[ML] Change logging level used in NLP request processing (#98497)

Author: davidkyle

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/inference/deployment/InferencePyTorchAction.java

@@ -90,7 +90,7 @@ protected void doRun() throws Exception {
             NlpConfig nlpConfig = (NlpConfig) config;
             NlpTask.Request request = processor.getRequestBuilder(nlpConfig)
                 .buildRequest(text, requestIdStr, nlpConfig.getTokenization().getTruncate(), nlpConfig.getTokenization().getSpan());
-            logger.trace(() -> format("handling request [%s]", requestIdStr));
+            logger.debug(() -> format("handling request [%s]", requestIdStr));
 
             // Tokenization is non-trivial, so check for cancellation one last time before sending request to the native process
             if (isCancelled()) {
@@ -110,6 +110,7 @@ protected void doRun() throws Exception {
             logger.error(() -> "[" + getDeploymentId() + "] error writing to inference process", e);
             onFailure(ExceptionsHelper.serverError("Error writing to inference process", e));
         } catch (Exception e) {
+            logger.error(() -> "[" + getDeploymentId() + "] error running inference", e);
             onFailure(e);
         }
     }
@@ -141,7 +142,7 @@ private void processResult(
             return;
         }
         InferenceResults results = inferenceResultsProcessor.processResult(tokenization, pyTorchResult.inferenceResult());
-        logger.trace(() -> format("[%s] processed result for request [%s]", getDeploymentId(), getRequestId()));
+        logger.debug(() -> format("[%s] processed result for request [%s]", getDeploymentId(), getRequestId()));
         onSuccess(results);
     }
 

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/inference/pytorch/process/PyTorchResultProcessor.java

@@ -149,7 +149,7 @@ void processInferenceResult(PyTorchResult result) {
             timeMs = 0L;
         }
 
-        logger.trace(() -> format("[%s] Parsed inference result with id [%s]", modelId, result.requestId()));
+        logger.debug(() -> format("[%s] Parsed inference result with id [%s]", modelId, result.requestId()));
         updateStats(timeMs, Boolean.TRUE.equals(result.isCacheHit()));
         PendingResult pendingResult = pendingResults.remove(result.requestId());
         if (pendingResult == null) {
@@ -163,7 +163,7 @@ void processThreadSettings(PyTorchResult result) {
         ThreadSettings threadSettings = result.threadSettings();
         assert threadSettings != null;
 
-        logger.trace(() -> format("[%s] Parsed thread settings result with id [%s]", modelId, result.requestId()));
+        logger.debug(() -> format("[%s] Parsed thread settings result with id [%s]", modelId, result.requestId()));
         PendingResult pendingResult = pendingResults.remove(result.requestId());
         if (pendingResult == null) {
             logger.debug(() -> format("[%s] no pending result for thread settings [%s]", modelId, result.requestId()));
@@ -176,7 +176,7 @@ void processAcknowledgement(PyTorchResult result) {
         AckResult ack = result.ackResult();
         assert ack != null;
 
-        logger.trace(() -> format("[%s] Parsed ack result with id [%s]", modelId, result.requestId()));
+        logger.debug(() -> format("[%s] Parsed ack result with id [%s]", modelId, result.requestId()));
         PendingResult pendingResult = pendingResults.remove(result.requestId());
         if (pendingResult == null) {
             logger.debug(() -> format("[%s] no pending result for ack [%s]", modelId, result.requestId()));
@@ -194,7 +194,7 @@ void processErrorResult(PyTorchResult result) {
             errorCount++;
         }
 
-        logger.trace(() -> format("[%s] Parsed error with id [%s]", modelId, result.requestId()));
+        logger.debug(() -> format("[%s] Parsed error with id [%s]", modelId, result.requestId()));
         PendingResult pendingResult = pendingResults.remove(result.requestId());
         if (pendingResult == null) {
             logger.debug(() -> format("[%s] no pending result for error [%s]", modelId, result.requestId()));