使用Apache Camel Multicast组件遇到的一个问题
1 前言
本文翻译自华为加拿大研究所的Reji Mathews发表于Apache Camel社区的《ROUTING MULTICAST OUTPUT AFTER ENCOUNTERING PARTIAL FAILURES》一文。在征得原作者同意后，本文对原文的部分内容作了少许修改。
2 Multicast组件简介
Multicast是Apache Camel（以下简称“Camel”）中一个功能强大的EIP组件，可以将消息发送至多条子路径，然后并行地执行它们。
参考官网文档，我们可以使用两种方式配置Multicast组件：
- 独立执行所有子路径，并将最后响应的子路径的结果作为最终输出。这也是Multicast组件的默认配置。
- 通过实现Camel的聚合策略（AggregationStrategy），使用自定义的聚合器来处理所有子路径的输出。
3 问题描述
本文使用案例如下：使用Jetty组件发布一个API，调用该API后，消息会分别发送至"direct:A"和"direct:B"两条子路径。在使用自定义的聚合策略处理后，继续执行后续步骤。其中在"direct:A"中抛出一个异常，来模拟运行失败；"direct:B"正常运行。同时在onException中定义了异常处理策略。
本文使用的Camel版本为3.8.0
@Override
public void configure() throws Exception {
onException(Exception.class)
.useOriginalMessage()
.handled(true)
.log("Exception handler invoked")
.transform().constant("{\"data\" : \"err\"}")
.end();
from("jetty:http://localhost:8081/myapi?httpMethodRestrict=GET")
.log("received request")
.log("Entering multicast")
.multicast(new SimpleFlowMergeAggregator())
.parallelProcessing().to("direct:A", "direct:B")
.end()
.log("Aggregated results ${body}")
.log("Another log")
.transform(simple("{\"result\" : \"success\"}"))
.end();
from("direct:A")
.log("Executing PATH_1 - exception path")
.transform(constant("DATA_FROM_PATH_1"))
.log("Starting exception throw")
.throwException(new Exception("USER INITIATED EXCEPTION"))
.log("PATH_1")
.end();
from("direct:B")
.log("Executing PATH_2 - success path")
.delayer(1000)
.transform(constant("DATA_FROM_PATH_2"))
.log("PATH_2")
.end();
}
自定义聚合器SimpleFlowMergeAggregator定义如下，其中我们将所有子路径的结果放入一个list对象。
public class SimpleFlowMergeAggregator implements AggregationStrategy {
private static final Logger LOGGER = LoggerFactory.getLogger(SimpleFlowMergeAggregator.class.getName());
@Override
public Exchange aggregate(Exchange oldExchange, Exchange newExchange) {
LOGGER.info("Inside aggregator " + newExchange.getIn().getBody());
if(oldExchange == null) {
String data = newExchange.getIn().getBody(String.class);
List<String> aggregatedDataList = new ArrayList<>();
aggregatedDataList.add(data);
newExchange.getIn().setBody(aggregatedDataList);
return newExchange;
}
List<String> oldData = oldExchange.getIn().getBody(List.class);
oldData.add(newExchange.getIn().getBody(String.class));
oldExchange.getIn().setBody(oldData);
return oldExchange;
}
}
基于对Multicast组件执行逻辑的理解，我们认为存在多个子路径时，其运行结果应该为：如果其中有一条子路径能运行成功，则使用聚合的结果继续执行后续步骤；如果所有子路径都运行失败，则停止整个路由（route）。本案例中，由于子路径"direct:A"运行异常，子路径"direct:B"运行正常，则应该正常执行后续两个步骤日志（log）和转换（transform）。
运行上述案例，日志信息如下：
2021-05-06 12:43:18.565 INFO 13956 --- [qtp916897446-42] route1 : received request
2021-05-06 12:43:18.566 INFO 13956 --- [qtp916897446-42] route1 : Entering multicast
2021-05-06 12:43:18.575 INFO 13956 --- [ #4 - Multicast] route2 : Executing PATH_1 - exception path
2021-05-06 12:43:18.575 INFO 13956 --- [ #4 - Multicast] route2 : Starting exception throw
2021-05-06 12:43:18.578 INFO 13956 --- [ #4 - Multicast] route2 : Exception handler invoked
2021-05-06 12:43:18.579 INFO 13956 --- [ #4 - Multicast] c.e.d.m.SimpleFlowMergeAggregator : Inside aggregator {"data" : "err"}
2021-05-06 12:43:19.575 INFO 13956 --- [ #3 - Multicast] route3 : Executing PATH_2 - success path
2021-05-06 12:43:21.576 INFO 13956 --- [ #3 - Multicast] route3 : PATH_2
2021-05-06 12:43:21.576 INFO 13956 --- [ #3 - Multicast] c.e.d.m.SimpleFlowMergeAggregator : Inside aggregator DATA_FROM_PATH_2
观察上述日志，我们发现完成两条子路径结果的聚合后，后续的两个步骤日志（log）和转换（transform）并未执行。这并不符合我们期望的结果。
经过多次测试，我们还发现，只有当到达聚合器SimpleFlowMergeAggregator的第一个子路径（"direct:A"）执行异常时，便会发生这种后续步骤未执行的情况；而如果第一个子路径（"direct:A"）执行成功，即使另一个子路径（"direct:B"）执行失败，也会继续执行后续的步骤。
4 问题分析
接下来，我们通过查看Camel源代码，来找出上述现象的原因。
在camel-core-processors模块的Pipeline.java 中，其run()方法中有这样一段代码：
@Override
public void run() {
boolean stop = exchange.isRouteStop();
int num = index;
boolean more = num < size;
boolean first = num == 0;
if (!stop && more && (first || continueProcessing(exchange, "so breaking out of pipeline", LOG))) {
// prepare for next run
if (exchange.hasOut()) {
exchange.setIn(exchange.getOut());
exchange.setOut(null);
}
// get the next processor
AsyncProcessor processor = processors.get(index++);
processor.process(exchange, this);
} else {
// copyResults is needed in case MEP is OUT and the message is not an OUT message
ExchangeHelper.copyResults(exchange, exchange);
// logging nextExchange as it contains the exchange that might have altered the payload and since
// we are logging the completion if will be confusing if we log the original instead
// we could also consider logging the original and the nextExchange then we have *before* and *after* snapshots
if (LOG.isTraceEnabled()) {
LOG.trace("Processing complete for exchangeId: {} >>> {}", exchange.getExchangeId(), exchange);
}
AsyncCallback cb = callback;
taskFactory.release(this);
reactiveExecutor.schedule(cb);
}
}
其中，这个if判断决定了是否继续执行后续步骤：
if (!stop && more && (first || continueProcessing(exchange, "so breaking out of pipeline", LOG)))
可以看出，在如下三种情况下，后续步骤将不会被执行：
1. 之前的步骤已经将exchange 对象标记为停止状态。
boolean stop = exchange.isRouteStop();
2. 后续没有步骤可执行。
boolean more = num < size;
3. continueProcessing()方法返回false。
我们来看看continueProcessing()方法的代码。
public final class PipelineHelper {
public static boolean continueProcessing(Exchange exchange, String message, Logger log) {
ExtendedExchange ee = (ExtendedExchange) exchange;
boolean stop = ee.isFailed() || ee.isRollbackOnly() || ee.isRollbackOnlyLast()
|| (ee.isErrorHandlerHandledSet() && ee.isErrorHandlerHandled());
if (stop) {
if (log.isDebugEnabled()) {
StringBuilder sb = new StringBuilder();
sb.append("Message exchange has failed: ").append(message).append(" for exchange: ").append(exchange);
if (exchange.isRollbackOnly() || exchange.isRollbackOnlyLast()) {
sb.append(" Marked as rollback only.");
}
if (exchange.getException() != null) {
sb.append(" Exception: ").append(exchange.getException());
}
if (ee.isErrorHandlerHandledSet() && ee.isErrorHandlerHandled()) {
sb.append(" Handled by the error handler.");
}
log.debug(sb.toString());
}
return false;
}
if (ee.isRouteStop()) {
if (log.isDebugEnabled()) {
log.debug("ExchangeId: {} is marked to stop routing: {}", exchange.getExchangeId(), exchange);
}
return false;
}
return true;
}
}
可以看出，当执行过程发生异常并且被异常处理器捕获时，continueProcessing()方法将返回false。
再回到我们的案例，第一个到达聚合器SimpleFlowMergeAggregator的子路径（"direct:A"），会作为后续聚合的基础，其它子路径（"direct:B"）会在此基础上追加各自的body数据。实际上，很多Camel用户都会采用这种方式来实现自定义聚合策略。但这样做存在一个问题：在异常处理时，子路径"direct:A"的exchange对象会被设置一个状态标识，而此状态标识会被传递到下游，用于判断是否继续执行后续步骤。由于作为聚合基础的"direct:A"子路径的exchange对象状态为“异常”，最终continueProcessing()方法将返回false，后续的步骤也就不会再执行。
5 解决方案
对于上述问题，用户可以使用多种方式来设置异常处理时exchange对象的状态。本文采用如下解决方案：如果第一个子路径执行正常，则继续执行后续步骤；如果第一个子路径执行异常，则将其与其它执行成功的子路径交换，然后继续执行后续步骤。
更新后的自定义聚合器SimpleFlowMergeAggregator如下：
public class SimpleFlowMergeAggregator implements AggregationStrategy {
private static final Logger LOGGER = LoggerFactory.getLogger(SimpleFlowMergeAggregator.class.getName());
@Override
public Exchange aggregate(Exchange oldExchange, Exchange newExchange) {
LOGGER.info("Inside aggregator " + newExchange.getIn().getBody());
if(oldExchange == null) {
String data = newExchange.getIn().getBody(String.class);
List<String> aggregatedDataList = new ArrayList<>();
aggregatedDataList.add(data);
newExchange.getIn().setBody(aggregatedDataList);
return newExchange;
}
if(hadException(oldExchange)) {
if(!hadException(newExchange)) {
// aggregate and swap the base
LOGGER.info("Found new exchange with success. swapping the base exchange");
List<String> oldData = oldExchange.getIn().getBody(List.class);
oldData.add(newExchange.getIn().getBody(String.class));
// swapped the base here
newExchange.getIn().setBody(oldData);
return newExchange;
}
}
List<String> oldData = oldExchange.getIn().getBody(List.class);
oldData.add(newExchange.getIn().getBody(String.class));
oldExchange.getIn().setBody(oldData);
return oldExchange;
}
private boolean hadException(Exchange exchange) {
if(exchange.isFailed()) {
return true;
}
if(exchange.isRollbackOnly()) {
return true;
}
if(exchange.isRollbackOnlyLast()) {
return true;
}
if(((ExtendedExchange)exchange).isErrorHandlerHandledSet()
&& ((ExtendedExchange)exchange).isErrorHandlerHandled()) {
return true;
}
return false;
}
}
再次运行上述案例，日志信息如下：
2021-05-06 12:46:19.122 INFO 2576 --- [qtp174245837-45] route1 : received request
2021-05-06 12:46:19.123 INFO 2576 --- [qtp174245837-45] route1 : Entering multicast
2021-05-06 12:46:19.130 INFO 2576 --- [ #3 - Multicast] route2 : Executing PATH_1 - exception path
2021-05-06 12:46:19.130 INFO 2576 --- [ #3 - Multicast] route2 : Starting exception throw
2021-05-06 12:46:19.134 INFO 2576 --- [ #3 - Multicast] route2 : Exception handler invoked
2021-05-06 12:46:19.135 INFO 2576 --- [ #3 - Multicast] c.e.d.m.SimpleFlowMergeAggregator : Inside aggregator {"data" : "err"}
2021-05-06 12:46:20.130 INFO 2576 --- [ #4 - Multicast] route3 : Executing PATH_2 - success path
2021-05-06 12:46:22.132 INFO 2576 --- [ #4 - Multicast] route3 : PATH_2
2021-05-06 12:46:22.132 INFO 2576 --- [ #4 - Multicast] c.e.d.m.SimpleFlowMergeAggregator : Inside aggregator DATA_FROM_PATH_2
2021-05-06 12:46:22.132 INFO 2576 --- [ #4 - Multicast] c.e.d.m.SimpleFlowMergeAggregator : Found new exchange with success. swapping the base exchange
2021-05-06 12:46:22.133 INFO 2576 --- [ #4 - Multicast] route1 : Aggregated results {"data" : "err"},DATA_FROM_PATH_2
2021-05-06 12:46:22.133 INFO 2576 --- [ #4 - Multicast] route1 : Another log
可以看出，使用新的自定义聚合策略后，后续的日志（log）和转换（transform）步骤都成功执行。
6 结语
本文通过案例，发现了一个Camel Multicast组件聚合策略相关的问题。通过查看Camel源代码，找到了问题原因并给出了解决方案。
希望本文可以帮助到遇到同样问题的Camel用户。
- 点赞
- 收藏
- 关注作者
评论（0）