Feature healthcheck

data/common.yaml

@@ -44,6 +44,11 @@ fluentbit::storage_backlog_mem_limit: 5M
 fluentbit::storage_metrics: true
 fluentbit::storage_delete_irrecoverable_chunks: false
 
+fluentbit::health_check: false
+fluentbit::hc_errors_count: 5
+fluentbit::hc_retry_failure_count: 5
+fluentbit::hc_period: 60
+
 fluentbit::manage_plugins_file: true
 fluentbit::plugins_file: '/etc/fluent-bit/plugins.conf'
 

manifests/config.pp

@@ -105,6 +105,10 @@
   $storage_max_chunks_up = $fluentbit::storage_max_chunks_up
   $storage_metrics = bool2str($fluentbit::storage_metrics, 'On', 'Off')
   $storage_delete_irrecoverable_chunks = bool2str($fluentbit::storage_delete_irrecoverable_chunks, 'On', 'Off')
+  $health_check = bool2str($fluentbit::health_check, 'On', 'Off')
+  $hc_errors_count = $fluentbit::hc_errors_count
+  $hc_retry_failure_count = $fluentbit::hc_retry_failure_count
+  $hc_period = $fluentbit::hc_period
   $scheduler_cap = $fluentbit::scheduler_cap
   $scheduler_base = $fluentbit::scheduler_base
   $json_convert_nan_to_null = $fluentbit::json_convert_nan_to_null
@@ -124,6 +128,15 @@
       'storage.delete_irrecoverable_chunks' => $storage_delete_irrecoverable_chunks,
     },
   }
+  $health_config = $health_check ? {
+    undef           => {},
+    default         => {
+      'health.check'                        => $health_check,
+      'hc.errors_count'                     => $hc_errors_count,
+      'hc.retry_failure_count'              => $hc_retry_failure_count,
+      'hc.period'                           => $hc_period,
+    },
+  }
 
   file { $fluentbit::config_file:
     mode    => $fluentbit::config_file_mode,
@@ -147,7 +160,8 @@
           'scheduler.cap'            => $scheduler_cap,
           'scheduler.base'           => $scheduler_base,
           'json.convert_nan_to_null' => $json_convert_nan_to_null,
-        } + $storage_config,
+        } + $storage_config
+          + $health_config,
       },
     ),
   }

manifests/init.pp

@@ -111,6 +111,18 @@
 #   located in the configured storage path directory will be deleted when Fluent-Bit starts.
 #   Default value: false
 #
+# @param health_check
+#   Enable or disable health_check
+#   Default Off
+# @param hc_errors_count
+#   Only in use if health_check is enabled. The error count after which the healcheck returns an error.
+#   Default 5
+# @param hc_retry_failure_count
+#   Only in use if health_check is enabled. Retry count till a check returns an error
+#   Default 5
+# @param hc_period
+#   Only in use if health_check is enabled. Time period by second to count the error and retry failure data point
+#   Default 60
 # @param manage_plugins_file
 #   Whether to manage the enabled external plugins
 #
@@ -232,6 +244,12 @@
   Boolean                          $storage_metrics,
   Boolean                          $storage_delete_irrecoverable_chunks,
   Optional[String[1]]              $storage_backlog_mem_limit,
+
+  Boolean $health_check,
+  Integer $hc_errors_count,
+  Integer $hc_retry_failure_count,
+  Integer $hc_period,
+
   Integer $coro_stack_size,
   String $plugins_dir,
   String $scripts_dir,

templates/fluentbit.conf.epp

@@ -34,6 +34,12 @@
     storage.max_chunks_up     <%= $service['storage.max_chunks_up'] %>
     storage.metrics           <%= $service['storage.metrics'] %>
     storage.delete_irrecoverable_chunks <%= $service['storage.delete_irrecoverable_chunks'] %>
+<% if $service['health.check'] { %>
+    Health_Check              <%= $service['health.check'] %>
+    HC_Errors_Count           <%= $service['hc.errors_count'] %>
+    HC_Rerty_Failure_Count    <%= $service['hc.retry_failure_count'] %>
+    HC_Period                 <%= $service['hc.period'] %>
+<% } -%>
 
 <%- if $manage_config_dir { %>
 @INCLUDE pipelines/*.conf