[
https://issues.apache.org/jira/browse/HADOOP-17377?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Brandon updated HADOOP-17377:
-----------------------------
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. 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.
{noformat}
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){noformat}
CC [~mackrorysd], [[email protected]]
was:
*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. 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 my logs a few hundred to low thousands of times a day.
{noformat}
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){noformat}
CC [~mackrorysd], [[email protected]]
> ABFS: MsiTokenProvider doesn't retry HTTP 429 from the Instance Metadata
> Service
> --------------------------------------------------------------------------------
>
> Key: HADOOP-17377
> URL: https://issues.apache.org/jira/browse/HADOOP-17377
> Project: Hadoop Common
> Issue Type: Bug
> Components: fs/azure
> Affects Versions: 3.2.1
> Reporter: Brandon
> Priority: Major
>
> *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. 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.
> {noformat}
> 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){noformat}
> CC [~mackrorysd], [[email protected]]
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
