Disable context propagation on kprobes (#556)

bpf/http_sock.h

@@ -285,10 +285,7 @@ static __always_inline void handle_buf_with_connection(pid_connection_info_t *pi
             http_connection_metadata_t *meta = bpf_map_lookup_elem(&filtered_connections, pid_conn);
             get_or_create_trace_info(meta, pid_conn->pid, &pid_conn->conn, u_buf, bytes_len, capture_header_buffer);
 
-            if (!meta) {
-                bpf_dbg_printk("No META!");
-            }
-
+#if 0 // disabled for now, until we have better way to detect when it's safe to assume same thread
             if (meta) {            
                 tp_info_pid_t *tp_p = trace_info_for_connection(&pid_conn->conn);
                 if (tp_p) {
@@ -309,7 +306,10 @@ static __always_inline void handle_buf_with_connection(pid_connection_info_t *pi
                 } else {
                     bpf_dbg_printk("Can't find trace info, this is a bug!");
                 }
+            } else {
+                bpf_dbg_printk("No META!");
             }
+#endif            
             // we copy some small part of the buffer to the info trace event, so that we can process an event even with
             // incomplete trace info in user space.
             bpf_probe_read(info->buf, FULL_BUF_SIZE, u_buf);

bpf/trace_common.h

@@ -160,6 +160,7 @@ static __always_inline void get_or_create_trace_info(http_connection_metadata_t
 
     u8 found_tp = 0;
 
+#if 0 // disabled for now, until we have better way to detect when it's safe to assume same thread
     if (meta) {
         if (meta->type == EVENT_HTTP_CLIENT) {
             tp_p->pid = -1; // we only want to prevent correlation of duplicate server calls by PID
@@ -183,7 +184,8 @@ static __always_inline void get_or_create_trace_info(http_connection_metadata_t
             } 
         }
     }
-
+#endif    
+
     if (!found_tp) {
         bpf_dbg_printk("Generating new traceparent id");
         urand_bytes(tp_p->tp.trace_id, TRACE_ID_SIZE_BYTES);        

test/integration/multiprocess_test.go

@@ -86,13 +86,15 @@ func TestMultiProcess(t *testing.T) {
 		assert.Empty(t, results)
 	})
 
-	t.Run("Nested traces with kprobes: node python rails", func(t *testing.T) {
-		testNestedHTTPTracesKProbes(t)
-	})
-
-	t.Run("Nested traces with kprobes: SSL node python rails", func(t *testing.T) {
-		testNestedHTTPSTracesKProbes(t)
-	})
+	if kprobeTraces {
+		t.Run("Nested traces with kprobes: node python rails", func(t *testing.T) {
+			testNestedHTTPTracesKProbes(t)
+		})
+
+		t.Run("Nested traces with kprobes: SSL node python rails", func(t *testing.T) {
+			testNestedHTTPSTracesKProbes(t)
+		})
+	}
 
 	t.Run("BPF pinning folders mounted", func(t *testing.T) {
 		// 1 beyla pinned map folder for all processes

test/integration/red_test_nodeclient.go

@@ -84,7 +84,9 @@ func testNodeClientWithMethodAndStatusCode(t *testing.T, method string, statusCo
 	 We check that the traceID has that 16 character 0 suffix and then we
 	 use the first 16 characters for looking up by Parent span.
 	*/
-	require.True(t, span.TraceID != "")
-	require.True(t, strings.HasSuffix(span.TraceID, traceIDLookup))
-	require.True(t, strings.HasPrefix(span.SpanID, "00"))
+	if kprobeTraces {
+		require.True(t, span.TraceID != "")
+		require.True(t, strings.HasSuffix(span.TraceID, traceIDLookup))
+		require.True(t, strings.HasPrefix(span.SpanID, "00"))
+	}
 }

test/integration/red_test_rust.go

@@ -85,10 +85,12 @@ func testREDMetricsForRustHTTPLibrary(t *testing.T, url, comm, namespace string,
 	require.Len(t, res, 1)
 	parent := res[0]
 	require.NotEmpty(t, parent.TraceID)
-	require.Equal(t, traceID, parent.TraceID)
-	// Validate that "parent" is a CHILD_OF the traceparent's "parent-id"
-	childOfPID := trace.ChildrenOf(parentID)
-	require.Len(t, childOfPID, 1)
+	if kprobeTraces {
+		require.Equal(t, traceID, parent.TraceID)
+		// Validate that "parent" is a CHILD_OF the traceparent's "parent-id"
+		childOfPID := trace.ChildrenOf(parentID)
+		require.Len(t, childOfPID, 1)
+	}
 	require.NotEmpty(t, parent.SpanID)
 	// check duration is at least 2us
 	assert.Less(t, (2 * time.Microsecond).Microseconds(), parent.Duration)

test/integration/suites_test.go

@@ -14,6 +14,8 @@ import (
 	"github.com/grafana/beyla/test/integration/components/docker"
 )
 
+var kprobeTraces = false // disabled for now until we figure a safe way to determine server client relationships
+
 func TestSuite(t *testing.T) {
 	compose, err := docker.ComposeSuite("docker-compose.yml", path.Join(pathOutput, "test-suite.log"))
 	require.NoError(t, err)

test/integration/traces_test.go

@@ -334,10 +334,12 @@ func testHTTPTracesKProbes(t *testing.T) {
 	require.Len(t, res, 1)
 	parent := res[0]
 	require.NotEmpty(t, parent.TraceID)
-	require.Equal(t, traceID, parent.TraceID)
-	// Validate that "parent" is a CHILD_OF the traceparent's "parent-id"
-	childOfPID := trace.ChildrenOf(parentID)
-	require.Len(t, childOfPID, 1)
+	if kprobeTraces {
+		require.Equal(t, traceID, parent.TraceID)
+		// Validate that "parent" is a CHILD_OF the traceparent's "parent-id"
+		childOfPID := trace.ChildrenOf(parentID)
+		require.Len(t, childOfPID, 1)
+	}
 	require.NotEmpty(t, parent.SpanID)
 	// check duration is at least 2us
 	assert.Less(t, (2 * time.Microsecond).Microseconds(), parent.Duration)