Sentinel入门（一） Sentinel在Dubbo的实现

Demo

Consumer.java

public static void main(String[] args) {

		FlowRule flowRule = new FlowRule();
		flowRule.setResource(
				"com.alibaba.cloud.examples.FooService:hello(java.lang.String)");
		// 设置1s内只能调用10次
        flowRule.setCount(10);
		flowRule.setGrade(RuleConstant.FLOW_GRADE_QPS);
		flowRule.setLimitApp("default");
        // 注册限流规则
		FlowRuleManager.loadRules(Collections.singletonList(flowRule));

		SpringApplicationBuilder consumerBuilder = new SpringApplicationBuilder();
		ApplicationContext applicationContext = consumerBuilder
				.web(WebApplicationType.NONE).sources(SentinelDubboConsumerApp.class)
				.run(args);

		FooServiceConsumer service = applicationContext.getBean(FooServiceConsumer.class);
        // 测试调用15次，期望应该是有5调用失败的
		for (int i = 0; i < 15; i++) {
			try {
				String message = service.hello("Jim");
				System.out.println((i + 1) + " -> Success: " + message);
			}
			catch (SentinelRpcException ex) {
				System.out.println("Blocked");
			}
			catch (Exception ex) {
				ex.printStackTrace();
			}
		}
	}

FooService服务在Sentinel下对应的资源名是 com.alibaba.cloud.examples.FooService:hello(java.lang.String) 。详细见下面的Filter的getResourceName， 定义该资源名对应的限流规则：

 public static String getResourceName(Invoker<?> invoker, Invocation invocation) {
        StringBuilder buf = new StringBuilder(64);
        buf.append(invoker.getInterface().getName())
            .append(":")
            .append(invocation.getMethodName())
            .append("(");
        boolean isFirst = true;
        for (Class<?> clazz : invocation.getParameterTypes()) {
            if (!isFirst) {
                buf.append(",");
            }
            buf.append(clazz.getName());
            isFirst = false;
        }
        buf.append(")");
        return buf.toString();
    }

由于设置的qps是10。调用15次查看是否被限流

1 -> Success: hello, Jim
2 -> Success: hello, Jim
3 -> Success: hello, Jim
4 -> Success: hello, Jim
5 -> Success: hello, Jim
6 -> Success: hello, Jim
7 -> Success: hello, Jim
8 -> Success: hello, Jim
9 -> Success: hello, Jim
10 -> Success: hello, Jim
Blocked
Blocked
Blocked
Blocked
Blocked

可以看到是正常被限流的，那么看看是如何限流的。

Dubbo Consumer Sentinal Filter

@Activate(group = "consumer")
public class SentinelDubboConsumerFilter extends AbstractDubboFilter implements Filter {

    public SentinelDubboConsumerFilter() {
        RecordLog.info("Sentinel Dubbo consumer filter initialized");
    }

    @Override
    public Result invoke(Invoker<?> invoker, Invocation invocation) throws RpcException {
        Entry interfaceEntry = null;
        Entry methodEntry = null;
        try {
            String resourceName = getResourceName(invoker, invocation);
            interfaceEntry = SphU.entry(invoker.getInterface().getName(), ResourceTypeConstants.COMMON_RPC,
                EntryType.OUT);
            methodEntry = SphU.entry(resourceName, ResourceTypeConstants.COMMON_RPC, EntryType.OUT);

            Result result = invoker.invoke(invocation);
            if (result.hasException()) {
                Throwable e = result.getException();
                // Record common exception.
                Tracer.traceEntry(e, interfaceEntry);
                Tracer.traceEntry(e, methodEntry);
            }
            return result;
        } catch (BlockException e) {
            return DubboFallbackRegistry.getConsumerFallback().handle(invoker, invocation, e);
        } catch (RpcException e) {
            Tracer.traceEntry(e, interfaceEntry);
            Tracer.traceEntry(e, methodEntry);
            throw e;
        } finally {
            if (methodEntry != null) {
                methodEntry.exit();
            }
            if (interfaceEntry != null) {
                interfaceEntry.exit();
            }
        }
    }
}

重点是：

1. String resourceName = getResourceName(invoker, invocation); resource的生成
2. 接口级别的流控： interfaceEntry = SphU.entry(invoker.getInterface().getName(), ResourceTypeConstants.COMMON_RPC,
3. 方法级别的流控： methodEntry = SphU.entry(resourceName, ResourceTypeConstants.COMMON_RPC, EntryType.OUT);
4. EntryType.OUT代表的是出网流量，即对外调用次数，EntryType.IN代表的是入网流量，即被调用次数

Dubbo Provider Sentinal

public class SentinelDubboProviderFilter implements Filter {

    public SentinelDubboProviderFilter() {
        RecordLog.info("Sentinel Apache Dubbo provider filter initialized");
    }

    @Override
    public Result invoke(Invoker<?> invoker, Invocation invocation) throws RpcException {
        // Get origin caller.
        String application = DubboUtils.getApplication(invocation, "");

        Entry interfaceEntry = null;
        Entry methodEntry = null;
        try {
            String resourceName = DubboUtils.getResourceName(invoker, invocation);
            String interfaceName = invoker.getInterface().getName();
            // Only need to create entrance context at provider side, as context will take effect
            // at entrance of invocation chain only (for inbound traffic).
            ContextUtil.enter(resourceName, application);
            interfaceEntry = SphU.entry(interfaceName, ResourceTypeConstants.COMMON_RPC, EntryType.IN);
            methodEntry = SphU.entry(resourceName, ResourceTypeConstants.COMMON_RPC,
                EntryType.IN, invocation.getArguments());

            Result result = invoker.invoke(invocation);
            if (result.hasException()) {
                Throwable e = result.getException();
                // Record common exception.
                Tracer.traceEntry(e, interfaceEntry);
                Tracer.traceEntry(e, methodEntry);
            }
            return result;
        } catch (BlockException e) {
            return DubboFallbackRegistry.getProviderFallback().handle(invoker, invocation, e);
        } catch (RpcException e) {
            Tracer.traceEntry(e, interfaceEntry);
            Tracer.traceEntry(e, methodEntry);
            throw e;
        } finally {
            if (methodEntry != null) {
                methodEntry.exit(1, invocation.getArguments());
            }
            if (interfaceEntry != null) {
                interfaceEntry.exit();
            }
            ContextUtil.exit();
        }
    }
}

与消费者主要有以下的不同

1. ContextUtil.enter(resourceName, application); 因为上下文只会在调用链的入口才生效（对于入站流量）。
2. 对资源的保护方向不一样。

Sentinel实现原理

可以联想到我们之前定义的那个规则：

QPS：10

那么可以之间看到FlowSlot

public class FlowSlot extends AbstractLinkedProcessorSlot<DefaultNode> {

    private final FlowRuleChecker checker;

    public FlowSlot() {
        this(new FlowRuleChecker());
    }

    /**
     * Package-private for test.
     *
     * @param checker flow rule checker
     * @since 1.6.1
     */
    FlowSlot(FlowRuleChecker checker) {
        AssertUtil.notNull(checker, "flow checker should not be null");
        this.checker = checker;
    }

    @Override
    public void entry(Context context, ResourceWrapper resourceWrapper, DefaultNode node, int count,
                      boolean prioritized, Object... args) throws Throwable {
        checkFlow(resourceWrapper, context, node, count, prioritized);

        fireEntry(context, resourceWrapper, node, count, prioritized, args);
    }

    void checkFlow(ResourceWrapper resource, Context context, DefaultNode node, int count, boolean prioritized)
        throws BlockException {
        // 真正的流控逻辑
        checker.checkFlow(ruleProvider, resource, context, node, count, prioritized);
    }

    @Override
    public void exit(Context context, ResourceWrapper resourceWrapper, int count, Object... args) {
        fireExit(context, resourceWrapper, count, args);
    }

    private final Function<String, Collection<FlowRule>> ruleProvider = new Function<String, Collection<FlowRule>>() {
        @Override
        public Collection<FlowRule> apply(String resource) {
            // 捞取我们刚才定义的FlowRule
            Map<String, List<FlowRule>> flowRules = FlowRuleManager.getFlowRuleMap();
            return flowRules.get(resource);
        }
    };
}

---

1. 在多线程情况下，他的访问次数应该是一个竟态资源，那么它是怎么处理的呢？

以上两个问题可能需要读代码来解决了。。。。

未完待续。。。。