Downgrade PMD error logs for systemmetrics pipeline (#2073)

Author: sky333999

plugins/outputs/cloudwatch/cloudwatch.go

@@ -367,8 +367,13 @@ func (c *CloudWatch) backoffSleep() {
 		d = c.config.BackoffRetryBase * time.Duration(1<<c.retries)
 	}
 	d = (d / 2) + publishJitter(d/2)
-	log.Printf("W! cloudwatch: %v retries, going to sleep %v ms before retrying.",
-		c.retries, d.Milliseconds())
+	if !c.config.DowngradeErrors {
+		log.Printf("W! cloudwatch: %v retries, going to sleep %v ms before retrying.",
+			c.retries, d.Milliseconds())
+	} else {
+		log.Printf("I! cloudwatch: %v retries, going to sleep %v ms before retrying.",
+			c.retries, d.Milliseconds())
+	}
 	c.retries++
 	time.Sleep(d)
 }
@@ -423,7 +428,11 @@ func (c *CloudWatch) WriteToCloudWatch(req interface{}) {
 				continue
 
 			default:
-				log.Printf("E! cloudwatch: code: %s, message: %s, original error: %+v", awsErr.Code(), awsErr.Message(), awsErr.OrigErr())
+				if !c.config.DowngradeErrors {
+					log.Printf("E! cloudwatch: code: %s, message: %s, original error: %+v", awsErr.Code(), awsErr.Message(), awsErr.OrigErr())
+				} else {
+					log.Printf("I! cloudwatch: PutMetricData unsuccessful")
+				}
 				c.backoffSleep()
 			}
 		} else {
@@ -432,7 +441,11 @@ func (c *CloudWatch) WriteToCloudWatch(req interface{}) {
 		break
 	}
 	if err != nil {
-		log.Println("E! cloudwatch: WriteToCloudWatch failure, err: ", err)
+		if !c.config.DowngradeErrors {
+			log.Println("E! cloudwatch: WriteToCloudWatch failure, err: ", err)
+		} else {
+			log.Println("I! cloudwatch: WriteToCloudWatch unsuccessful, batch dropped")
+		}
 	}
 }
 

plugins/outputs/cloudwatch/config.go

@@ -42,6 +42,8 @@ type Config struct {
 	BackoffRetryBase time.Duration `mapstructure:"backoff_retry_base,omitempty"`
 	// MaxConcurrentPublishers is the number of concurrent workers making PMD calls. Defaults to 10.
 	MaxConcurrentPublishers int `mapstructure:"max_concurrent_publishers,omitempty"`
+	// DowngradeErrors suppresses PutMetricData failure logs: E! and W! are downgraded to I! level.
+	DowngradeErrors bool `mapstructure:"downgrade_errors,omitempty"`
 }
 
 var _ component.Config = (*Config)(nil)

translator/tocwconfig/sampleConfig/system_metrics_config.yaml

@@ -26,6 +26,7 @@ exporters:
     rollup_dimensions:
       - - InstanceId
       - [ ]
+    downgrade_errors: true
 extensions:
   agenthealth/metrics:
     is_usage_data_enabled: true

translator/translate/otel/pipeline/systemmetrics/cloudwatch.go

@@ -47,6 +47,7 @@ func (t *cloudWatchTranslator) Translate(_ *confmap.Conf) (component.Config, err
 	cfg.MaxRetryCount = 2
 	cfg.BackoffRetryBase = time.Minute
 	cfg.MaxConcurrentPublishers = 1
+	cfg.DowngradeErrors = true
 
 	return cfg, nil
 }