Java Bytecode Assembly, Interesting concepts in the art of Computer Science/Software Development, maybe from time to time other pesky corp. related development issues.

Wednesday, January 5, 2011

Recursive delete utility for version enabled S3 bucket

Deleting S3 bucket that contains anything more then the most trivial collection of keys (>1000) is a never ending saga. Deleting S3 bucket that has versioning enabled is with today's tools simply not possible.

We had 2 buckets that were taking up storage space: the first had ~300,000,000 keys, the second ~2500 but with versioning enabled and keys that contain "\n" (new line), "//" (double slash) and " " (space) in their names. Trying to use tools such as jets3, boto, bucket explorer, aws s3 console, s3fox on either of these buckets failed in the most pathetic/exotic of ways.

I quickly understood that this requires a custom implementation.

Doing some googling I found this excellent post proving some pseudo-code for multi threaded delete operation against S3. Good, the correct post in the correct time! Based on Mark's code I've implemented BucketDestroy which has the sole purpose of doing everything it takes to delete an S3 bucket, a seek & destroy S3 bucket commando if you like.

I've enhanced Mark code with the ability to delete not just bucket keys but bucket key versions, this is required because Amazon will not let you delete S3 bucket that has old versions of a Key.

The code is provided below, I've also hosted it on google code where you can find maven based setup instructions and the latest version.


import java.util.List;
import java.util.concurrent.CountDownLatch;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.atomic.AtomicInteger;

import com.amazonaws.auth.BasicAWSCredentials;

 * Efficient delete of version enabled bucket.
 * This code will delete all keys from gives bucket and all previous versions of given key.
 * This is done in an efficient multi-threaded ExecutorService environment
 * Launching:
 * $java -Daws.key=AK7895IH1234X2GW12IQ -Daws.secret=1234567890123456789012345678901234456789 BucketDestroy "bucketName"

 * @author Maxim Veksler <>
 * @author Mark S. Kolich
 * @version 0.1 - First pseudo code version, Written by Mark S. Kolich. 
 * @version 0.2 - Maxim Veksler - Fix paging. Implement deletion of versioned keys.
 * For the latest version see
public class BucketDestroy {
    private static final String AWS_ACCESS_KEY_PROPERTY = "aws.key";
    private static final String AWS_SECRET_PROPERTY = "aws.secret";
    private static AmazonS3Client s3;
    public static void main(String[] args) {

        final String accessKey = System.getProperty(AWS_ACCESS_KEY_PROPERTY);
        final String secret = System.getProperty(AWS_SECRET_PROPERTY);
        if (accessKey == null || secret == null) {
            System.err.println("You're missing the -Daws.key and -Daws.secret required VM properties.");

        if (args.length < 1) {
            System.err.println("Missing required program argument: bucketName.");
        // S3 Initialization
        s3 = new AmazonS3Client(new BasicAWSCredentials(accessKey, secret));

        // Instantiate BucketDestroy and call destroyBucket() - Bucket deleting beast!
        new BucketDestroy().destroyBucket(args[0]);

     * Do everything it takes to delete amazon S3 bucket. 
     * Currently it means delete all versions of bucket keys.
     * TODO: 
     *  - Support distributed operation by doing shuffle on first level of bucketName/../
     *  - Seek a method to obtain MFA to delete MFA enabled bucket
     *  - Implement starvation prevention for deleting threads (do not wait until next s3.listVersions finishes)
    public void destroyBucket(final String bucketName) {
        // Set up a new thread pool to delete 20 objects at a time.
        ExecutorService _pool = Executors.newFixedThreadPool(20);

        // Get counter, just for status
        final AtomicInteger ATOMIC_INTEGER = new AtomicInteger();
        // List all key in the bucket
        VersionListing versionListing = s3.listVersions(bucketName, "");
        List<S3VersionSummary> versionSummaries = versionListing.getVersionSummaries();
        while(versionSummaries != null && versionSummaries.size() > 0) {
            final CountDownLatch latch = new CountDownLatch(versionSummaries.size());
            for(final S3VersionSummary objectSummary : versionSummaries) {
                _pool.execute(new Runnable() {
                    public void run() {
                        String keyName = null;
                        String versionId = null;
                        try {
                            keyName = objectSummary.getKey();
                            versionId = objectSummary.getVersionId();
                            String info = ">>>>   INFO: " + ATOMIC_INTEGER.incrementAndGet() + " deleting: (" + bucketName + "/" + keyName + "@" + versionId + ")";
                            s3.deleteVersion(new DeleteVersionRequest(bucketName, keyName, versionId));
                        } catch (Exception e) {
                            String err = ">>>> FAILED delete: (" + bucketName + "/" + keyName + "@" + versionId + ")";
                        } finally {
            // After sending current batch of delete tasks we block until Latch reaches zero, this 
            // allows to not over populate ExecutorService tasks queue. 
            try {
            } catch (Exception exception) {

            // Paging over all S3 keys...
            versionListing = s3.listNextBatchOfVersions(versionListing);
            versionSummaries = versionListing.getVersionSummaries();

        try {
            // This code fails with HTTP 400 from aws API 
//            SetBucketVersioningConfigurationRequest setBucketVersioningConfigurationRequest = new SetBucketVersioningConfigurationRequest(bucketName, new BucketVersioningConfiguration().withStatus(BucketVersioningConfiguration.OFF));
//            s3.setBucketVersioningConfiguration(setBucketVersioningConfigurationRequest);
            // So we do this:
            System.out.println(">>>>   INFO: bucket " + bucketName + " deleted!");
        } catch (Exception e) { 
            System.err.println("Failed to ultimately delete bucket: " + bucketName);



  1. This was a life saver, thanks!

    One note: you are depending on Apache commons logging, which might not be in the user's class path, which then produces an ugly error (java.lang.NoClassDefFoundError: org/apache/commons/logging/LogFactory).

    The fix, if anyone experiences it, is to just download from, unzip, add to classpath (or just put it in your lib/, you will need it later).

    Thanks again!

    1. Hi Filip,

      Thanks for the comment. I've written an easier integration method based on Maven here



About Me

My photo
Tel Aviv, Israel
I work in one of Israel's Hi-Tech company's. I do 5 nines system development with some system administration. This blog will focus on the technical side of things, with some philosophical strings attached.