Skip to content

[Bug] some clients might always encounter the RpcException with 'No provider available from registry' if restarting all clients and servers which register-mode are interface #15107

@zrlw

Description

@zrlw

Pre-check

  • I am sure that all the content I provide is in English.

Search before asking

  • I had searched in the issues and found no similar issues.

Apache Dubbo Component

Java SDK (apache/dubbo)

Dubbo Version

Dubbo 3.2.12 + nacos-client 2.3.2, registration mode of the dubbo servers are interface,Jdk17

Steps to reproduce this issue

  1. start dubbo servers which register-mode are interface only.
  2. login nacos registation center, switch dubbo server interface status to offline.
  3. set dubbo client app breakpoints.
    (1) at org.apache.dubbo.registry.integration.RegistryDirectory#refreshInvoker
    private void refreshInvoker(List<URL> invokerUrls) {
        Assert.notNull(invokerUrls, "invokerUrls should not be null");

        if (invokerUrls.size() == 1
                && invokerUrls.get(0) != null
                && EMPTY_PROTOCOL.equals(invokerUrls.get(0).getProtocol())) {
            //  set first breakpoint at next refreshRouter(...) statement. <<<============================= SEE HERE !!!
            refreshRouter(
                    BitList.emptyList(), () -> this.forbidden = true // Forbid to access
                    );
            destroyAllInvokers(); // Close all invokers
    
    (2) at org.apache.dubbo.registry.nacos.NacosRegistry#doSubscribe
    private void doSubscribe(final URL url, final NacosAggregateListener listener, final Set<String> serviceNames) {
        try {
            if (isServiceNamesWithCompatibleMode(url)) {

                /**
                 * Get all instances with serviceNames to avoid instance overwrite and but with empty instance mentioned
                 * in https://github.com/apache/dubbo/issues/5885 and https://github.com/apache/dubbo/issues/5899
                 *
                 * namingService.getAllInstances with
                 * {@link org.apache.dubbo.registry.support.AbstractRegistry#registryUrl}
                 * default {@link DEFAULT_GROUP}
                 *
                 * in https://github.com/apache/dubbo/issues/5978
                 */
                for (String serviceName : serviceNames) {
                    List<Instance> instances =
                            namingService.getAllInstances(serviceName, geturl("").getGroup(Constants.DEFAULT_GROUP));
                    notifySubscriber(url, serviceName, listener, instances);
                }
                //  set second breakpoint at next for(...) statement. <<<============================= SEE HERE !!!
                for (String serviceName : serviceNames) {
                    subscribeEventListener(serviceName, url, listener);
                }
  1. start dubbo client app under debug mode, and it will run to first breakpoint which will set the forbidden flag to true.
  2. continue dubbo client app running to second breakpoint which will subscribe event listener.
  3. switch all dubbo server interface status to online at nacos registation center, and wait about a few seconds (let's say 10 seconds) to ensure nacos registration center finish notifying providers changing.
  4. continue dubbo client app running.
  5. as forbidden value of RegistryDirectory is always true, the dubbo client app will encouter the RpcException with 'No provider available from registry' whenever calling the dubbo server interface.
    at org.apache.dubbo.registry.integration.DynamicDirectory#doList
@Override
    public List<Invoker<T>> doList(
            SingleRouterChain<T> singleRouterChain, BitList<Invoker<T>> invokers, Invocation invocation) {
        //  >>> the value of forbidden is always true. <<<
        if (forbidden && shouldFailFast) {
            // 1. No service provider 2. Service providers are disabled
            throw new RpcException(
                    RpcException.FORBIDDEN_EXCEPTION,
                    "No provider available from registry " + this
                            + " for service " + getConsumerurl("").getServiceKey() + " on consumer "
                            + NetUtils.getLocalHost()
                            + " use dubbo version " + Version.getVersion()
                            + ", please check status of providers(disabled, not registered or in blocklist).");
        }

What you expected to happen

moving subscribeEventListener calling statement of NacosRegistry#doSubscribe prior to notifySubscriber might solve this issue. just like ZookeeperRegistry#doSubscribe did.

Anything else

this issue has existed in dubbo2 and dubbo3 for a long time and many contributers proposed revision suggestions. I wonder why not made any revisions to this issue.

Are you willing to submit a pull request to fix on your own?

  • Yes I am willing to submit a pull request on my own!

Code of Conduct

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    Status

    Done

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions