Details
-
Bug
-
Status: Open
-
Major
-
Resolution: Unresolved
-
3.2.1
-
None
Description
Summary
The instance metadata service has its own guidance for error handling and retry which are different from the Blob store. https://docs.microsoft.com/en-us/azure/active-directory/managed-identities-azure-resources/how-to-use-vm-token#error-handling
In particular, it responds with HTTP 429 if request rate is too high. Whereas Blob store will respond with HTTP 503. The retry policy used only accounts for the latter as it will retry any status >=500. This can result in job instability when running multiple processes on the same host.
Environment
- Spark talking to an ABFS store
- Hadoop 3.2.1
- Running on an Azure VM with user-assigned identity, ABFS configured to use MsiTokenProvider
- 6 executor processes on each VM
Example
Here's an example error message and stack trace. It's always the same stack trace. This appears in logs a few hundred to low thousands of times a day. It's luckily skating by since the download operation is wrapped in 3 retries.
AADToken: HTTP connection failed for getting token from AzureAD. Http response: 429 null Content-Type: application/json; charset=utf-8 Content-Length: 90 Request ID: Proxies: none First 1K of Body: {"error":"invalid_request","error_description":"Temporarily throttled, too many requests"} at org.apache.hadoop.fs.azurebfs.services.AbfsRestOperation.executeHttpOperation(AbfsRestOperation.java:190) at org.apache.hadoop.fs.azurebfs.services.AbfsRestOperation.execute(AbfsRestOperation.java:125) at org.apache.hadoop.fs.azurebfs.services.AbfsClient.getAclStatus(AbfsClient.java:506) at org.apache.hadoop.fs.azurebfs.services.AbfsClient.getAclStatus(AbfsClient.java:489) at org.apache.hadoop.fs.azurebfs.AzureBlobFileSystemStore.getIsNamespaceEnabled(AzureBlobFileSystemStore.java:208) at org.apache.hadoop.fs.azurebfs.AzureBlobFileSystemStore.getFileStatus(AzureBlobFileSystemStore.java:473) at org.apache.hadoop.fs.azurebfs.AzureBlobFileSystem.getFileStatus(AzureBlobFileSystem.java:437) at org.apache.hadoop.fs.FileSystem.isFile(FileSystem.java:1717) at org.apache.spark.util.Utils$.fetchHcfsFile(Utils.scala:747) at org.apache.spark.util.Utils$.doFetchFile(Utils.scala:724) at org.apache.spark.util.Utils$.fetchFile(Utils.scala:496) at org.apache.spark.executor.Executor.$anonfun$updateDependencies$7(Executor.scala:812) at org.apache.spark.executor.Executor.$anonfun$updateDependencies$7$adapted(Executor.scala:803) at scala.collection.TraversableLike$WithFilter.$anonfun$foreach$1(TraversableLike.scala:792) at scala.collection.mutable.HashMap.$anonfun$foreach$1(HashMap.scala:149) at scala.collection.mutable.HashTable.foreachEntry(HashTable.scala:237) at scala.collection.mutable.HashTable.foreachEntry$(HashTable.scala:230) at scala.collection.mutable.HashMap.foreachEntry(HashMap.scala:44) at scala.collection.mutable.HashMap.foreach(HashMap.scala:149) at scala.collection.TraversableLike$WithFilter.foreach(TraversableLike.scala:791) at org.apache.spark.executor.Executor.org$apache$spark$executor$Executor$$updateDependencies(Executor.scala:803) at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:375) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624) at java.lang.Thread.run(Thread.java:748)
Attachments
Issue Links
- is blocked by
-
HADOOP-18860 Upgrade mockito to 4.11.0
-
- Open
-
- relates to
-
HADOOP-16857 ABFS: Optimize HttpRequest retry triggers
-
- Resolved
-
-
HADOOP-17092 ABFS: Long waits and unintended retries when multiple threads try to fetch token using ClientCreds
-
- Resolved
-
-
HIVE-27884 LLAP: Reuse FileSystem objects from cache across different tasks in the same LLAP daemon
-
- In Progress
-
- links to