feat(service): Allow configuration of systemd service Restart directive (#813)

* Add restart_policy and restart_sec properties to elasticsearch_service

Co-authored-by: damacus <[email protected]>

* Add documentation and integration tests for restart functionality

Co-authored-by: damacus <[email protected]>

* Fix RSpec failure by using proper test recipe instead of converge_dsl

Co-authored-by: damacus <[email protected]>

* Revert to inline testing with converge_block method

Co-authored-by: damacus <[email protected]>

* Fixes

Signed-off-by: Dan Webb <[email protected]>

---------

Signed-off-by: Dan Webb <[email protected]>
Co-authored-by: copilot-swe-agent[bot] <[email protected]>
Co-authored-by: damacus <[email protected]>
Co-authored-by: Dan Webb <[email protected]>

Author: Copilot

.rubocop.yml

@@ -0,0 +1,2 @@
+require:
+  - cookstyle

Dangerfile

@@ -232,6 +232,37 @@ a matching `elasticsearch_config` resource in the collection.
 elasticsearch_service 'elasticsearch'
 ```
 
+#### Restart Configuration
+
+You can configure systemd restart behavior using the `restart_policy` and `restart_sec` properties:
+
+```ruby
+elasticsearch_service 'elasticsearch' do
+  restart_policy 'on-failure'    # Restart only on failure
+  restart_sec 30                 # Wait 30 seconds before restart
+end
+```
+
+```ruby
+elasticsearch_service 'elasticsearch' do
+  restart_policy 'always'        # Always restart
+  restart_sec '5min'             # Wait 5 minutes before restart
+end
+```
+
+Valid restart policies:
+
+- `''` (empty string) - No automatic restart (default, maintains backward compatibility)
+- `'no'` - Never restart
+- `'always'` - Always restart regardless of exit status
+- `'on-success'` - Restart only when process exits cleanly
+- `'on-failure'` - Restart when process exits with non-zero code, killed by signal, timeout, or watchdog
+- `'on-abnormal'` - Restart when process is terminated by signal, timeout, or watchdog
+- `'on-abort'` - Restart when process exits due to uncaught signal
+- `'on-watchdog'` - Restart when watchdog timeout occurs
+
+The `restart_sec` property accepts either an integer (seconds) or a systemd time span string (e.g., "5min", "30s", "1m 30s").
+
 If you'd like to skip init scripts and systemd scripts, simply pass `nil` for
 the template file (init_source or systemd_source) and this cookbook will
 entirely skip trying to setup those scripts. Combined with changing the default

README.md

@@ -0,0 +1,5 @@
+# .mise.toml
+
+[env]
+PATH = "/opt/chef-workstation/bin:/opt/chef-workstation/embedded/bin:{{env.PATH}}"
+KITCHEN_LOCAL_YAML = "kitchen.dokken.yml"

mise.toml

@@ -14,6 +14,16 @@
         [Symbol, String, Array],
         default: [:enable, :start]
 
+property :restart_policy,
+        String,
+        default: '',
+        regex: /^(no|always|on-success|on-failure|on-abnormal|on-abort|on-watchdog)?$/,
+        description: 'Systemd restart policy. Valid values: no, always, on-success, on-failure, on-abnormal, on-abort, on-watchdog. Defaults to empty string (no restart).'
+
+property :restart_sec,
+        [Integer, String],
+        description: 'Configures the time to sleep before restarting a service (seconds). Takes an integer or time span value (e.g., "5min 20s").'
+
 action :configure do
   es_user = find_es_resource(Chef.run_context, :elasticsearch_user, new_resource)
   es_conf = find_es_resource(Chef.run_context, :elasticsearch_configure, new_resource)
@@ -30,6 +40,38 @@
 
   default_conf_dir = platform_family?('rhel', 'amazon') ? '/etc/sysconfig' : '/etc/default'
 
+  service_config = {
+    Type: 'notify',
+    RuntimeDirectory: 'elasticsearch',
+    PrivateTmp: 'true',
+    Environment: [
+      "ES_HOME=#{es_conf.path_home}",
+      'ES_PATH_CONF=/etc/elasticsearch',
+      "PID_DIR=#{es_conf.path_pid}",
+      'ES_SD_NOTIFY=true',
+    ],
+    EnvironmentFile: "-#{default_conf_dir}/#{new_resource.service_name}",
+    WorkingDirectory: es_conf.path_home.to_s,
+    User: es_user.username,
+    Group: es_user.groupname,
+    ExecStart: "#{es_conf.path_home}/bin/systemd-entrypoint -p ${PID_DIR}/elasticsearch.pid --quiet",
+    StandardOutput: 'journal',
+    StandardError: 'inherit',
+    LimitNOFILE: '65535',
+    LimitNPROC: '4096',
+    LimitAS: 'infinity',
+    LimitFSIZE: 'infinity',
+    TimeoutStopSec: '0',
+    KillSignal: 'SIGTERM',
+    KillMode: 'process',
+    SendSIGKILL: 'no',
+    SuccessExitStatus: '143',
+    TimeoutStartSec: '900',
+  }
+
+  service_config[:Restart] = new_resource.restart_policy if new_resource.restart_policy && !new_resource.restart_policy.empty?
+  service_config[:RestartSec] = new_resource.restart_sec if new_resource.restart_sec
+
   systemd_unit new_resource.service_name do
     content(
       Unit: {
@@ -38,34 +80,7 @@
         Wants: 'network-online.target',
         After: 'network-online.target',
       },
-      Service: {
-        Type: 'notify',
-        RuntimeDirectory: 'elasticsearch',
-        PrivateTmp: 'true',
-        Environment: [
-          "ES_HOME=#{es_conf.path_home}",
-          'ES_PATH_CONF=/etc/elasticsearch',
-          "PID_DIR=#{es_conf.path_pid}",
-          'ES_SD_NOTIFY=true',
-        ],
-        EnvironmentFile: "-#{default_conf_dir}/#{new_resource.service_name}",
-        WorkingDirectory: es_conf.path_home.to_s,
-        User: es_user.username,
-        Group: es_user.groupname,
-        ExecStart: "#{es_conf.path_home}/bin/systemd-entrypoint -p ${PID_DIR}/elasticsearch.pid --quiet",
-        StandardOutput: 'journal',
-        StandardError: 'inherit',
-        LimitNOFILE: '65535',
-        LimitNPROC: '4096',
-        LimitAS: 'infinity',
-        LimitFSIZE: 'infinity',
-        TimeoutStopSec: '0',
-        KillSignal: 'SIGTERM',
-        KillMode: 'process',
-        SendSIGKILL: 'no',
-        SuccessExitStatus: '143',
-        TimeoutStartSec: '900',
-      },
+      Service: service_config,
       Install: {
         WantedBy: 'multi-user.target',
       }

resources/service.rb

@@ -0,0 +1,59 @@
+require_relative 'spec_helper'
+
+describe 'elasticsearch_service' do
+  before { stub_resources }
+
+  supported_platforms.each do |platform, versions|
+    versions.each do |version|
+      context "on #{platform.capitalize} #{version}" do
+        let(:chef_run) do
+          ChefSpec::ServerRunner.new(platform: platform, version: version, step_into: ['elasticsearch_service']) do |node, server|
+            node_resources(node)
+            stub_chef_zero(platform, version, server)
+          end.converge('test::restart_policy')
+        end
+
+        it 'creates systemd unit with restart policy' do
+          expect(chef_run).to create_systemd_unit('elasticsearch').with(
+            content: hash_including(
+              Service: hash_including(
+                Restart: 'on-failure',
+                RestartSec: 30
+              )
+            )
+          )
+        end
+
+        it 'enables and starts the elasticsearch service' do
+          expect(chef_run).to enable_service('elasticsearch')
+          expect(chef_run).to start_service('elasticsearch')
+        end
+      end
+    end
+  end
+
+  # Test default behavior (no restart)
+  # rubocop:disable Style/MultilineBlockChain
+  context 'with default configuration' do
+    let(:chef_run) do
+      ChefSpec::ServerRunner.new(platform: 'ubuntu', version: '20.04', step_into: ['elasticsearch_service']) do |node, server|
+        node_resources(node)
+        stub_chef_zero('ubuntu', '20.04', server)
+      end.converge_block do
+        elasticsearch_user 'elasticsearch'
+        elasticsearch_install 'elasticsearch' do
+          type 'package'
+        end
+        elasticsearch_configure 'elasticsearch'
+        elasticsearch_service 'elasticsearch'
+      end
+    end
+
+    it 'creates systemd unit without restart policy' do
+      systemd_unit = chef_run.systemd_unit('elasticsearch')
+      expect(systemd_unit.content[:Service]).not_to have_key(:Restart)
+      expect(systemd_unit.content[:Service]).not_to have_key(:RestartSec)
+    end
+  end
+  # rubocop:enable Style/MultilineBlockChain
+end

spec/service_spec.rb

@@ -0,0 +1,21 @@
+# test fixture to validate restart_policy functionality
+
+# Basic installation and user setup
+elasticsearch_user 'elasticsearch'
+
+elasticsearch_install 'elasticsearch' do
+  type 'package'
+end
+
+elasticsearch_configure 'elasticsearch' do
+  allocated_memory '256m'
+  configuration('node.name' => 'restart_test_node')
+  action :manage
+end
+
+# Test service with restart policy
+elasticsearch_service 'elasticsearch' do
+  restart_policy 'on-failure'
+  restart_sec 30
+  service_actions [:enable, :start]
+end

test/fixtures/cookbooks/test/recipes/restart_policy.rb

@@ -0,0 +1,38 @@
+control 'Service restart configuration' do
+  impact 1.0
+  title 'Elasticsearch service restart policy configuration'
+  desc 'Verify that restart policy is correctly configured in systemd unit'
+
+  # Test that the systemd service file exists
+  describe file('/etc/systemd/system/elasticsearch.service') do
+    it { should exist }
+  end
+
+  # Test restart policy configuration if the test recipe was used
+  if file('/etc/systemd/system/elasticsearch.service').exist?
+    service_content = file('/etc/systemd/system/elasticsearch.service').content
+
+    # Check if this is the restart_policy test
+    if service_content.include?('Restart=on-failure')
+      describe 'Elasticsearch service restart configuration' do
+        it 'should have restart policy configured' do
+          expect(service_content).to match(/^Restart=on-failure$/)
+        end
+
+        it 'should have restart delay configured' do
+          expect(service_content).to match(/^RestartSec=30$/)
+        end
+      end
+    else
+      describe 'Elasticsearch service default configuration' do
+        it 'should not have restart policy by default' do
+          expect(service_content).not_to match(/^Restart=/)
+        end
+
+        it 'should not have restart delay by default' do
+          expect(service_content).not_to match(/^RestartSec=/)
+        end
+      end
+    end
+  end
+end