Update default stack version to 8.6.1 (#1118)

Add also some options to try to reduce flakiness in elastic-package stack up:
* Retry limits have been replaced with start periods.
* Fleet-server and elastic-agent healthcheck retries have been increased.
* When docker-compose up fails and elastic-agent container is the only container
  failing, retry once, this may happen if it tries to enroll while fleet-server is being
  restarted or reconfigured.

Co-authored-by: Mario Rodriguez Molins <mario.rodriguez@elastic.co>

Author: jsoriano

internal/install/stack_version.go

@@ -6,5 +6,5 @@ package install
 
 const (
 	// DefaultStackVersion is the default version of the stack
-	DefaultStackVersion = "8.5.1"
+	DefaultStackVersion = "8.6.1"
 )

internal/profile/_static/docker-compose-stack.yml

@@ -4,8 +4,8 @@ services:
     image: "${ELASTICSEARCH_IMAGE_REF}"
     healthcheck:
       test: "curl -s --cacert /usr/share/elasticsearch/config/certs/ca-cert.pem -f -u elastic:changeme https://127.0.0.1:9200/_cat/health | cut -f4 -d' ' | grep -E '(green|yellow)'"
-      retries: 300
-      interval: 1s
+      start_period: 300s
+      interval: 5s
     environment:
       - "ES_JAVA_OPTS=-Xms1g -Xmx1g"
       - "ELASTIC_PASSWORD=changeme"
@@ -32,8 +32,8 @@ services:
         condition: service_healthy
     healthcheck:
       test: "sh /usr/share/kibana/healthcheck.sh"
-      retries: 600
-      interval: 1s
+      start_period: 600s
+      interval: 5s
     env_file:
       - "../certs/ca.env"
     environment:
@@ -60,8 +60,8 @@ services:
         PROFILE: "${PROFILE_NAME}"
     healthcheck:
       test: ["CMD", "curl", "--cacert", "/etc/ssl/package-registry/ca-cert.pem", "-f", "https://localhost:8080"]
-      retries: 300
-      interval: 1s
+      start_period: 300s
+      interval: 5s
     environment:
       - "EPR_LOG_LEVEL=debug"
       - "EPR_ADDRESS=0.0.0.0:8080"
@@ -89,7 +89,7 @@ services:
         condition: service_healthy
     healthcheck:
       test: "curl --cacert /etc/ssl/elastic-agent/ca-cert.pem -f https://localhost:8220/api/status | grep -i healthy 2>&1 >/dev/null"
-      retries: 60
+      start_period: 60s
       interval: 5s
     hostname: docker-fleet-server
     environment:
@@ -124,8 +124,10 @@ services:
         condition: service_healthy
     healthcheck:
       test: "elastic-agent status"
+      timeout: 2s
+      start_period: 360s
       retries: 180
-      interval: 1s
+      interval: 5s
     hostname: docker-fleet-agent
     env_file: "./elastic-agent.${STACK_VERSION_VARIANT}.env"
     volumes:

internal/stack/boot.go

@@ -60,12 +60,41 @@ func BootUp(options Options) error {
 
 	err = dockerComposeUp(options)
 	if err != nil {
+		// At least starting on 8.6.0, fleet-server may be reconfigured or
+		// restarted after being healthy. If elastic-agent tries to enroll at
+		// this moment, it fails inmediately, stopping and making `docker-compose up`
+		// to fail too.
+		// As a workaround, try to give another chance to docker-compose if only
+		// elastic-agent failed.
+		if onlyElasticAgentFailed() {
+			fmt.Println("Elastic Agent failed to start, trying again.")
+			err = dockerComposeUp(options)
+		}
 		return errors.Wrap(err, "running docker-compose failed")
 	}
 
 	return nil
 }
 
+func onlyElasticAgentFailed() bool {
+	status, err := Status()
+	if err != nil {
+		fmt.Printf("Failed to check status of the stack after failure: %v\n", err)
+		return false
+	}
+
+	for _, service := range status {
+		if strings.Contains(service.Name, "elastic-agent") {
+			continue
+		}
+		if !strings.HasPrefix(service.Status, "running") {
+			return false
+		}
+	}
+
+	return true
+}
+
 // TearDown function takes down the testing stack.
 func TearDown(options Options) error {
 	err := dockerComposeDown(options)

test/packages/benchmarks/pipeline_benchmark/manifest.yml

@@ -7,7 +7,6 @@ title: Pipeline benchmarks
 version: 999.999.999
 description: Test for pipeline test runner
 categories: ["network"]
-release: experimental
 license: basic
 type: integration
 conditions:

test/packages/benchmarks/use_pipeline_tests/manifest.yml

@@ -7,7 +7,6 @@ title: Use pipeline tests for the benchmark
 version: 999.999.999
 description: Test for pipeline test runner
 categories: ["network"]
-release: experimental
 license: basic
 type: integration
 conditions:

test/packages/other/multiinput/manifest.yml

@@ -7,7 +7,6 @@ title: Multi-input test
 version: 999.999.999
 description: Test for multiple input tests
 categories: ["network"]
-release: experimental
 license: basic
 type: integration
 conditions:

test/packages/other/pipeline_tests/manifest.yml

@@ -7,7 +7,6 @@ title: Pipeline tests
 version: 999.999.999
 description: Test for pipeline test runner
 categories: ["network"]
-release: experimental
 license: basic
 type: integration
 conditions: