org.apache.hadoop.fs.s3a

Class S3ARetryPolicy

java.lang.Object

org.apache.hadoop.fs.s3a.S3ARetryPolicy

All Implemented Interfaces:

org.apache.hadoop.io.retry.RetryPolicy

Direct Known Subclasses:

S3GuardDataAccessRetryPolicy, S3GuardExistsRetryPolicy

public class S3ARetryPolicy
extends Object
implements org.apache.hadoop.io.retry.RetryPolicy

The S3A request retry policy. This uses the retry options in the configuration file to determine retry count and delays for "normal" retries and separately, for throttling; the latter is best handled for longer with an exponential back-off.

1. Those exceptions considered unrecoverable (networking) are failed fast.
2. All non-IOEs are failed immediately. Assumed: bugs in code, unrecoverable errors, etc

For non-idempotent operations, only failures due to throttling or from failures which are known to only arise prior to talking to S3 are retried. The retry policy is all built around that of the normal IO exceptions, particularly those extracted from S3AUtils.translateException(String, Path, AmazonClientException). Because the shouldRetry(Exception, int, int, boolean) method does this translation if an AmazonClientException is processed, the policy defined for the IOEs also applies to the original exceptions. Put differently: this retry policy aims to work for handlers of the untranslated exceptions, as well as the translated ones.

See Also:

S3 Error responses, Amazon S3 Error Best Practices, Dynamo DB Commmon errors

Nested Class Summary

Nested classes/interfaces inherited from interface org.apache.hadoop.io.retry.RetryPolicy

org.apache.hadoop.io.retry.RetryPolicy.RetryAction

Field Summary

Fields

Modifier and Type	Field and Description
protected org.apache.hadoop.io.retry.RetryPolicy	baseExponentialRetry Exponential policy for the base of normal failures.
protected org.apache.hadoop.io.retry.RetryPolicy	connectivityFailure Client connectivity: baseExponentialRetry without worrying about whether or not the command is idempotent.
protected org.apache.hadoop.io.retry.RetryPolicy	fail No retry on network and tangible API issues.
protected org.apache.hadoop.io.retry.RetryPolicy	retryIdempotentCalls Idempotent calls which raise IOEs are retried.
protected org.apache.hadoop.io.retry.RetryPolicy	throttlePolicy Policy for throttle requests, which are considered repeatable, even for non-idempotent calls, as the service rejected the call entirely.

Constructor Summary

Constructors

Constructor and Description
S3ARetryPolicy(org.apache.hadoop.conf.Configuration conf) Instantiate.

Method Summary

Modifier and Type	Method and Description
protected Map<Class<? extends Exception>,org.apache.hadoop.io.retry.RetryPolicy>	createExceptionMap() Subclasses can override this like a constructor to change behavior: call superclass method, then modify it as needed, and return it.
protected org.apache.hadoop.io.retry.RetryPolicy	createThrottleRetryPolicy(org.apache.hadoop.conf.Configuration conf) Create the throttling policy.
protected org.apache.hadoop.conf.Configuration	getConfiguration() Get the configuration this policy was created from.
org.apache.hadoop.io.retry.RetryPolicy.RetryAction	shouldRetry(Exception exception, int retries, int failovers, boolean idempotent)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

baseExponentialRetry

protected final org.apache.hadoop.io.retry.RetryPolicy baseExponentialRetry

Exponential policy for the base of normal failures.

retryIdempotentCalls

protected final org.apache.hadoop.io.retry.RetryPolicy retryIdempotentCalls

Idempotent calls which raise IOEs are retried.

throttlePolicy

protected final org.apache.hadoop.io.retry.RetryPolicy throttlePolicy

Policy for throttle requests, which are considered repeatable, even for non-idempotent calls, as the service rejected the call entirely.

fail

protected final org.apache.hadoop.io.retry.RetryPolicy fail

No retry on network and tangible API issues.

connectivityFailure

protected final org.apache.hadoop.io.retry.RetryPolicy connectivityFailure

Client connectivity: baseExponentialRetry without worrying about whether or not the command is idempotent.

Constructor Detail

S3ARetryPolicy

public S3ARetryPolicy(org.apache.hadoop.conf.Configuration conf)

Instantiate.

Parameters:

conf - configuration to read.

Method Detail

createThrottleRetryPolicy

protected org.apache.hadoop.io.retry.RetryPolicy createThrottleRetryPolicy(org.apache.hadoop.conf.Configuration conf)

Create the throttling policy. This will be called from the S3ARetryPolicy constructor, so subclasses must assume they are not initialized.

Parameters:

conf - configuration to use.

Returns:

the retry policy for throttling events.

createExceptionMap

protected Map<Class<? extends Exception>,org.apache.hadoop.io.retry.RetryPolicy> createExceptionMap()

Subclasses can override this like a constructor to change behavior: call superclass method, then modify it as needed, and return it.

Returns:

Map from exception type to RetryPolicy

shouldRetry

public org.apache.hadoop.io.retry.RetryPolicy.RetryAction shouldRetry(Exception exception,
                                                                      int retries,
                                                                      int failovers,
                                                                      boolean idempotent)
                                                               throws Exception

Specified by:

shouldRetry in interface org.apache.hadoop.io.retry.RetryPolicy

Throws:

Exception

getConfiguration

protected org.apache.hadoop.conf.Configuration getConfiguration()

Get the configuration this policy was created from.

Returns:

the configuration.