Java Antithesis log before assertion

Author: robertomonteromiguel

dd-trace-core/src/main/java/datadog/trace/common/writer/PayloadDispatcherImpl.java

@@ -66,6 +66,7 @@ public void onDroppedTrace(int spanCount) {
     dropDetails.put("total_dropped_traces", droppedTraceCount.sum() + 1);
     dropDetails.put("total_dropped_spans", droppedSpanCount.sum() + spanCount);
 
+    log.debug("ANTITHESIS_ASSERT: Traces dropped before sending (unreachable) - span_count: {}, total_dropped: {}", spanCount, droppedTraceCount.sum() + 1);
     Assert.unreachable(
         "Traces should not be dropped before attempting to send - indicates buffer overflow or backpressure",
         dropDetails);
@@ -117,7 +118,9 @@ public void accept(int messageCount, ByteBuffer buffer) {
     // or when the packer is flushed at a heartbeat
     if (messageCount > 0) {
       // Antithesis: Verify that we're attempting to send traces
+      log.debug("ANTITHESIS_ASSERT: Trace sending code path exercised (reachable) - message_count: {}", messageCount);
       Assert.reachable("Trace sending code path is exercised", null);
+      log.debug("ANTITHESIS_ASSERT: Checking if traces are being sent to API (sometimes) - message_count: {}", messageCount);
       Assert.sometimes(
           messageCount > 0,
           "Traces are being sent to the API",
@@ -141,6 +144,7 @@ public void accept(int messageCount, ByteBuffer buffer) {
       });
       response.status().ifPresent(status -> sendDetails.put("http_status", status));
 
+      log.debug("ANTITHESIS_ASSERT: Checking trace sending success (always) - success: {}, trace_count: {}", response.success(), messageCount);
       Assert.always(
           response.success(),
           "Trace sending to API should always succeed - no traces should be lost",
@@ -162,6 +166,7 @@ public void accept(int messageCount, ByteBuffer buffer) {
         });
         response.status().ifPresent(status -> failureDetails.put("http_status", status));
 
+        log.debug("ANTITHESIS_ASSERT: Trace sending failed (unreachable) - trace_count: {}, size: {} bytes", messageCount, sizeInBytes);
         Assert.unreachable(
             "Trace sending failure path should never be reached - indicates traces are being lost",
             failureDetails);

dd-trace-core/src/main/java/datadog/trace/common/writer/RemoteWriter.java

@@ -76,6 +76,7 @@ public void write(final List<DDSpan> trace) {
     writeAttemptDetails.put("trace_size", trace.size());
     writeAttemptDetails.put("has_traces", !trace.isEmpty());
 
+    log.debug("ANTITHESIS_ASSERT: Checking writer not closed when writing (always) - closed: {}, trace_size: {}", closed, trace.size());
     Assert.always(
         !closed,
         "Writer should never be closed when attempting to write traces",
@@ -90,6 +91,7 @@ public void write(final List<DDSpan> trace) {
       shutdownDetails.put("trace_size", trace.size());
       shutdownDetails.put("reason", "writer_closed_during_shutdown");
 
+      log.debug("ANTITHESIS_ASSERT: Traces dropped due to shutdown (sometimes) - closed: {}, trace_size: {}", closed, trace.size());
       Assert.sometimes(
           closed && !trace.isEmpty(),
           "Traces are dropped due to writer shutdown - tracking shutdown behavior",
@@ -123,6 +125,7 @@ public void write(final List<DDSpan> trace) {
             overflowDetails.put("buffer_capacity", traceProcessingWorker.getCapacity());
             overflowDetails.put("reason", "buffer_overflow_backpressure");
 
+            log.debug("ANTITHESIS_ASSERT: Buffer overflow occurred (unreachable) - trace_size: {}, capacity: {}", trace.size(), traceProcessingWorker.getCapacity());
             Assert.unreachable(
                 "Buffer overflow should never occur - traces are being dropped due to backpressure",
                 overflowDetails);

dd-trace-core/src/main/java/datadog/trace/common/writer/ddagent/DDAgentApi.java

@@ -94,7 +94,9 @@ public Response sendSerializedTraces(final Payload payload) {
     final int sizeInBytes = payload.sizeInBytes();
 
     // Antithesis: Track that agent API send is being exercised
+    log.debug("ANTITHESIS_ASSERT: Verifying DDAgentApi trace sending is exercised (reachable) with {} traces", payload.traceCount());
     Assert.reachable("DDAgentApi trace sending is exercised", null);
+    log.debug("ANTITHESIS_ASSERT: Checking if traces are being sent through DDAgentApi (sometimes) - count: {}", payload.traceCount());
     Assert.sometimes(
         payload.traceCount() > 0,
         "Traces are being sent through DDAgentApi",
@@ -112,6 +114,7 @@ public Response sendSerializedTraces(final Payload payload) {
         agentDetectionDetails.put("agent_url", agentUrl.toString());
         agentDetectionDetails.put("failure_reason", "agent_not_detected");
 
+        log.debug("ANTITHESIS_ASSERT: Agent not detected (unreachable) - url: {}, traces: {}", agentUrl, payload.traceCount());
         Assert.unreachable(
             "Datadog agent should always be detected - agent communication failure",
             agentDetectionDetails);
@@ -154,6 +157,7 @@ public Response sendSerializedTraces(final Payload payload) {
         httpResponseDetails.put("success", response.code() == 200);
         httpResponseDetails.put("agent_url", tracesUrl.toString());
 
+        log.debug("ANTITHESIS_ASSERT: Checking HTTP response status (always) - code: {}, traces: {}", response.code(), payload.traceCount());
         Assert.always(
             response.code() == 200,
             "HTTP response from Datadog agent should always be 200 - API communication failure",
@@ -168,6 +172,7 @@ public Response sendSerializedTraces(final Payload payload) {
           errorDetails.put("http_message", response.message());
           errorDetails.put("failure_reason", "http_error_response");
 
+          log.debug("ANTITHESIS_ASSERT: Non-200 HTTP response (unreachable) - code: {}, message: {}, traces: {}", response.code(), response.message(), payload.traceCount());
           Assert.unreachable(
               "Non-200 HTTP response from agent indicates API failure - traces may be lost",
               errorDetails);
@@ -204,6 +209,7 @@ public Response sendSerializedTraces(final Payload payload) {
       networkErrorDetails.put("agent_url", agentUrl.toString());
       networkErrorDetails.put("failure_reason", "network_io_exception");
 
+      log.debug("ANTITHESIS_ASSERT: Network/IO exception (unreachable) - type: {}, message: {}, traces: {}", e.getClass().getName(), e.getMessage(), payload.traceCount());
       Assert.unreachable(
           "Network/IO exceptions should not occur when sending to agent - indicates connectivity issues",
           networkErrorDetails);

telemetry/src/main/java/datadog/telemetry/TelemetryClient.java

@@ -99,6 +99,7 @@ public Result sendHttpRequest(Request.Builder httpRequestBuilder) {
     String requestType = httpRequest.header(DD_TELEMETRY_REQUEST_TYPE);
 
     // Antithesis: Track telemetry sending attempts
+    log.debug("ANTITHESIS_ASSERT: Telemetry sending exercised (reachable) - request_type: {}", requestType);
     Assert.reachable("Telemetry sending is exercised", null);
 
     try (okhttp3.Response response =
@@ -119,6 +120,7 @@ public Result sendHttpRequest(Request.Builder httpRequestBuilder) {
         notFoundDetails.put("url", url.toString());
         notFoundDetails.put("reason", "endpoint_disabled_404");
 
+        log.debug("ANTITHESIS_ASSERT: Telemetry endpoint 404 (sometimes) - request_type: {}, url: {}", requestType, url);
         Assert.sometimes(
             response.code() == 404,
             "Telemetry endpoint returns 404 - endpoint may be disabled",
@@ -137,6 +139,7 @@ public Result sendHttpRequest(Request.Builder httpRequestBuilder) {
         failureDetails.put("url", url.toString());
         failureDetails.put("reason", "http_error_response");
 
+        log.debug("ANTITHESIS_ASSERT: Telemetry HTTP request failed (unreachable) - request_type: {}, status: {}", requestType, response.code());
         Assert.unreachable(
             "Telemetry HTTP request failed - telemetry data should not be dropped, should retry",
             failureDetails);
@@ -150,6 +153,7 @@ public Result sendHttpRequest(Request.Builder httpRequestBuilder) {
       }
 
       // Antithesis: Assert success
+      log.debug("ANTITHESIS_ASSERT: Checking telemetry request success (always) - successful: {}, request_type: {}", response.isSuccessful(), requestType);
       Assert.always(
           response.isSuccessful(),
           "Telemetry requests should always succeed - no telemetry data should be lost",
@@ -168,6 +172,7 @@ public Result sendHttpRequest(Request.Builder httpRequestBuilder) {
       ioErrorDetails.put("url", url.toString());
       ioErrorDetails.put("reason", "network_io_exception");
 
+      log.debug("ANTITHESIS_ASSERT: Telemetry network/IO exception (unreachable) - request_type: {}, exception: {}", requestType, e.getClass().getName());
       Assert.unreachable(
           "Telemetry network/IO failure - telemetry data should not be dropped, should retry",
           ioErrorDetails);

telemetry/src/main/java/datadog/telemetry/TelemetryRouter.java

@@ -53,6 +53,7 @@ public TelemetryClient.Result sendRequest(TelemetryRequest request) {
     routingDetails.put("has_fallback", intakeClient != null);
     routingDetails.put("url", currentClient.getUrl().toString());
 
+    log.debug("ANTITHESIS_ASSERT: Checking telemetry routing success (always) - result: {}, client: {}", result, currentClient == agentClient ? "agent" : "intake");
     Assert.always(
         result == TelemetryClient.Result.SUCCESS || result == TelemetryClient.Result.INTERRUPTED,
         "Telemetry routing should always succeed - failures indicate data loss without retry mechanism",
@@ -67,6 +68,7 @@ public TelemetryClient.Result sendRequest(TelemetryRequest request) {
         agentFailureDetails.put("has_intake_fallback", intakeClient != null);
         agentFailureDetails.put("reason", "agent_telemetry_failure");
 
+        log.debug("ANTITHESIS_ASSERT: Agent telemetry endpoint failed (unreachable) - result: {}, has_fallback: {}", result, intakeClient != null);
         Assert.unreachable(
             "Agent telemetry endpoint failed - switching to intake but current request data is lost",
             agentFailureDetails);
@@ -87,6 +89,7 @@ public TelemetryClient.Result sendRequest(TelemetryRequest request) {
         intakeFailureDetails.put("will_fallback_to_agent", true);
         intakeFailureDetails.put("reason", "intake_telemetry_failure");
 
+        log.debug("ANTITHESIS_ASSERT: Intake telemetry endpoint failed (unreachable) - result: {}, will_fallback: true", result);
         Assert.unreachable(
             "Intake telemetry endpoint failed - switching to agent but current request data is lost",
             intakeFailureDetails);