skywalking trace跟踪原理：
1.先理解trace的几个概念，EntrySpan EntrySpan ExitSpan
2.以百度的 brpc插件为例
ClientInterceptor
@Override
public void beforeMethod(EnhancedInstance objInst, Method method, Object[] allArguments, Class<?>[] argumentsTypes, MethodInterceptResult result) throws Throwable {
Request request = (Request) allArguments[0];
InetSocketAddress remoteAddress = (InetSocketAddress) request.getChannel().remoteAddress();
InetAddress address = remoteAddress.getAddress();
final ContextCarrier contextCarrier = new ContextCarrier();
# createExitSpan创建本地线程变量，底层是使用的 threadLocal技术
AbstractSpan span = ContextManager.createExitSpan(generateOperationName(request), contextCarrier, address.getHostAddress() + ":" + remoteAddress.getPort());
CarrierItem next = contextCarrier.items();
while (next.hasNext()) {
next = next.next();
if (request.getKvAttachment() == null) {
request.setKvAttachment(new HashMap<>());
}
# 将rpc请求放到request的请求头中，服务端获取就会从请求头中获取
request.getKvAttachment().put(next.getHeadKey(), next.getHeadValue());
}
span.setComponent(ComponentsDefine.BRPC_JAVA);
SpanLayer.asRPCFramework(span);
}
private static AbstractTracerContext getOrCreate(String operationName, boolean forceSampling) {
#这里的get就是调用的ThreadLocal的get方法获取本线程的ThreadLcalMap对象，将来里面会放[{threadLocal1,value1},{threadLocal2,value2}],比如就可以放traceId
AbstractTracerContext context = CONTEXT.get();
if (context == null) {
if (StringUtil.isEmpty(operationName)) {
if (LOGGER.isDebugEnable()) {
LOGGER.debug("No operation name, ignore this trace.");
}
context = new IgnoredTracerContext();
} else {
if (EXTEND_SERVICE == null) {
EXTEND_SERVICE = ServiceManager.INSTANCE.findService(ContextManagerExtendService.class);
}
context = EXTEND_SERVICE.createTraceContext(operationName, forceSampling);
}
CONTEXT.set(context);
}
return context;
}
ServerInterceptor
public class ServerInterceptor implements InstanceMethodsAroundInterceptor {
@Override
public void beforeMethod(EnhancedInstance objInst, Method method, Object[] allArguments, Class<?>[] argumentsTypes,
MethodInterceptResult result) throws Throwable {
Request request = (Request) allArguments[0];
ContextCarrier contextCarrier = new ContextCarrier();
CarrierItem next = contextCarrier.items();
if (request.getKvAttachment() == null) {
request.setKvAttachment(new HashMap<>());
}
while (next.hasNext()) {
next = next.next();
#将客户端带过来的request的getHeadKey放到自己的本地线程ThreadLocalMap中(通过createEntrySpan创建)
next.setHeadValue((String) request.getKvAttachment().get(next.getHeadKey()));
}
AbstractSpan span = ContextManager.createEntrySpan(generateOperationName(request), contextCarrier);
SpanLayer.asRPCFramework(span);
span.setComponent(ComponentsDefine.BRPC_JAVA);
}
。。。
}
3.springmvc是如何完成跟踪的？
原理是一样的
public class InvokeInterceptor implements InstanceMethodsAroundInterceptor {
@Override
public void beforeMethod(final EnhancedInstance objInst,
final Method method,
final Object[] allArguments,
final Class<?>[] argumentsTypes,
final MethodInterceptResult result) throws Throwable {
ServerWebExchange exchange = (ServerWebExchange) allArguments[0];
ContextManager.getRuntimeContext()
.put(RESPONSE_KEY_IN_RUNTIME_CONTEXT, new ReactiveResponseHolder(exchange.getResponse()));
ContextManager.getRuntimeContext()
.put(REQUEST_KEY_IN_RUNTIME_CONTEXT, new ReactiveRequestHolder(exchange.getRequest()));
}
4.kafka的原理呢？
KafkaProducerInterceptor
你会发现底层还是放到head里面
public class KafkaProducerInterceptor implements InstanceMethodsAroundInterceptor {
public static final String OPERATE_NAME_PREFIX = "Kafka/";
public static final String PRODUCER_OPERATE_NAME_SUFFIX = "/Producer";
@Override
public void beforeMethod(EnhancedInstance objInst, Method method, Object[] allArguments, Class<?>[] argumentsTypes,
MethodInterceptResult result) throws Throwable {
ContextCarrier contextCarrier = new ContextCarrier();
ProducerRecord record = (ProducerRecord) allArguments[0];
String topicName = record.topic();
AbstractSpan activeSpan = ContextManager.createExitSpan(OPERATE_NAME_PREFIX + topicName + PRODUCER_OPERATE_NAME_SUFFIX, contextCarrier, (String) objInst
.getSkyWalkingDynamicField());
Tags.MQ_BROKER.set(activeSpan, (String) objInst.getSkyWalkingDynamicField());
Tags.MQ_TOPIC.set(activeSpan, topicName);
contextCarrier.extensionInjector().injectSendingTimestamp();
SpanLayer.asMQ(activeSpan);
activeSpan.setComponent(ComponentsDefine.KAFKA_PRODUCER);
CarrierItem next = contextCarrier.items();
while (next.hasNext()) {
next = next.next();
record.headers().add(next.getHeadKey(), next.getHeadValue().getBytes(StandardCharsets.UTF_8));
}
//when use lambda expression, not to generate inner class,
// and not to trigger kafka CallBack class define, so allArguments[1] can't to cast EnhancedInstance
Object shouldCallbackInstance = allArguments[1];
if (null != shouldCallbackInstance) {
if (shouldCallbackInstance instanceof EnhancedInstance) {
EnhancedInstance callbackInstance = (EnhancedInstance) shouldCallbackInstance;
ContextSnapshot snapshot = ContextManager.capture();
if (null != snapshot) {
CallbackCache cache = new CallbackCache();
cache.setSnapshot(snapshot);
callbackInstance.setSkyWalkingDynamicField(cache);
}
} else if (shouldCallbackInstance instanceof Callback) {
Callback callback = (Callback) shouldCallbackInstance;
ContextSnapshot snapshot = ContextManager.capture();
if (null != snapshot) {
CallbackCache cache = new CallbackCache();
cache.setSnapshot(snapshot);
cache.setCallback(callback);
allArguments[1] = new CallbackAdapterInterceptor(cache);
}
}
}
}
KafkaConsumerInterceptor
也是从record的header中获取的，完美搞定。
@Override
public Object afterMethod(EnhancedInstance objInst, Method method, Object[] allArguments, Class<?>[] argumentsTypes,
Object ret) throws Throwable {
/*
* If the intercepted method throws exception, the ret will be null
*/
if (ret == null) {
return ret;
}
Map<TopicPartition, List<ConsumerRecord<?, ?>>> records = (Map<TopicPartition, List<ConsumerRecord<?, ?>>>) ret;
//
// The entry span will only be created when the consumer received at least one message.
//
if (records.size() > 0) {
ConsumerEnhanceRequiredInfo requiredInfo = (ConsumerEnhanceRequiredInfo) objInst.getSkyWalkingDynamicField();
KafkaContext context = (KafkaContext) ContextManager.getRuntimeContext().get(Constants.KAFKA_FLAG);
if (context != null) {
ContextManager.createEntrySpan(context.getOperationName(), null);
context.setNeedStop(true);
}
String operationName = OPERATE_NAME_PREFIX + requiredInfo.getTopics() + CONSUMER_OPERATE_NAME + requiredInfo.getGroupId();
AbstractSpan activeSpan = ContextManager.createEntrySpan(operationName, null).start(requiredInfo.getStartTime());
activeSpan.setComponent(ComponentsDefine.KAFKA_CONSUMER);
SpanLayer.asMQ(activeSpan);
Tags.MQ_BROKER.set(activeSpan, requiredInfo.getBrokerServers());
Tags.MQ_TOPIC.set(activeSpan, requiredInfo.getTopics());
for (List<ConsumerRecord<?, ?>> consumerRecords : records.values()) {
for (ConsumerRecord<?, ?> record : consumerRecords) {
ContextCarrier contextCarrier = new ContextCarrier();
CarrierItem next = contextCarrier.items();
while (next.hasNext()) {
next = next.next();
Iterator<Header> iterator = record.headers().headers(next.getHeadKey()).iterator();
if (iterator.hasNext()) {
next.setHeadValue(new String(iterator.next().value(), StandardCharsets.UTF_8));
}
}
ContextManager.extract(contextCarrier);
}
}
ContextManager.stopSpan();
}
return ret;
}