Minor: Adding JSON_PARSER_ERROR and CONFIG_OVERRIDES_SER_FAILED (open…

…search-project#366) Signed-off-by: Khushboo Rajput <khushbr@amazon.com>
CoderJeffrey · May 17, 2023 · 744f4e3 · 744f4e3
1 parent e420988
commit 744f4e3
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 11 deletions.
diff --git a/...in/java/org/opensearch/performanceanalyzer/rca/framework/metrics/ExceptionsAndErrors.java b/...in/java/org/opensearch/performanceanalyzer/rca/framework/metrics/ExceptionsAndErrors.java
@@ -15,9 +15,11 @@ public enum ExceptionsAndErrors implements MeasurementSet {
     INVALID_CONFIG_RCA_AGENT_STOPPED("InvalidConfigRCAAgentStopped"),
     RCA_FRAMEWORK_CRASH("RcaFrameworkCrash"),
     CONFIG_DIR_NOT_FOUND("ConfigDirectoryNotFound"),
+    CONFIG_OVERRIDES_SER_FAILED("ConfigOverridesSerFailed"),
     WRITE_UPDATED_RCA_CONF_ERROR("WriteUpdatedRcaConfError"),
     BATCH_METRICS_CONFIG_ERROR("BatchMetricsConfigError"),
     MUTE_ERROR("MuteError"),
+    JSON_PARSER_ERROR("JsonParserError", "namedCount", Statistics.NAMED_COUNTERS),
 
     /**
      * Aggregate metrics across RCA Graph nodes tracking failure in various staged of RCA Graph
@@ -54,15 +56,10 @@ public enum ExceptionsAndErrors implements MeasurementSet {
     /** Thread ID is no loner exists */
     JVM_THREAD_ID_NO_LONGER_EXISTS("JVMThreadIdNoLongerExists"),
 
-    /** This metric indicates failure in collecting ClusterManagerServiceEventMetrics */
-    CLUSTER_MANAGER_METRICS_ERROR("ClusterManagerMetricsError"),
-
-    /** This metric indicates cluster_manager is not up */
-    CLUSTER_MANAGER_NODE_NOT_UP("ClusterManagerNodeNotUp"),
-
     /** This metric indicates faiure in intercepting opensearch requests at transport channel */
     OPENSEARCH_REQUEST_INTERCEPTOR_ERROR("OpenSearchRequestInterceptorError"),
 
+    /** Metrics tracking RCA Agent level: 1. Errors 2. Exceptions */
     /** This metric indicates metric entry insertion to event log queue failed */
     METRICS_WRITE_ERROR("MetricsWriteError", "namedCount", Statistics.NAMED_COUNTERS),
 
@@ -85,6 +82,9 @@ public enum ExceptionsAndErrors implements MeasurementSet {
     /** This metric indicates that error occurred while closing metrics db. */
     METRICS_DB_CLOSURE_ERROR("MetricsDbClosureError", "namedCount", Statistics.NAMED_COUNTERS),
 
+    /** When the reader encounters errors accessing metricsdb files. */
+    READER_METRICSDB_ACCESS_ERRORS("ReaderMetricsdbAccessError"),
+
     /** This metric indicates that error occurred while closing database connection. */
     IN_MEMORY_DATABASE_CONN_CLOSURE_ERROR(
             "InMemoryDatabaseConnClosureError", "namedCount", Statistics.NAMED_COUNTERS),
@@ -94,11 +94,13 @@ public enum ExceptionsAndErrors implements MeasurementSet {
     BATCH_METRICS_HTTP_HOST_ERROR("BatchMetricsHttpHostError"),
     BATCH_METRICS_EXCEEDED_MAX_DATAPOINTS("ExceededBatchMetricsMaxDatapoints"),
 
-    /** When the reader encounters errors accessing metricsdb files. */
-    READER_METRICSDB_ACCESS_ERRORS("ReaderMetricsdbAccessError"),
+    /** Below track Collector specific Errors. */
+    THREADPOOL_METRICS_COLLECTOR_ERROR("ThreadPoolMetricsCollectorError"),
     SHARD_STATE_COLLECTOR_ERROR("ShardStateCollectorError"),
     ADMISSION_CONTROL_COLLECTOR_ERROR("AdmissionControlCollectorError"),
     CIRCUIT_BREAKER_COLLECTOR_ERROR("CircuitBreakerCollectorError"),
+    CLUSTER_MANAGER_METRICS_ERROR("ClusterManagerMetricsError"),
+    CLUSTER_MANAGER_NODE_NOT_UP("ClusterManagerNodeNotUp"),
     CLUSTER_MANAGER_THROTTLING_COLLECTOR_ERROR("ClusterManagerThrottlingMetricsCollectorError"),
     FAULT_DETECTION_COLLECTOR_ERROR("FaultDetectionMetricsCollectorError"),
     CLUSTER_APPLIER_SERVICE_STATS_COLLECTOR_ERROR("ClusterApplierServiceStatsCollectorError"),

diff --git a/src/main/java/org/opensearch/performanceanalyzer/rca/framework/metrics/WriterMetrics.java b/src/main/java/org/opensearch/performanceanalyzer/rca/framework/metrics/WriterMetrics.java
@@ -111,8 +111,8 @@ public enum WriterMetrics implements MeasurementSet {
             "ClusterApplierServiceStatsCollectorExecutionTime",
             "millis",
             Arrays.asList(Statistics.MAX, Statistics.MEAN, Statistics.SUM)),
-    ADMISSION_CONTROL_METRICS_COLLECTOR_EXECUTION_TIME(
-            "AdmissionControlMetricsCollectorExecutionTime",
+    ADMISSION_CONTROL_COLLECTOR_EXECUTION_TIME(
+            "AdmissionControlCollectorExecutionTime",
             "millis",
             Arrays.asList(Statistics.MAX, Statistics.MEAN, Statistics.SUM)),
     ELECTION_TERM_COLLECTOR_EXECUTION_TIME(

diff --git a/src/main/java/org/opensearch/performanceanalyzer/reader/NodeMetricsEventProcessor.java b/src/main/java/org/opensearch/performanceanalyzer/reader/NodeMetricsEventProcessor.java
@@ -16,10 +16,12 @@
 import org.apache.logging.log4j.LogManager;
 import org.apache.logging.log4j.Logger;
 import org.jooq.BatchBindStep;
+import org.opensearch.performanceanalyzer.PerformanceAnalyzerApp;
 import org.opensearch.performanceanalyzer.collectors.StatExceptionCode;
 import org.opensearch.performanceanalyzer.collectors.StatsCollector;
 import org.opensearch.performanceanalyzer.metrics.AllMetrics;
 import org.opensearch.performanceanalyzer.metrics.PerformanceAnalyzerMetrics;
+import org.opensearch.performanceanalyzer.rca.framework.metrics.ExceptionsAndErrors;
 import org.opensearch.performanceanalyzer.reader_writer_shared.Event;
 import org.opensearch.performanceanalyzer.util.JsonConverter;
 import org.opensearch.performanceanalyzer.util.JsonPathNotFoundException;
@@ -157,7 +159,8 @@ private boolean processEvent(
                             "Malformed json (%s) ExceptionCode: %s",
                             lines[0], StatExceptionCode.JSON_PARSER_ERROR.toString()),
                     ex);
-            StatsCollector.instance().logException(StatExceptionCode.JSON_PARSER_ERROR);
+            PerformanceAnalyzerApp.ERRORS_AND_EXCEPTIONS_AGGREGATOR.updateStat(
+                    ExceptionsAndErrors.JSON_PARSER_ERROR, this.getClass().getSimpleName(), 1);
             return false;
         } catch (IOException ex) {
             LOG.warn(