Google Storage API中的死锁

Question

我在Dataproc上运行一个spark作业，它从一个桶中读取大量文件并将它们合并到一个大文件中。我通过阴影使用google-api-services-storage 1.29.0。到目前为止它工作正常，整合~20-30K文件。今天我用大约5倍的文件尝试了它，突然间我陷入了僵局（在东方，我想我是，因为似乎所有的执行者都在无限期地等待）。

这是线程转储：

org.conscrypt.NativeCrypto.SSL_read(Native Method)
org.conscrypt.NativeSsl.read(NativeSsl.java:416)
org.conscrypt.ConscryptFileDescriptorSocket$SSLInputStream.read(ConscryptFileDescriptorSocket.java:547) => holding Monitor(java.lang.Object@1638155334})
java.io.BufferedInputStream.fill(BufferedInputStream.java:246)
java.io.BufferedInputStream.read1(BufferedInputStream.java:286)
java.io.BufferedInputStream.read(BufferedInputStream.java:345) => holding Monitor(java.io.BufferedInputStream@1513035694})
sun.net.www.http.HttpClient.parseHTTPHeader(HttpClient.java:735)
sun.net.www.http.HttpClient.parseHTTP(HttpClient.java:678)
sun.net.www.protocol.http.HttpURLConnection.getInputStream0(HttpURLConnection.java:1587) => holding Monitor(sun.net.www.protocol.https.DelegateHttpsURLConnection@995846771})
sun.net.www.protocol.http.HttpURLConnection.getInputStream(HttpURLConnection.java:1492) => holding Monitor(sun.net.www.protocol.https.DelegateHttpsURLConnection@995846771})
java.net.HttpURLConnection.getResponseCode(HttpURLConnection.java:480)
sun.net.www.protocol.https.HttpsURLConnectionImpl.getResponseCode(HttpsURLConnectionImpl.java:347)
com.shaded.google.api.client.http.javanet.NetHttpResponse.<init>(NetHttpResponse.java:37)
com.shaded.google.api.client.http.javanet.NetHttpRequest.execute(NetHttpRequest.java:105)
com.shaded.google.api.client.http.HttpRequest.execute(HttpRequest.java:981)
com.shaded.google.api.client.googleapis.services.AbstractGoogleClientRequest.executeUnparsed(AbstractGoogleClientRequest.java:419)
com.shaded.google.api.client.googleapis.services.AbstractGoogleClientRequest.executeUnparsed(AbstractGoogleClientRequest.java:352)
com.shaded.google.api.client.googleapis.services.AbstractGoogleClientRequest.executeMedia(AbstractGoogleClientRequest.java:380)
com.shaded.google.api.services.storage.Storage$Objects$Get.executeMedia(Storage.java:6189)
com.shaded.google.cloud.storage.spi.v1.HttpStorageRpc.load(HttpStorageRpc.java:584)
com.shaded.google.cloud.storage.StorageImpl$16.call(StorageImpl.java:464)
com.shaded.google.cloud.storage.StorageImpl$16.call(StorageImpl.java:461)
com.shaded.google.api.gax.retrying.DirectRetryingExecutor.submit(DirectRetryingExecutor.java:89)
com.shaded.google.cloud.RetryHelper.run(RetryHelper.java:74)
com.shaded.google.cloud.RetryHelper.runWithRetries(RetryHelper.java:51)
com.shaded.google.cloud.storage.StorageImpl.readAllBytes(StorageImpl.java:461)
com.shaded.google.cloud.storage.Blob.getContent(Blob.java:455)
my.package.with.my.StorageAPI.readFetchedLocation(StorageAPI.java:71)
...

最终我必须杀死这份工作，因为没有任何反应。知道是什么原因引起的吗？我尝试在我的代码中使用ThreadLocal<Storage>和单个Storage实例，它似乎没有什么区别。

Answer 1

这个工作实际上并没有陷入僵局，只是Spark UI由于某种原因在阶段完成之前没有显示任务的任何进展。我以为什么都没发生，但如果我反复进行线程转储，那么我可以看到它正在做的事情。

正如tix在评论中建议的那样，在直接使用存储库时实现exponential backoff可能是明智的，如果我得到StorageException的isRetryable()则重试。