fix(broker/bam): detection of downtime end was bad (#319)

* fix(tests/bam): sometimes broker is too long to stop
* fix(tests/summary): new options -f and -s
* fix(conanfile): openssl updated

REFS: MON-14091

broker/bam/src/ba.cc

@@ -1,5 +1,5 @@
 /*
-** Copyright 2014-2016, 2021 Centreon
+** Copyright 2014-2016, 2021-2022 Centreon
 **
 ** Licensed under the Apache License, Version 2.0 (the "License");
 ** you may not use this file except in compliance with the License.

broker/bam/src/kpi_service.cc

@@ -1,5 +1,5 @@
 /*
-** Copyright 2014-2015, 2021 Centreon
+** Copyright 2014-2015, 2021-2022 Centreon
 **
 ** Licensed under the Apache License, Version 2.0 (the "License");
 ** you may not use this file except in compliance with the License.
@@ -181,7 +181,7 @@ bool kpi_service::is_acknowledged() const {
  *  @param[out] visitor  Object that will receive events.
  */
 void kpi_service::service_update(
-    std::shared_ptr<neb::service_status> const& status,
+    const std::shared_ptr<neb::service_status>& status,
     io::stream* visitor) {
   if (status && status->host_id == _host_id &&
       status->service_id == _service_id) {
@@ -298,15 +298,15 @@ void kpi_service::service_update(
  *  @param[in]  dt
  *  @param[out] visitor  Object that will receive events.
  */
-void kpi_service::service_update(std::shared_ptr<neb::downtime> const& dt,
+void kpi_service::service_update(const std::shared_ptr<neb::downtime>& dt,
                                  io::stream* visitor) {
   assert(dt && dt->host_id == _host_id && dt->service_id == _service_id);
   // Update information.
   bool downtimed = dt->was_started && dt->actual_end_time.is_null();
   if (!_downtimed && downtimed)
     _downtimed = true;
 
-  if (_downtime_ids.contains(dt->internal_id) && !dt->was_cancelled) {
+  if (_downtime_ids.contains(dt->internal_id) && dt->deletion_time.is_null()) {
     log_v2::bam()->trace("Downtime {} already handled in this kpi service",
                          dt->internal_id);
     return;

broker/bam/test/ba/kpi_service.cc

@@ -968,6 +968,7 @@ TEST_F(BamBA, KpiServiceDt) {
     std::cout << "service_update 1" << std::endl;
     kpis[0]->service_update(dt, _visitor.get());
 
+    dt->deletion_time = now + 2 + 10 * i + 5;
     dt->actual_end_time = now + 2 + 10 * i + 5;
     dt->was_cancelled = true;
     std::cout << "service_update 2" << std::endl;

ci/docker/Dockerfile.collect-debian11-dependencies

@@ -1,6 +1,6 @@
 FROM debian:bullseye
 RUN apt-get -y update && \
-    apt-get -y install cmake \ 
+    apt-get -y install cmake \
     curl \
     gcc \
     g++ \
@@ -36,5 +36,3 @@ COPY conanfile.txt .
 RUN cat conanfile.txt
 RUN conan install . -s compiler.cppstd=14 -s compiler.libcxx=libstdc++11 --build='missing'
 WORKDIR /src
-
-

ci/scripts/collect-unit-tests.sh

@@ -26,4 +26,4 @@ tests/ut_broker --gtest_output=xml:ut_broker.xml
 tests/ut_engine --gtest_output=xml:ut_engine.xml
 tests/ut_clib --gtest_output=xml:ut_clib.xml
 tests/ut_connector --gtest_output=xml:ut_connector.xml
-echo "----------------------------------------------------------   end of ut tests ------------------------------------------------"
+echo "----------------------------------------------------------   end of ut tests ------------------------------------------------"

conanfile.txt

@@ -4,8 +4,8 @@ asio/1.22.1
 fmt/8.1.1
 spdlog/1.10.0
 nlohmann_json/3.10.5
-openssl/1.1.1n
-grpc/1.43.0
+openssl/1.1.1o
+grpc/1.46.3
 mariadb-connector-c/3.1.12
 zlib/1.2.12
 boost/1.79.0
@@ -15,7 +15,13 @@ libssh2/1.10.0
 cmake
 
 [options]
-grpc:secure=True
 boost:header_only=True
+grpc:csharp_ext=False
+grpc:csharp_plugin=False
+grpc:node_plugin=False
+grpc:objective_c_plugin=False
+grpc:php_plugin=False
+grpc:python_plugin=True
+grpc:ruby_plugin=False
+grpc:secure=True
 libssh2:shared=False
-

engine/inc/com/centreon/engine/configuration/state.hh

@@ -293,13 +293,11 @@ class state {
   int perfdata_timeout() const noexcept;
   void perfdata_timeout(int value);
   std::string const& poller_name() const noexcept;
-  void poller_name(std::string const& value) noexcept;
+  void poller_name(std::string const& value);
   uint32_t poller_id() const noexcept;
-  void poller_id(uint32_t value) noexcept;
+  void poller_id(uint32_t value);
   uint16_t rpc_port() const noexcept;
-  void rpc_port(uint16_t value) noexcept;
-  const std::string& rpc_listen_address() const noexcept;
-  void rpc_listen_address(const std::string& listen_address) noexcept;
+  void rpc_port(uint16_t value);
   bool process_performance_data() const noexcept;
   void process_performance_data(bool value);
   std::list<std::string> const& resource_file() const noexcept;

engine/src/configuration/state.cc

@@ -2798,7 +2798,7 @@ std::string const& state::poller_name() const noexcept {
  *
  *  @param[in] value The new poller_name value.
  */
-void state::poller_name(std::string const& value) noexcept {
+void state::poller_name(std::string const& value) {
   _poller_name = value;
 }
 
@@ -2816,7 +2816,7 @@ uint32_t state::poller_id() const noexcept {
  *
  *  @param[in] value The new poller_id value.
  */
-void state::poller_id(uint32_t value) noexcept {
+void state::poller_id(uint32_t value) {
   _poller_id = value;
 }
 
@@ -2834,7 +2834,7 @@ uint16_t state::rpc_port() const noexcept {
  *
  *  @param[in] value The new poller_id value.
  */
-void state::rpc_port(uint16_t value) noexcept {
+void state::rpc_port(uint16_t value) {
   _rpc_port = value;
 }
 
@@ -2852,7 +2852,7 @@ const std::string& state::rpc_listen_address() const noexcept {
  *
  *  @param[in] value The new grpc api listen address.
  */
-void state::rpc_listen_address(const std::string& listen_address) noexcept {
+void state::rpc_listen_address(const std::string& listen_address) {
   _rpc_listen_address = listen_address;
 }
 

tests/bam/inherited_downtime.robot

@@ -2,10 +2,12 @@
 Resource	../resources/resources.robot
 Suite Setup	Clean Before Suite
 Suite Teardown	Clean After Suite
-Test Setup	Stop Processes
+Test Setup	BAM Setup
 
 Documentation	Centreon Broker and BAM
 Library	Process
+Library	DatabaseLibrary
+Library	DateTime
 Library	OperatingSystem
 Library	../resources/Broker.py
 Library	../resources/Engine.py
@@ -60,7 +62,7 @@ BEBAMIDT1
 	Should Be True	${result}	msg=The BA ba_1 is in downtime as it should not
 
 	Stop Engine
-	Stop Broker
+	Kindly Stop Broker
 
 BEBAMIDT2
 	[Documentation]	A BA of type 'worst' with one service is configured. The BA is in critical state, because of its service. Then we set a downtime on this last one. An inherited downtime is set to the BA. Engine is restarted. Broker is restarted. The two downtimes are still there with no duplicates. The downtime is removed from the service, the inherited downtime is then deleted.
@@ -105,7 +107,7 @@ BEBAMIDT2
 	  Start Engine
 	  Sleep	3s
 	  # Broker is restarted
-	  Stop Broker
+	  Kindly Stop Broker
 	  Start Broker
 	END
 
@@ -126,7 +128,7 @@ BEBAMIDT2
 	Should Be True	${result}	msg=We should have no more downtime
 
 	Stop Engine
-	Stop Broker
+	Kindly Stop Broker
 
 BEBAMIGNDT1
 	[Documentation]	A BA of type 'worst' with two services is configured. The downtime policy on this ba is "Ignore the indicator in the calculation". The BA is in critical state, because of the second critical service. Then we apply two downtimes on this last one. The BA state is ok because of the policy on indicators. A first downtime is cancelled, the BA is still OK, but when the second downtime is cancelled, the BA should be CRITICAL.
@@ -209,3 +211,94 @@ BEBAMIGNDT1
 	Stop Engine
 	Kindly Stop Broker
 
+BEBAMIGNDT2
+	[Documentation]	A BA of type 'worst' with two services is configured. The downtime policy on this ba is "Ignore the indicator in the calculation". The BA is in critical state, because of the second critical service. Then we apply two downtimes on this last one. The BA state is ok because of the policy on indicators. The first downtime reaches its end, the BA is still OK, but when the second downtime reaches its end, the BA should be CRITICAL.
+	[Tags]	broker	downtime	engine	bam
+	Clear Commands Status
+	Config Broker	module
+	Config Broker	central
+	Broker Config Log	central	core	error
+	Broker Config Log	central	bam	trace
+	Config Broker	rrd
+	Config Engine	${1}
+
+	Clone Engine Config To DB
+	Add Bam Config To Engine
+
+	@{svc}=	Set Variable	${{ [("host_16", "service_313"), ("host_16", "service_314")] }}
+	Create BA With Services	test	worst	${svc}  ignore
+	Add Bam Config To Broker	central
+	# Command of service_314 is set to critical
+	${cmd_1}=	Get Command Id	313
+	Log To Console	service_314 has command id ${cmd_1}
+	Set Command Status	${cmd_1}	0
+	${cmd_2}=	Get Command Id	314
+	Log To Console	service_314 has command id ${cmd_2}
+	Set Command Status	${cmd_2}	2
+	Start Broker
+	Start Engine
+	Sleep	5s
+
+	# KPI set to ok
+	Repeat Keyword	3 times	Process Service Check Result	host_16	service_313	0	output critical for 313
+	${result}=	Check Service Status With Timeout	host_16	service_313	0	60
+	Should Be True	${result}	msg=The service (host_16,service_313) is not OK as expected
+
+	# KPI set to critical
+	Repeat Keyword	3 times	Process Service Check Result	host_16	service_314	2	output critical for 314
+	${result}=	Check Service Status With Timeout	host_16	service_314	2	60
+	Should Be True	${result}	msg=The service (host_16,service_314) is not CRITICAL as expected
+
+	# The BA should become critical
+	${result}=	Check Ba Status With Timeout	test	2	60
+	Should Be True	${result}	msg=The BA ba_1 is not CRITICAL as expected
+	Log To console	The BA is critical.
+
+	# Two downtimes are applied on service_314
+	Schedule Service Downtime	host_16	service_314	60
+	${result}=	Check Service Downtime With Timeout	host_16	service_314	1	60
+	Should Be True	${result}	msg=The service (host_16, service_314) is not in downtime as it should be
+	Log to console	One downtime applied to service_314.
+
+	Schedule Service Downtime	host_16	service_314	30
+	${result}=	Check Service Downtime With Timeout	host_16	service_314	2	60
+	Should Be True	${result}	msg=The service (host_16, service_314) is not in downtime as it should be
+	Log to console	Two downtimes applied to service_314.
+
+	${result}=	Check Service Downtime With Timeout	_Module_BAM_1	ba_1	0	60
+	Should Be True	${result}	msg=The BA ba_1 is in downtime but should not
+	Log to console	The BA is configured to ignore kpis in downtime
+
+	${result}=	Check Ba Status With Timeout	test	0	60
+	Should Be True	${result}	msg=The service in downtime should be ignored while computing the state of this BA.
+	Log to console	The BA is OK, since the critical service is in downtime.
+
+	# The first downtime should reach its end
+
+        Log to console  After 30s, the first downtime should be finished.
+	${result}=	Check Service Downtime With Timeout	host_16	service_314	1	60
+	Should Be True	${result}	msg=The service (host_16, service_314) does not contain 1 downtime as it should
+	Log to console	Still one downtime applied to service_314.
+
+        Log to console  After 30s, the second downtime should be finished.
+	${result}=	Check Ba Status With Timeout	test	0	60
+	Should Be True	${result}	msg=The BA is not OK whereas the service_314 is still in downtime.
+	Log to console	The BA is still OK
+
+	# The second downtime finishes
+	${result}=	Check Ba Status With Timeout	test	2	60
+	Should Be True	${result}	msg=The critical service is no more in downtime, the BA should be critical.
+	Log to console	The BA is now critical (no more downtime)
+
+	Stop Engine
+	Kindly Stop Broker
+
+*** Keywords ***
+BAM Setup
+	Stop Processes
+        Connect To Database	pymysql	${DBName}	${DBUser}	${DBPass}	${DBHost}	${DBPort}
+        ${date}=	Get Current Date  result_format=epoch
+        log to console    date=${date}
+        Query	UPDATE downtimes SET deletion_time=${date}, actual_end_time=${date} WHERE actual_end_time is null
+
+

tests/broker-engine/bbdo-protobuf.robot

@@ -32,7 +32,7 @@ BEPBBEE1
 	${result}=	Find In Log with timeout	${centralLog}	${start}	${content}	30
 	Should Be True	${result}	msg=Message about not matching bbdo versions not available
 	Stop Engine
-	Stop Broker
+	Kindly Stop Broker
 
 BEPBBEE2
 	[Documentation]	bbdo_version 3 not compatible with sql/storage
@@ -74,7 +74,7 @@ BEPBBEE3
 	Start Engine
         Wait Until Created	/tmp/pbservicestatus.log	1m
 	Stop Engine
-	Stop Broker
+	Kindly Stop Broker
 
 BEPBBEE4
 	[Documentation]	bbdo_version 3 generates new bbdo protobuf host status messages.
@@ -96,7 +96,7 @@ BEPBBEE4
 	Start Engine
         Wait Until Created	/tmp/pbhoststatus.log	1m
 	Stop Engine
-	Stop Broker
+	Kindly Stop Broker
 
 BEPBBEE5
 	[Documentation]	bbdo_version 3 generates new bbdo protobuf service messages.
@@ -118,4 +118,4 @@ BEPBBEE5
 	Start Engine
         Wait Until Created	/tmp/pbservice.log	1m
 	Stop Engine
-	Stop Broker
+	Kindly Stop Broker

tests/broker-engine/compression.robot

@@ -33,7 +33,7 @@ BECC1
 		Start Engine
 		${result}=	Check Connections
 		Should Be True	${result}	msg=Engine and Broker not connected
-		Stop Broker
+		Kindly Stop Broker
 		Stop Engine
 		${content1}=	Create List	we have extensions '${ext["${comp1}"]}' and peer has '${ext["${comp2}"]}'
 		${content2}=	Create List	we have extensions '${ext["${comp2}"]}' and peer has '${ext["${comp1}"]}'

tests/broker-engine/hostgroups.robot

@@ -40,7 +40,7 @@ EBNHG1
 	${result}=	Find In Log With Timeout	${centralLog}	${start}	${content}	45
 	Should Be True	${result}	msg=One of the new host groups not found in logs.
 	Stop Engine
-	Stop Broker
+	Kindly Stop Broker
 
 EBNHGU1
 	[Documentation]	New host group with several pollers and connections to DB with broker configured with unified_sql