[HADOOP-17377] ABFS: MsiTokenProvider doesn't retry HTTP 429 from the Instance Metadata Service

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 3.2.1
• Component/s: fs/azure
• Labels: pull-request-available
• Assignee: Unassigned
• Reporter: Brandon
• Created: 13/Nov/20 03:16
• Updated: 07/Dec/23 17:33

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)

 CC mackrorysd, [e-mail]

Issue Links

• is blocked by: HADOOP-18860 Upgrade mockito to 4.11.0
• relates to: HADOOP-16857 ABFS: Optimize HttpRequest retry triggers
• relates to: HADOOP-17092 ABFS: Long waits and unintended retries when multiple threads try to fetch token using ClientCreds
• relates to: HIVE-27884 LLAP: Reuse FileSystem objects from cache across different tasks in the same LLAP daemon
• links to: GitHub Pull Request #5273