ZookeeperDynamicConfiguration will be busy wait if zk is unavailable.

[beiwei30]

    ZookeeperDynamicConfiguration(URL url, ZookeeperTransporter zookeeperTransporter) {
        this.url = url;
        rootPath = "/" + url.getParameter(CONFIG_NAMESPACE_KEY, DEFAULT_GROUP) + "/config";

        initializedLatch = new CountDownLatch(1);
        this.cacheListener = new CacheListener(rootPath, initializedLatch);
        this.executor = Executors.newFixedThreadPool(1, new NamedThreadFactory(this.getClass().getSimpleName(), true));

        zkClient = zookeeperTransporter.connect(url);
        zkClient.addDataListener(rootPath, cacheListener, executor);
        try {
            // Wait for connection
            this.initializedLatch.await();
        } catch (InterruptedException e) {
            logger.warn("Failed to build local cache for config center (zookeeper)." + url);
        }
    }

Instead, should we introduce a timeout for this.initializedLatch.await()?

[Leishunyu]

I think it is not particularly necessary, it is not available at the time of initialization. This problem should be discovered in time. If the configuration center is used, the connection is necessary.

[beiwei30]

there are two issues in this implementation:

1. When zookeeper address is configured wrong, the latch will wait forever. This problem is also mentioned in ZookeeperDynamicConfiguration 严重 BUG #4125
2. Even when zookeeper address is configured correctly, there's chance that INITIALIZED event is fired before the listener is registered, which will cause latch waits forever as well.

For issue 2, it is confirmed there's tiny time window which may cause the dead wait, after I patched both dubbo and curator:

[16/07/19 03:04:56:056 CST] main-EventThread DEBUG cache.TreeCache: publishEvent: TreeCacheEvent{type=INITIALIZED, data=null}
[16/07/19 03:04:56:056 CST] Curator-TreeCache-0  INFO cache.TreeCache: >>> listeners count: 0
[16/07/19 03:04:56:056 CST] Curator-TreeCache-0  INFO cache.TreeCache: >>> event: TreeCacheEvent{type=INITIALIZED, data=null}
[16/07/19 03:04:56:056 CST] main  INFO zookeeper.ZookeeperDynamicConfiguration: >>> start to wait the latch