health_mgmtapi/expected_list/SILENCE_ALL-list.json b/tests/health_mgmtapi/expected_list/SILENCE_ALL-list.json
new file mode 100644
index 000000000..c88ef9fde
--- /dev/null
+++ b/tests/health_mgmtapi/expected_list/SILENCE_ALL-list.json
@@ -0,0 +1 @@
+{ "all": true, "type": "SILENCE", "silencers": [] }
diff --git a/tests/health_mgmtapi/health-cmdapi-test.sh.in b/tests/health_mgmtapi/health-cmdapi-test.sh.in
index 1bbe269d5..5abf2b170 100755
--- a/tests/health_mgmtapi/health-cmdapi-test.sh.in
+++ b/tests/health_mgmtapi/health-cmdapi-test.sh.in
@@ -1,56 +1,72 @@
 #!/usr/bin/env bash
 # shellcheck disable=SC1117,SC2034,SC2059,SC2086,SC2181
 
-NETDATA_USER_CONFIG_DIR="@configdir_POST@"
-NETDATA_STOCK_CONFIG_DIR="@libconfigdir_POST@"
 NETDATA_VARLIB_DIR="@varlibdir_POST@"
 
-printhelp () {
-	echo "Usage: health-cmdapi-test.sh [OPTIONS]
-	-s SETUP config files for python example tests
-	-c CLEANUP config files from python example tests
-	-r RESTART netdata after SETUP and CLEANUP, using systemctl restart netdata.
-	-t TEST scenarios execution
-	-u <URL> changes the host:port from localhost:19999 to <URL>
-	"
-}
-
 check () {
-	echo -e "${GRAY}Check: '${1}' in 2 sec"
-	sleep 2
-	resp=$(curl -s "http://$URL/api/v1/alarms?all")
+	sec=1
+	echo -e "  ${GRAY}Check: '${1}' in $sec sec"
+	sleep $sec
+    number=$RANDOM
+	resp=$(curl -s "http://$URL/api/v1/alarms?all&$number")
 	r=$(echo "${resp}" | \
 	python3 -c "import sys, json; d=json.load(sys.stdin); \
 	print(\
-	d['alarms']['example.random.example_alarm1']['disabled'], \
-	d['alarms']['example.random.example_alarm1']['silenced'] , \
-	d['alarms']['example.random.example_alarm2']['disabled'], \
-	d['alarms']['example.random.example_alarm2']['silenced'], \
+	d['alarms']['system.cpu.10min_cpu_usage']['disabled'], \
+	d['alarms']['system.cpu.10min_cpu_usage']['silenced'] , \
+	d['alarms']['system.cpu.10min_cpu_iowait']['disabled'], \
+	d['alarms']['system.cpu.10min_cpu_iowait']['silenced'], \
 	d['alarms']['system.load.load_trigger']['disabled'], \
 	d['alarms']['system.load.load_trigger']['silenced'], \
 	);" 2>&1)
 	if [ $? -ne 0 ] ; then
-		echo -e "${RED}ERROR: Unexpected response '$resp'"
+		echo -e "  ${RED}ERROR: Unexpected response stored in /tmp/resp-$number.json"
+		echo "$resp" > /tmp/resp-$number.json
 		err=$((err+1))
+		iter=0
 	elif [ "${r}" != "${2}" ] ; then
-		echo -e "${RED}ERROR: 'Got ${r}'. Expected '${2}'"
-		err=$((err+1))
+		echo -e "  ${GRAY}WARNING: 'Got ${r}'. Expected '${2}'"
+		iter=$((iter+1))
+		if [ $iter -lt 10 ] ; then
+			echo -e "  ${GRAY}Repeating test "
+			check "$1" "$2"
+		else
+			echo -e "  ${RED}ERROR: 'Got ${r}'. Expected '${2}'"
+			iter=0
+			err=$((err+1))
+		fi
 	else
-		echo -e "${GREEN}Success"
+		echo -e "  ${GREEN}Success"
+		iter=0
 	fi
 }
 
 cmd () {
-	echo -e "${WHITE}Cmd '${1}', expecting '${2}'"
+	echo -e "${WHITE}Cmd '${1}'"
+	echo -en "  ${GRAY}Expecting '${2}' : "
 	RESPONSE=$(curl -s "http://$URL/api/v1/manage/health?${1}" -H "X-Auth-Token: $TOKEN" 2>&1)
 	if [ "${RESPONSE}" != "${2}" ] ; then
-		echo -e "${RED}ERROR: Response '${RESPONSE}' != '${2}'"
+		echo -e "${RED}ERROR: Response '${RESPONSE}'"
 		err=$((err+1))
 	else
 		echo -e "${GREEN}Success"
 	fi
 }
 
+check_list() {
+    RESPONSE=$(curl -s "http://$URL/api/v1/manage/health?cmd=LIST" -H "X-Auth-Token: $TOKEN" 2>&1)
+
+    NAME="$1-list.json"
+    echo $RESPONSE > $NAME
+    diff $NAME expected_list/$NAME 1>/dev/null 2>&1
+    if [ $? -eq 0 ]; then
+        echo -e "${GREEN}Success: The list command got the correct answer for $NAME!"
+    else
+        echo -e "${RED}ERROR: the files $NAME and expected_list/$NAME does not match."
+        exit 1
+    fi
+}
+
 WHITE='\033[0;37m'
 RED='\033[0;31m'
 GREEN='\033[0;32m'
@@ -62,63 +78,8 @@ CLEANUP=0
 TEST=0
 URL="localhost:19999"
 
-while getopts :srctu: option
-do
-    case "$option" in
-    s)
-        SETUP=1
-        ;;
-	r)
-		RESTART=1
-		;;
-	c)
-		CLEANUP=1
-		;;
-	t)
-		TEST=1
-		;;
-	u)
-		URL=$OPTARG
-		;;
-	*)
-		printhelp
-		exit 1
-		;;
-	esac
-done
-
-if [ $SETUP -eq 1 ] ; then
-	echo "Preparing netdata configuration for testing"
-	# Prep netdata for tests
-	if [ -f "${NETDATA_USER_CONFIG_DIR}/python.d.conf" ] ; then
-		cp -f "${NETDATA_USER_CONFIG_DIR}/python.d.conf"  /tmp/python.d.conf
-	else
-		cp "${NETDATA_STOCK_CONFIG_DIR}/python.d.conf" "${NETDATA_USER_CONFIG_DIR}/"
-	fi
-	sed -i -e "s/example: no/example: yes/g" "${NETDATA_USER_CONFIG_DIR}/python.d.conf"
-
-	mypath=$(cd ${0%/*} && echo $PWD)
-
-	cp -f "${mypath}/python-example.conf" "${NETDATA_USER_CONFIG_DIR}/health.d/"
-
-	# netdata.conf
-	if [ -f "${NETDATA_USER_CONFIG_DIR}/netdata.conf" ] ; then
-		cp -f "${NETDATA_USER_CONFIG_DIR}/netdata.conf"  /tmp/netdata.conf
-	fi
-	printf "[health]\nrun at least every seconds = 1\n" > "${NETDATA_USER_CONFIG_DIR}/netdata.conf"
-
-	chmod +r "${NETDATA_USER_CONFIG_DIR}/python.d.conf" "${NETDATA_USER_CONFIG_DIR}/netdata.conf" "${NETDATA_USER_CONFIG_DIR}/health.d/python-example.conf" "${NETDATA_STOCK_CONFIG_DIR}/health.d/load.conf"
-	# Restart netdata
-	if [ $RESTART -eq 1 ] ; then
-		echo "Restarting netdata"
-		systemctl restart netdata
-	fi
-fi
-
 err=0
 
-# Execute tests
-if [ $TEST -eq 1 ] ; then
 
 	HEALTH_CMDAPI_MSG_AUTHERROR="Auth Error"
 	HEALTH_CMDAPI_MSG_SILENCEALL="All alarm notifications are silenced"
@@ -143,11 +104,13 @@ if [ $TEST -eq 1 ] ; then
 	# Test default state
 	cmd "cmd=RESET" "$HEALTH_CMDAPI_MSG_RESET"
 	check "Default State" "False False False False False False"
+	check_list "RESET"
 
 	# Test auth failure
 	TOKEN="Wrong token"
 	cmd "cmd=DISABLE ALL" "$HEALTH_CMDAPI_MSG_AUTHERROR"
 	check "Default State" "False False False False False False"
+	check_list "DISABLE_ALL_ERROR"
 
 	# Set correct token
 	TOKEN="${CORRECT_TOKEN}"
@@ -155,108 +118,107 @@ if [ $TEST -eq 1 ] ; then
 	# Test disable
 	cmd "cmd=DISABLE ALL" "$HEALTH_CMDAPI_MSG_DISABLEALL"
 	check "All disabled" "True False True False True False"
+	check_list "DISABLE_ALL"
 
 	# Reset
 	cmd "cmd=RESET" "$HEALTH_CMDAPI_MSG_RESET"
 	check "Default State" "False False False False False False"
+	check_list "RESET"
 
 	# Test silence
 	cmd "cmd=SILENCE ALL" "$HEALTH_CMDAPI_MSG_SILENCEALL"
 	check "All silenced" "False True False True False True"
+	check_list "SILENCE_ALL"
 
 	# Reset
 	cmd "cmd=RESET" "$HEALTH_CMDAPI_MSG_RESET"
 	check "Default State" "False False False False False False"
+	check_list "RESET"
 
 	# Add silencer by name
 	printf -v resp "$HEALTH_CMDAPI_MSG_SILENCE\n$HEALTH_CMDAPI_MSG_ADDED"
-	cmd "cmd=SILENCE&alarm=*example_alarm1 *load_trigger" "${resp}"
+	cmd "cmd=SILENCE&alarm=*10min_cpu_usage *load_trigger" "${resp}"
 	check "Silence notifications for alarm1 and load_trigger" "False True False False False True"
+	check_list "SILENCE_ALARM_CPU_USAGE_LOAD_TRIGGER"
 
 	# Convert to disable health checks
 	cmd "cmd=DISABLE" "$HEALTH_CMDAPI_MSG_DISABLE"
 	check "Disable notifications for alarm1 and load_trigger" "True False False False True False"
+	check_list "DISABLE"
 
 	# Convert back to silence notifications
 	cmd "cmd=SILENCE" "$HEALTH_CMDAPI_MSG_SILENCE"
 	check "Silence notifications for alarm1 and load_trigger" "False True False False False True"
+	check_list "SILENCE"
 
 	# Add second silencer by name
-	cmd "alarm=*example_alarm2" "$HEALTH_CMDAPI_MSG_ADDED"
+	cmd "alarm=*10min_cpu_iowait" "$HEALTH_CMDAPI_MSG_ADDED"
 	check "Silence notifications for alarm1,alarm2 and load_trigger" "False True False True False True"
+	check_list "ALARM_CPU_IOWAIT"
 
 	# Reset
 	cmd "cmd=RESET" "$HEALTH_CMDAPI_MSG_RESET"
+	check_list "RESET"
 
 	# Add silencer by chart
 	printf -v resp "$HEALTH_CMDAPI_MSG_DISABLE\n$HEALTH_CMDAPI_MSG_ADDED"
 	cmd "cmd=DISABLE&chart=system.load" "${resp}"
 	check "Default State" "False False False False True False"
+	check_list "DISABLE_SYSTEM_LOAD"
 
 	# Add silencer by context
-	cmd "context=random" "$HEALTH_CMDAPI_MSG_ADDED"
+	cmd "context=system.cpu" "$HEALTH_CMDAPI_MSG_ADDED"
 	check "Default State" "True False True False True False"
+	check_list "CONTEXT_SYSTEM_CPU"
 
 	# Reset
 	cmd "cmd=RESET" "$HEALTH_CMDAPI_MSG_RESET"
+	check_list "RESET"
 
 	# Add second condition to a selector (AND)
 	printf -v resp "$HEALTH_CMDAPI_MSG_SILENCE\n$HEALTH_CMDAPI_MSG_ADDED"
-	cmd "cmd=SILENCE&alarm=*example_alarm1 *load_trigger&chart=system.load" "${resp}"
+	cmd "cmd=SILENCE&alarm=*10min_cpu_usage *load_trigger&chart=system.load" "${resp}"
 	check "Silence notifications load_trigger" "False False False False False True"
+	check_list "SILENCE_ALARM_CPU_USAGE"
 
 	# Add second selector with two conditions
-	cmd "alarm=*example_alarm1 *load_trigger&context=random" "$HEALTH_CMDAPI_MSG_ADDED"
+	cmd "alarm=*10min_cpu_usage *load_trigger&context=system.cpu" "$HEALTH_CMDAPI_MSG_ADDED"
 	check "Silence notifications load_trigger" "False True False False False True"
+	check_list "ALARM_CPU_USAGE"
 
 	# Reset
 	cmd "cmd=RESET" "$HEALTH_CMDAPI_MSG_RESET"
+	check_list "RESET"
 
 	# Add silencer without a command to disable or silence alarms
 	printf -v resp "$HEALTH_CMDAPI_MSG_ADDED\n$HEALTH_CMDAPI_MSG_STYPEWARNING"
 	cmd "families=load" "${resp}"
 	check "Family selector with no command" "False False False False False False"
+	check_list "FAMILIES_LOAD"
 
 	# Add silence command
 	cmd "cmd=SILENCE" "$HEALTH_CMDAPI_MSG_SILENCE"
 	check "Silence family load" "False False False False False True"
+	check_list "SILENCE_2"
 
 	# Reset
 	cmd "cmd=RESET" "$HEALTH_CMDAPI_MSG_RESET"
+	check_list "RESET"
 
 	# Add command without silencers
 	printf -v resp "$HEALTH_CMDAPI_MSG_SILENCE\n$HEALTH_CMDAPI_MSG_NOSELECTORWARNING"
 	cmd "cmd=SILENCE" "${resp}"
 	check "Command with no selector" "False False False False False False"
+	check_list "SILENCE_3"
 
 	# Add hosts silencer
 	cmd "hosts=*" "$HEALTH_CMDAPI_MSG_ADDED"
 	check "Silence all hosts" "False True False True False True"
+	check_list "HOSTS"
 
 	# Reset
 	cmd "cmd=RESET" "$HEALTH_CMDAPI_MSG_RESET"
-
-fi
-
-# Cleanup
-if [ $CLEANUP -eq 1 ] ; then
-	echo -e "${WHITE}Restoring netdata configuration"
-	for f in "python.d.conf" "netdata.conf" ; do
-		if [ -f "/tmp/$f" ] ; then
-			mv -f "/tmp/$f" "${NETDATA_USER_CONFIG_DIR}/"
-		else
-			rm -f "${NETDATA_USER_CONFIG_DIR}/$f"
-		fi
-	done
-
-	rm -f "${NETDATA_USER_CONFIG_DIR}/health.d/python-example.conf"
-
-	# Restart netdata
-	if [ $RESTART -eq 1 ] ; then
-		echo "Restarting netdata"
-		systemctl restart netdata
-	fi
-fi
+	check_list "RESET"
 
 if [ $err -gt 0 ] ; then
 	echo "$err error(s) found"
diff --git a/tests/health_mgmtapi/python-example.conf b/tests/health_mgmtapi/python-example.conf
deleted file mode 100644
index 66713208c..000000000
--- a/tests/health_mgmtapi/python-example.conf
+++ /dev/null
@@ -1,16 +0,0 @@
-alarm: example_alarm1
-      on: example.random
-   every: 2s
-    warn: $random1 > (($status >= $WARNING)  ? (55) : (75))
-    crit: $random1 > (($status == $CRITICAL) ? (75) : (95))
-    info: random
-      to: sysadmin
-
-alarm: example_alarm2
-      on: example.random
-   every: 2s
-    warn: $random2 > (($status >= $WARNING)  ? (55) : (75))
-    crit: $random2 > (($status == $CRITICAL) ? (75) : (95))
-    info: random
-      to: sysadmin
-