S3Guard: Consistency and Metadata Caching for S3A

S3A to warn or fail if S3Guard is disabled

A seemingly recurrent problem with S3Guard is that people think S3Guard is turned on but it isn’t. You can set org.apache.hadoop.fs.s3a.s3guard.disabled.warn.level to avoid this. The property sets what to do when an S3A FS is instantiated without S3Guard. The following values are available:

• SILENT: Do nothing.
• INFORM: Log at info level that FS is instantiated without S3Guard.
• WARN: Warn that data may be at risk in workflows.
• FAIL: S3AFileSystem instantiation will fail.

The default setting is SILENT. The setting is case insensitive. The required level can be set in the core-site.xml.

The latest configuration parameters are defined in core-default.xml. You should consult that file for full information, but a summary is provided here.

1. Choose the Database

A core concept of S3Guard is that the directory listing data of the object store, the metadata is replicated in a higher-performance, consistent, database. In S3Guard, this database is called The Metadata Store

By default, S3Guard is not enabled.

The Metadata Store to use in production is bonded to Amazon’s DynamoDB database service. The following setting will enable this Metadata Store:

<property>
    <name>fs.s3a.metadatastore.impl</name>
    <value>org.apache.hadoop.fs.s3a.s3guard.DynamoDBMetadataStore</value>
</property>

Note that the NullMetadataStore store can be explicitly requested if desired. This offers no metadata storage, and effectively disables S3Guard.

<property>
    <name>fs.s3a.metadatastore.impl</name>
    <value>org.apache.hadoop.fs.s3a.s3guard.NullMetadataStore</value>
</property>

2. Configure S3Guard Settings

More settings will may be added in the future. Currently the only Metadata Store-independent setting, besides the implementation class above, are the allow authoritative and fail-on-error flags.

<property>
    <name>fs.s3a.metadatastore.authoritative</name>
    <value>false</value>
</property>

<property>
    <name>fs.s3a.metadatastore.metadata.ttl</name>
    <value>15m</value>
</property>

hadoop s3guard import [-meta URI] s3a://my-bucket/file-with-bad-metadata

<property>
    <name>fs.s3a.metadatastore.fail.on.write.error</name>
    <value>false</value>
</property>

3. Configure the Metadata Store.

Here are the DynamoDBMetadataStore settings. Other Metadata Store implementations will have their own configuration parameters.

4. Name Your Table

First, choose the name of the table you wish to use for the S3Guard metadata storage in your DynamoDB instance. If you leave it unset/empty, a separate table will be created for each S3 bucket you access, and that bucket’s name will be used for the name of the DynamoDB table. For example, this sets the table name to my-ddb-table-name

<property>
  <name>fs.s3a.s3guard.ddb.table</name>
  <value>my-ddb-table-name</value>
  <description>
    The DynamoDB table name to operate. Without this property, the respective
    S3 bucket names will be used.
  </description>
</property>

It is good to share a table across multiple buckets for multiple reasons, especially if you are not using on-demand DynamoDB tables, and instead prepaying for provisioned I/O capacity.

1. You are billed for the provisioned I/O capacity allocated to the table, even when the table is not used. Sharing capacity can reduce costs.

2. You can share the “provision burden” across the buckets. That is, rather than allocating for the peak load on a single bucket, you can allocate for the peak load across all the buckets, which is likely to be significantly lower.

3. It’s easier to measure and tune the load requirements and cost of S3Guard, because there is only one table to review and configure in the AWS management console.

4. When you don’t grant the permission to create DynamoDB tables to users. A single pre-created table for all buckets avoids the needs for an administrator to create one for every bucket.

When wouldn’t you want to share a table?

1. When you are using on-demand DynamoDB and want to keep each table isolated.
2. When you do explicitly want to provision I/O capacity to a specific bucket and table, isolated from others.

3. When you are using separate billing for specific buckets allocated to specific projects.

4. When different users/roles have different access rights to different buckets. As S3Guard requires all users to have R/W access to the table, all users will be able to list the metadata in all buckets, even those to which they lack read access.

5. Locate your Table

You may also wish to specify the region to use for DynamoDB. If a region is not configured, S3A will assume that it is in the same region as the S3 bucket. A list of regions for the DynamoDB service can be found in Amazon’s documentation. In this example, to use the US West 2 region:

<property>
  <name>fs.s3a.s3guard.ddb.region</name>
  <value>us-west-2</value>
</property>

When working with S3Guard-managed buckets from EC2 VMs running in AWS infrastructure, using a local DynamoDB region ensures the lowest latency and highest reliability, as well as avoiding all long-haul network charges. The S3Guard tables, and indeed, the S3 buckets, should all be in the same region as the VMs.

6. Optional: Create your Table

Next, you can choose whether or not the table will be automatically created (if it doesn’t already exist). If you want this feature, set the fs.s3a.s3guard.ddb.table.create option to true.

<property>
  <name>fs.s3a.s3guard.ddb.table.create</name>
  <value>true</value>
  <description>
    If true, the S3A client will create the table if it does not already exist.
  </description>
</property>

7. If creating a table: Choose your billing mode (and perhaps I/O Capacity)

Next, you need to decide whether to use On-Demand DynamoDB and its pay-per-request billing (recommended), or to explicitly request a provisioned IO capacity.

Before AWS offered pay-per-request billing, the sole billing mechanism, was “provisioned capacity”. This mechanism requires you to choose the DynamoDB read and write throughput requirements you expect to need for your expected uses of the S3Guard table. Setting higher values cost you more money -even when the table was idle Note that these settings only affect table creation when fs.s3a.s3guard.ddb.table.create is enabled. To change the throughput for an existing table, use the AWS console or CLI tool.

For more details on DynamoDB capacity units, see the AWS page on Capacity Unit Calculations.

Provisioned IO capacity is billed per hour for the life of the table, even when the table and the underlying S3 buckets are not being used.

There are also charges incurred for data storage and for data I/O outside of the region of the DynamoDB instance. S3Guard only stores metadata in DynamoDB: path names and summary details of objects —the actual data is stored in S3, so billed at S3 rates.

With provisioned I/O capacity, attempting to perform more I/O than the capacity requested throttles the operation and may result in operations failing. Larger I/O capacities cost more.

With the introduction of On-Demand DynamoDB, you can now avoid paying for provisioned capacity by creating an on-demand table. With an on-demand table you are not throttled if your DynamoDB requests exceed any pre-provisioned limit, nor do you pay per hour even when a table is idle.

You do, however, pay more per DynamoDB operation. Even so, the ability to cope with sudden bursts of read or write requests, combined with the elimination of charges for idle tables, suit the use patterns made of S3Guard tables by applications interacting with S3. That is: periods when the table is rarely used, with intermittent high-load operations when directory trees are scanned (query planning and similar), or updated (rename and delete operations).

We recommending using On-Demand DynamoDB for maximum performance in operations such as query planning, and lowest cost when S3 buckets are not being accessed.

This is the default, as configured in the default configuration options.

<property>
  <name>fs.s3a.s3guard.ddb.table.capacity.read</name>
  <value>0</value>
  <description>
    Provisioned throughput requirements for read operations in terms of capacity
    units for the DynamoDB table. This config value will only be used when
    creating a new DynamoDB table.
    If set to 0 (the default), new tables are created with "per-request" capacity.
    If a positive integer is provided for this and the write capacity, then
    a table with "provisioned capacity" will be created.
    You can change the capacity of an existing provisioned-capacity table
    through the "s3guard set-capacity" command.
  </description>
</property>

<property>
  <name>fs.s3a.s3guard.ddb.table.capacity.write</name>
  <value>0</value>
  <description>
    Provisioned throughput requirements for write operations in terms of
    capacity units for the DynamoDB table.
    If set to 0 (the default), new tables are created with "per-request" capacity.
    Refer to related configuration option fs.s3a.s3guard.ddb.table.capacity.read
  </description>
</property>

8. If creating a table: Enable server side encryption (SSE)

Encryption at rest can help you protect sensitive data in your DynamoDB table. When creating a new table, you can set server side encryption on the table using the default AWS owned customer master key (CMK), AWS managed CMK, or customer managed CMK. S3Guard code accessing the table is all the same whether SSE is enabled or not. For more details on DynamoDB table server side encryption, see the AWS page on Encryption at Rest: How It Works.

These are the default configuration options, as configured in core-default.xml.

<property>
  <name>fs.s3a.s3guard.ddb.table.sse.enabled</name>
  <value>false</value>
  <description>
    Whether server-side encryption (SSE) is enabled or disabled on the table.
    By default it's disabled, meaning SSE is set to AWS owned CMK.
  </description>
</property>

<property>
  <name>fs.s3a.s3guard.ddb.table.sse.cmk</name>
  <value/>
  <description>
    The KMS Customer Master Key (CMK) used for the KMS encryption on the table.
    To specify a CMK, this config value can be its key ID, Amazon Resource Name
    (ARN), alias name, or alias ARN. Users only need to provide this config if
    the key is different from the default DynamoDB KMS Master Key, which is
    alias/aws/dynamodb.
  </description>
</property>

Out-of-band operations with S3Guard

We call an operation out-of-band (OOB) when a bucket is used by a client with S3Guard, and another client runs a write (e.g delete, move, rename, overwrite) operation on an object in the same bucket without S3Guard.

The definition of behaviour in S3AFileSystem/MetadataStore in case of OOBs: * A client with S3Guard * B client without S3Guard (Directly to S3)

• OOB OVERWRITE, authoritative mode:
• A client creates F1 file
• B client overwrites F1 file with F2 (Same, or different file size)
• A client’s getFileStatus returns F1 metadata

• OOB OVERWRITE, NOT authoritative mode:

• A client creates F1 file
• B client overwrites F1 file with F2 (Same, or different file size)
• A client’s getFileStatus returns F2 metadata. In not authoritative mode we check S3 for the file. If the modification time of the file in S3 is greater than in S3Guard, we can safely return the S3 file metadata and update the cache.

• OOB DELETE, authoritative mode:

• A client creates F file
• B client deletes F file
• A client’s getFileStatus returns that the file is still there

• OOB DELETE, NOT authoritative mode:

• A client creates F file
• B client deletes F file
• A client’s getFileStatus returns that the file is still there

Note: authoritative and NOT authoritative mode behaves the same at OOB DELETE case.

The behaviour in case of getting directory listings: * File status in metadata store gets updated during the listing the same way as in getFileStatus.

Create a table: s3guard init

hadoop s3guard init -meta URI ( -region REGION | s3a://BUCKET )

Creates and initializes an empty metadata store.

A DynamoDB metadata store can be initialized with additional parameters pertaining to capacity.

If these values are both zero, then an on-demand DynamoDB table is created; if positive values then they set the Provisioned Throughput of the table.

[-write PROVISIONED_WRITES] [-read PROVISIONED_READS]

Server side encryption (SSE) can be enabled with AWS managed customer master key (CMK), or customer managed CMK. By default the DynamoDB table will be encrypted with AWS owned CMK. To use a customer managed CMK, you can specify its KMS key ID, ARN, alias name, or alias ARN. If not specified, the default AWS managed CMK for DynamoDB “alias/aws/dynamodb” will be used.

[-sse [-cmk KMS_CMK_ID]]

Tag argument can be added with a key=value list of tags. The table for the metadata store will be created with these tags in DynamoDB.

[-tag key=value;]

Example 1

hadoop s3guard init -meta dynamodb://ireland-team -write 0 -read 0 s3a://ireland-1

Creates an on-demand table “ireland-team”, in the same location as the S3 bucket “ireland-1”.

Example 2

hadoop s3guard init -meta dynamodb://ireland-team -region eu-west-1 --read 0 --write 0

Creates a table “ireland-team” in the region “eu-west-1.amazonaws.com”

Example 3

hadoop s3guard init -meta dynamodb://ireland-team -tag tag1=first;tag2=second;

Creates a table “ireland-team” with tags “first” and “second”. The read and write capacity will be those of the site configuration’s values of fs.s3a.s3guard.ddb.table.capacity.read and fs.s3a.s3guard.ddb.table.capacity.write; if these are both zero then it will be an on-demand table.

Example 4

hadoop s3guard init -meta dynamodb://ireland-team -sse

Creates a table “ireland-team” with server side encryption enabled. The CMK will be using the default AWS managed “alias/aws/dynamodb”.

Import a bucket: s3guard import

hadoop s3guard import [-meta URI] [-authoritative] [-verbose] s3a://PATH

Pre-populates a metadata store according to the current contents of an S3 bucket/path. If the -meta option is omitted, the binding information is taken from the core-site.xml configuration.

Usage

hadoop s3guard import

import [OPTIONS] [s3a://PATH]
    import metadata from existing S3 data

Common options:
  -authoritative - Mark imported directory data as authoritative.
  -verbose - Verbose Output.
  -meta URL - Metadata repository details (implementation-specific)

Amazon DynamoDB-specific options:
  -region REGION - Service region for connections

  URLs for Amazon DynamoDB are of the form dynamodb://TABLE_NAME.
  Specifying both the -region option and an S3A path
  is not supported.

Example

Import all files and directories in a bucket into the S3Guard table.

hadoop s3guard import s3a://ireland-1

Import a directory tree, marking directories as authoritative.

hadoop s3guard import -authoritative -verbose s3a://ireland-1/fork-0008

2020-01-03 12:05:18,321 [main] INFO - Metadata store DynamoDBMetadataStore{region=eu-west-1,
 tableName=s3guard-metadata, tableArn=arn:aws:dynamodb:eu-west-1:980678866538:table/s3guard-metadata} is initialized.
2020-01-03 12:05:18,324 [main] INFO - Starting: Importing s3a://ireland-1/fork-0008
2020-01-03 12:05:18,324 [main] INFO - Importing directory s3a://ireland-1/fork-0008
2020-01-03 12:05:18,537 [main] INFO - Dir  s3a://ireland-1/fork-0008/test/doTestListFiles-0-0-0-false
2020-01-03 12:05:18,630 [main] INFO - Dir  s3a://ireland-1/fork-0008/test/doTestListFiles-0-0-0-true
2020-01-03 12:05:19,142 [main] INFO - Dir  s3a://ireland-1/fork-0008/test/doTestListFiles-2-0-0-false/dir-0
2020-01-03 12:05:19,191 [main] INFO - Dir  s3a://ireland-1/fork-0008/test/doTestListFiles-2-0-0-false/dir-1
2020-01-03 12:05:19,240 [main] INFO - Dir  s3a://ireland-1/fork-0008/test/doTestListFiles-2-0-0-true/dir-0
2020-01-03 12:05:19,289 [main] INFO - Dir  s3a://ireland-1/fork-0008/test/doTestListFiles-2-0-0-true/dir-1
2020-01-03 12:05:19,314 [main] INFO - Updated S3Guard with 0 files and 6 directory entries
2020-01-03 12:05:19,315 [main] INFO - Marking directory tree s3a://ireland-1/fork-0008 as authoritative
2020-01-03 12:05:19,342 [main] INFO - Importing s3a://ireland-1/fork-0008: duration 0:01.018s
Inserted 6 items into Metadata Store

Compare a S3Guard table and the S3 Store: s3guard diff

hadoop s3guard diff [-meta URI] s3a://BUCKET

Lists discrepancies between a metadata store and bucket. Note that depending on how S3Guard is used, certain discrepancies are to be expected.

Example

hadoop s3guard diff s3a://ireland-1

Display information about a bucket, s3guard bucket-info

Prints and optionally checks the s3guard and encryption status of a bucket.

hadoop s3guard bucket-info [-guarded] [-unguarded] [-auth] [-nonauth] [-magic] [-encryption ENCRYPTION] s3a://BUCKET

Options

The server side encryption options are not directly related to S3Guard, but it is often convenient to check them at the same time.

Example

hadoop s3guard bucket-info -guarded -magic s3a://ireland-1

List the details of bucket s3a://ireland-1, mandating that it must have S3Guard enabled (“-guarded”) and that support for the magic S3A committer is enabled (“-magic”)

Filesystem s3a://ireland-1
Location: eu-west-1
Filesystem s3a://ireland-1 is using S3Guard with store DynamoDBMetadataStore{region=eu-west-1, tableName=ireland-1}
Authoritative S3Guard: fs.s3a.metadatastore.authoritative=false
Metadata Store Diagnostics:
  ARN=arn:aws:dynamodb:eu-west-1:00000000:table/ireland-1
  billing-mode=provisioned
  description=S3Guard metadata store in DynamoDB
  name=ireland-1
  read-capacity=20
  region=eu-west-1
  retryPolicy=ExponentialBackoffRetry(maxRetries=9, sleepTime=100 MILLISECONDS)
  size=12812
  status=ACTIVE
  table={AttributeDefinitions: [{AttributeName: child,AttributeType: S},
    {AttributeName: parent,AttributeType: S}],TableName: ireland-1,
    KeySchema: [{AttributeName: parent,KeyType: HASH}, {AttributeName: child,KeyType: RANGE}],
    TableStatus: ACTIVE,
    CreationDateTime: Fri Aug 25 19:07:25 BST 2017,
    ProvisionedThroughput: {LastIncreaseDateTime: Tue Aug 29 11:45:18 BST 2017,
    LastDecreaseDateTime: Wed Aug 30 15:37:51 BST 2017,
    NumberOfDecreasesToday: 1,
    ReadCapacityUnits: 20,WriteCapacityUnits: 20},
    TableSizeBytes: 12812,ItemCount: 91,
    TableArn: arn:aws:dynamodb:eu-west-1:00000000:table/ireland-1,}
  write-capacity=20
The "magic" committer is supported

S3A Client
  Signing Algorithm: fs.s3a.signing-algorithm=(unset)
  Endpoint: fs.s3a.endpoint=s3-eu-west-1.amazonaws.com
  Encryption: fs.s3a.server-side-encryption-algorithm=none
  Input seek policy: fs.s3a.experimental.input.fadvise=normal
  Change Detection Source: fs.s3a.change.detection.source=etag
  Change Detection Mode: fs.s3a.change.detection.mode=server
Delegation token support is disabled

This listing includes all the information about the table supplied from

hadoop s3guard bucket-info -unguarded -encryption none s3a://landsat-pds

List the S3Guard status of clients of the public landsat-pds bucket, and verifies that the data is neither tracked with S3Guard nor encrypted.

Filesystem s3a://landsat-pdsLocation: us-west-2
Filesystem s3a://landsat-pds is not using S3Guard
Endpoint: fs.s3a.endpoints3.amazonaws.com
Encryption: fs.s3a.server-side-encryption-algorithm=none
Input seek policy: fs.s3a.experimental.input.fadvise=normal

Note that other clients may have a S3Guard table set up to store metadata on this bucket; the checks are all done from the perspective of the configuration settings of the current client.

hadoop s3guard bucket-info -guarded -auth s3a://landsat-pds

Require the bucket to be using S3Guard in authoritative mode. This will normally fail against this specific bucket.

List or Delete Leftover Multipart Uploads: s3guard uploads

Lists or deletes all pending (uncompleted) multipart uploads older than given age.

hadoop s3guard uploads (-list | -abort | -expect <num-uploads>) [-verbose] \
    [-days <days>] [-hours <hours>] [-minutes <minutes>] [-seconds <seconds>] \
    [-force] s3a://bucket/prefix

The command lists or deletes all multipart uploads which are older than the given age, and that match the prefix supplied, if any.

For example, to delete all uncompleted multipart uploads older than two days in the folder at s3a://my-bucket/path/to/stuff, use the following command:

hadoop s3guard uploads -abort -days 2 s3a://my-bucket/path/to/stuff

We recommend running with -list first to confirm the parts shown are those that you wish to delete. Note that the command will prompt you with a “Are you sure?” prompt unless you specify the -force option. This is to safeguard against accidental deletion of data, which is especially risky without a long age parameter as it can affect in-fight uploads.

The -expect option is similar to -list, except it is silent by default, and terminates with a success or failure exit code depending on whether or not the supplied number matches the number of uploads found that match the given options (path, age).

Delete a table: s3guard destroy

Deletes a metadata store. With DynamoDB as the store, this means the specific DynamoDB table use to store the metadata.

hadoop s3guard destroy [-meta URI] ( -region REGION | s3a://BUCKET )

This does not delete the bucket, only the S3Guard table which it is bound to.

Examples

hadoop s3guard destroy s3a://ireland-1

Deletes the table which the bucket ireland-1 is configured to use as its MetadataStore.

hadoop s3guard destroy -meta dynamodb://ireland-team -region eu-west-1

Clean up a table, s3guard prune

Delete all file entries in the MetadataStore table whose object “modification time” is older than the specified age.

hadoop s3guard prune [-days DAYS] [-hours HOURS] [-minutes MINUTES]
    [-seconds SECONDS] [-tombstone] [-meta URI] ( -region REGION | s3a://BUCKET )

A time value of hours, minutes and/or seconds must be supplied.

1. This does not delete the entries in the bucket itself.
2. The modification time is effectively the creation time of the objects in the S3 Bucket.
3. If an S3A URI is supplied, only the entries in the table specified by the URI and older than a specific age are deleted.

The -tombstone option instructs the operation to only purge “tombstones”, markers of deleted files. These tombstone markers are only used briefly, to indicate that a recently deleted file should not be found in listings. As a result, there is no adverse consequences in regularly pruning old tombstones.

Example

hadoop s3guard prune -days 7 s3a://ireland-1

Deletes all entries in the S3Guard table for files older than seven days from the table associated with s3a://ireland-1.

hadoop s3guard prune -tombstone -days 7 s3a://ireland-1/path_prefix/

Deletes all entries in the S3Guard table for tombstones older than seven days from the table associated with s3a://ireland-1 and with the prefix path_prefix

hadoop s3guard prune -hours 1 -minutes 30 -meta dynamodb://ireland-team -region eu-west-1

Delete all file entries more than 90 minutes old from the table "ireland-team" in the region eu-west-1.

Audit the "authoritative state of a DynamoDB Table, s3guard authoritative

This recursively checks a S3Guard table to verify that all directories underneath are marked as “authoritative”, and/or that the configuration is set for the S3A client to treat files and directories urnder the path as authoritative.

hadoop s3guard authoritative

authoritative [OPTIONS] [s3a://PATH]
    Audits a DynamoDB S3Guard repository for all the entries being 'authoritative'

Options:
  -required Require directories under the path to be authoritative.
  -check-config Check the configuration for the path to be authoritative
  -verbose Verbose Output.

Verify that a path under an object store is declared to be authoritative in the cluster configuration -and therefore that file entries will not be validated against S3, and that directories marked as “authoritative” in the S3Guard table will be treated as complete.

hadoop s3guard authoritative -check-config s3a:///ireland-1/fork-0003/test/

2020-01-03 11:42:29,147 [main] INFO  Metadata store DynamoDBMetadataStore{
  region=eu-west-1, tableName=s3guard-metadata, tableArn=arn:aws:dynamodb:eu-west-1:980678866538:table/s3guard-metadata} is initialized.
Path /fork-0003/test is not configured to be authoritative

Scan a store and report which directories are not marked as authoritative.

hadoop s3guard authoritative s3a://ireland-1/

2020-01-03 11:51:58,416 [main] INFO  - Metadata store DynamoDBMetadataStore{region=eu-west-1, tableName=s3guard-metadata, tableArn=arn:aws:dynamodb:eu-west-1:980678866538:table/s3guard-metadata} is initialized.
2020-01-03 11:51:58,419 [main] INFO  - Starting: audit s3a://ireland-1/
2020-01-03 11:51:58,422 [main] INFO  - Root directory s3a://ireland-1/
2020-01-03 11:51:58,469 [main] INFO  -   files 4; directories 12
2020-01-03 11:51:58,469 [main] INFO  - Directory s3a://ireland-1/Users
2020-01-03 11:51:58,521 [main] INFO  -   files 0; directories 1
2020-01-03 11:51:58,522 [main] INFO  - Directory s3a://ireland-1/fork-0007
2020-01-03 11:51:58,573 [main] INFO  - Directory s3a://ireland-1/fork-0001
2020-01-03 11:51:58,626 [main] INFO  -   files 0; directories 1
2020-01-03 11:51:58,626 [main] INFO  - Directory s3a://ireland-1/fork-0006
2020-01-03 11:51:58,676 [main] INFO  - Directory s3a://ireland-1/path
2020-01-03 11:51:58,734 [main] INFO  -   files 0; directories 1
2020-01-03 11:51:58,735 [main] INFO  - Directory s3a://ireland-1/fork-0008
2020-01-03 11:51:58,802 [main] INFO  -   files 0; directories 1
2020-01-03 11:51:58,802 [main] INFO  - Directory s3a://ireland-1/fork-0004
2020-01-03 11:51:58,854 [main] INFO  -   files 0; directories 1
2020-01-03 11:51:58,855 [main] WARN   - Directory s3a://ireland-1/fork-0003 is not authoritative
2020-01-03 11:51:58,905 [main] INFO  -   files 0; directories 1
2020-01-03 11:51:58,906 [main] INFO  - Directory s3a://ireland-1/fork-0005
2020-01-03 11:51:58,955 [main] INFO  - Directory s3a://ireland-1/customsignerpath2
2020-01-03 11:51:59,006 [main] INFO  - Directory s3a://ireland-1/fork-0002
2020-01-03 11:51:59,063 [main] INFO  -   files 0; directories 1
2020-01-03 11:51:59,064 [main] INFO  - Directory s3a://ireland-1/customsignerpath1
2020-01-03 11:51:59,121 [main] INFO  - Directory s3a://ireland-1/Users/stevel
2020-01-03 11:51:59,170 [main] INFO  -   files 0; directories 1
2020-01-03 11:51:59,171 [main] INFO  - Directory s3a://ireland-1/fork-0001/test
2020-01-03 11:51:59,233 [main] INFO  - Directory s3a://ireland-1/path/style
2020-01-03 11:51:59,282 [main] INFO  -   files 0; directories 1
2020-01-03 11:51:59,282 [main] INFO  - Directory s3a://ireland-1/fork-0008/test
2020-01-03 11:51:59,338 [main] INFO  -   files 15; directories 10
2020-01-03 11:51:59,339 [main] INFO  - Directory s3a://ireland-1/fork-0004/test
2020-01-03 11:51:59,394 [main] WARN   - Directory s3a://ireland-1/fork-0003/test is not authoritative
2020-01-03 11:51:59,451 [main] INFO  -   files 35; directories 1
2020-01-03 11:51:59,451 [main] INFO  - Directory s3a://ireland-1/fork-0002/test
2020-01-03 11:51:59,508 [main] INFO  - Directory s3a://ireland-1/Users/stevel/Projects
2020-01-03 11:51:59,558 [main] INFO  -   files 0; directories 1
2020-01-03 11:51:59,559 [main] INFO  - Directory s3a://ireland-1/path/style/access
2020-01-03 11:51:59,610 [main] INFO  - Directory s3a://ireland-1/fork-0008/test/doTestListFiles-0-2-0-false
2020-01-03 11:51:59,660 [main] INFO  - Directory s3a://ireland-1/fork-0008/test/doTestListFiles-0-2-1-false
2020-01-03 11:51:59,719 [main] INFO  - Directory s3a://ireland-1/fork-0008/test/doTestListFiles-0-0-0-true
2020-01-03 11:51:59,773 [main] INFO  - Directory s3a://ireland-1/fork-0008/test/doTestListFiles-2-0-0-true
2020-01-03 11:51:59,824 [main] INFO  -   files 0; directories 2
2020-01-03 11:51:59,824 [main] INFO  - Directory s3a://ireland-1/fork-0008/test/doTestListFiles-0-2-1-true
2020-01-03 11:51:59,879 [main] INFO  - Directory s3a://ireland-1/fork-0008/test/doTestListFiles-0-0-1-false
2020-01-03 11:51:59,939 [main] INFO  - Directory s3a://ireland-1/fork-0008/test/doTestListFiles-0-0-0-false
2020-01-03 11:51:59,990 [main] INFO  - Directory s3a://ireland-1/fork-0008/test/doTestListFiles-0-2-0-true
2020-01-03 11:52:00,042 [main] INFO  - Directory s3a://ireland-1/fork-0008/test/doTestListFiles-2-0-0-false
2020-01-03 11:52:00,094 [main] INFO  -   files 0; directories 2
2020-01-03 11:52:00,094 [main] INFO  - Directory s3a://ireland-1/fork-0008/test/doTestListFiles-0-0-1-true
2020-01-03 11:52:00,144 [main] WARN   - Directory s3a://ireland-1/fork-0003/test/ancestor is not authoritative
2020-01-03 11:52:00,197 [main] INFO  - Directory s3a://ireland-1/Users/stevel/Projects/hadoop-trunk
2020-01-03 11:52:00,245 [main] INFO  -   files 0; directories 1
2020-01-03 11:52:00,245 [main] INFO  - Directory s3a://ireland-1/fork-0008/test/doTestListFiles-2-0-0-true/dir-0
2020-01-03 11:52:00,296 [main] INFO  - Directory s3a://ireland-1/fork-0008/test/doTestListFiles-2-0-0-true/dir-1
2020-01-03 11:52:00,346 [main] INFO  - Directory s3a://ireland-1/fork-0008/test/doTestListFiles-2-0-0-false/dir-0
2020-01-03 11:52:00,397 [main] INFO  - Directory s3a://ireland-1/fork-0008/test/doTestListFiles-2-0-0-false/dir-1
2020-01-03 11:52:00,479 [main] INFO  - Directory s3a://ireland-1/Users/stevel/Projects/hadoop-trunk/hadoop-tools
2020-01-03 11:52:00,530 [main] INFO  -   files 0; directories 1
2020-01-03 11:52:00,530 [main] INFO  - Directory s3a://ireland-1/Users/stevel/Projects/hadoop-trunk/hadoop-tools/hadoop-aws
2020-01-03 11:52:00,582 [main] INFO  -   files 0; directories 1
2020-01-03 11:52:00,582 [main] INFO  - Directory s3a://ireland-1/Users/stevel/Projects/hadoop-trunk/hadoop-tools/hadoop-aws/target
2020-01-03 11:52:00,636 [main] INFO  -   files 0; directories 1
2020-01-03 11:52:00,637 [main] INFO  - Directory s3a://ireland-1/Users/stevel/Projects/hadoop-trunk/hadoop-tools/hadoop-aws/target/test-dir
2020-01-03 11:52:00,691 [main] INFO  -   files 0; directories 3
2020-01-03 11:52:00,691 [main] INFO  - Directory s3a://ireland-1/Users/stevel/Projects/hadoop-trunk/hadoop-tools/hadoop-aws/target/test-dir/2
2020-01-03 11:52:00,752 [main] INFO  - Directory s3a://ireland-1/Users/stevel/Projects/hadoop-trunk/hadoop-tools/hadoop-aws/target/test-dir/5
2020-01-03 11:52:00,807 [main] INFO  - Directory s3a://ireland-1/Users/stevel/Projects/hadoop-trunk/hadoop-tools/hadoop-aws/target/test-dir/8
2020-01-03 11:52:00,862 [main] INFO   - Scanned 45 directories - 3 were not marked as authoritative
2020-01-03 11:52:00,863 [main] INFO  - audit s3a://ireland-1/: duration 0:02.444s

Scan the path/bucket and fail if any entry is non-authoritative.

hadoop s3guard authoritative -verbose -required s3a://ireland-1/

2020-01-03 11:47:40,288 [main] INFO  - Metadata store DynamoDBMetadataStore{region=eu-west-1, tableName=s3guard-metadata, tableArn=arn:aws:dynamodb:eu-west-1:980678866538:table/s3guard-metadata} is initialized.
2020-01-03 11:47:40,291 [main] INFO  - Starting: audit s3a://ireland-1/
2020-01-03 11:47:40,295 [main] INFO  - Root directory s3a://ireland-1/
2020-01-03 11:47:40,336 [main] INFO  -  files 4; directories 12
2020-01-03 11:47:40,336 [main] INFO  - Directory s3a://ireland-1/Users
2020-01-03 11:47:40,386 [main] INFO  -  files 0; directories 1
2020-01-03 11:47:40,386 [main] INFO  - Directory s3a://ireland-1/fork-0007
2020-01-03 11:47:40,435 [main] INFO  -  files 1; directories 0
2020-01-03 11:47:40,435 [main] INFO  - Directory s3a://ireland-1/fork-0001
2020-01-03 11:47:40,486 [main] INFO  -  files 0; directories 1
2020-01-03 11:47:40,486 [main] INFO  - Directory s3a://ireland-1/fork-0006
2020-01-03 11:47:40,534 [main] INFO  -  files 1; directories 0
2020-01-03 11:47:40,535 [main] INFO  - Directory s3a://ireland-1/path
2020-01-03 11:47:40,587 [main] INFO  -  files 0; directories 1
2020-01-03 11:47:40,588 [main] INFO  - Directory s3a://ireland-1/fork-0008
2020-01-03 11:47:40,641 [main] INFO  -  files 0; directories 1
2020-01-03 11:47:40,642 [main] INFO  - Directory s3a://ireland-1/fork-0004
2020-01-03 11:47:40,692 [main] INFO  -  files 0; directories 1
2020-01-03 11:47:40,693 [main] WARN  - Directory s3a://ireland-1/fork-0003 is not authoritative
2020-01-03 11:47:40,693 [main] INFO  - audit s3a://ireland-1/: duration 0:00.402s
2020-01-03 11:47:40,698 [main] INFO  - Exiting with status 46: `s3a://ireland-1/fork-0003': Directory is not marked as authoritative in the S3Guard store

This command is primarily for testing.

Tune the I/O capacity of the DynamoDB Table, s3guard set-capacity

Alter the read and/or write capacity of a s3guard table created with provisioned I/O capacity.

hadoop s3guard set-capacity [--read UNIT] [--write UNIT] ( -region REGION | s3a://BUCKET )

The --read and --write units are those of s3guard init.

It cannot be used to change the I/O capacity of an on demand table (there is no need), and nor can it be used to convert an existing table to being on-demand. For that the AWS console must be used.

Example

hadoop s3guard set-capacity  -read 20 -write 20 s3a://ireland-1

Set the capacity of the table used by bucket s3a://ireland-1 to 20 read and 20 write. (This is a low number, incidentally)

2017-08-30 16:21:26,343 [main] INFO  s3guard.S3GuardTool (S3GuardTool.java:initMetadataStore(229)) - Metadata store DynamoDBMetadataStore{region=eu-west-1, tableName=ireland-1} is initialized.
2017-08-30 16:21:26,344 [main] INFO  s3guard.DynamoDBMetadataStore (DynamoDBMetadataStore.java:updateParameters(1084)) - Current table capacity is read: 25, write: 25
2017-08-30 16:21:26,344 [main] INFO  s3guard.DynamoDBMetadataStore (DynamoDBMetadataStore.java:updateParameters(1086)) - Changing capacity of table to read: 20, write: 20
Metadata Store Diagnostics:
  ARN=arn:aws:dynamodb:eu-west-1:00000000000:table/ireland-1
  billing-mode=provisioned
  description=S3Guard metadata store in DynamoDB
  name=ireland-1
  read-capacity=25
  region=eu-west-1
  retryPolicy=ExponentialBackoffRetry(maxRetries=9, sleepTime=100 MILLISECONDS)
  size=12812
  status=UPDATING
  table={ ... }
  write-capacity=25

After the update, the table status changes to UPDATING; this is a sign that the capacity has been changed.

Repeating the same command will not change the capacity, as both read and write values match that already in use.

2017-08-30 16:24:35,337 [main] INFO  s3guard.DynamoDBMetadataStore (DynamoDBMetadataStore.java:updateParameters(1090)) - Table capacity unchanged at read: 20, write: 20
Metadata Store Diagnostics:
  ARN=arn:aws:dynamodb:eu-west-1:00000000000:table/ireland-1
  billing-mode=provisioned
  description=S3Guard metadata store in DynamoDB
  name=ireland-1
  read-capacity=20
  region=eu-west-1
  retryPolicy=ExponentialBackoffRetry(maxRetries=9, sleepTime=100 MILLISECONDS)
  size=12812
  status=ACTIVE
  table={ ... }
  write-capacity=20

Note: There is a limit to how many times in a 24 hour period the capacity of a bucket can be changed, either through this command or the AWS console.

Check the consistency of the metadata store, s3guard fsck

Compares S3 with MetadataStore, and returns a failure status if any rules or invariants are violated. Only works with DynamoDB metadata stores.

hadoop s3guard fsck [-check | -internal] [-fix] (s3a://BUCKET | s3a://PATH_PREFIX)

-check operation checks the metadata store from the S3 perspective, but does not fix any issues. The consistency issues will be logged in ERROR loglevel.

-internal operation checks the internal consistency of the metadata store, but does not fix any issues.

-fix operation fixes consistency issues between the metadatastore and the S3 bucket. This parameter is optional, and can be used together with check or internal parameters, but not alone. The following fix is implemented: - Remove orphan entries from DDB

The errors found will be logged at the ERROR log level.

Note: -check and -internal operations can be used only as separate commands. Running fsck with both will result in an error.

Example

hadoop s3guard fsck -check s3a://ireland-1/path_prefix/

Checks the metadata store while iterating through the S3 bucket. The path_prefix will be used as the root element of the check.

hadoop s3guard fsck -internal s3a://ireland-1/path_prefix/

Checks the metadata store internal consistency. The path_prefix will be used as the root element of the check.

Enabling a log message whenever S3Guard is disabled

When dealing with support calls related to the S3A connector, “is S3Guard on?” is the usual opening question. This can be determined by looking at the application logs for messages about S3Guard starting -the absence of S3Guard can only be inferred by the absence of such messages.

There is a another strategy: have the S3A Connector log whenever S3Guard is not enabled

This can be done in the configuration option fs.s3a.s3guard.disabled.warn.level

<property>
 <name>fs.s3a.s3guard.disabled.warn.level</name>
 <value>silent</value>
 <description>
   Level to print a message when S3Guard is disabled.
   Values: 
   "warn": log at WARN level
   "inform": log at INFO level
   "silent": log at DEBUG level
   "fail": raise an exception
 </description>
</property>

The fail option is clearly more than logging; it exists as an extreme debugging tool. Use with care.

Failure Semantics

Operations which modify metadata will make changes to S3 first. If, and only if, those operations succeed, the equivalent changes will be made to the Metadata Store.

These changes to S3 and Metadata Store are not fully-transactional: If the S3 operations succeed, and the subsequent Metadata Store updates fail, the S3 changes will not be rolled back. In this case, an error message will be logged.

Versioning

S3Guard tables are created with a version marker entry and table tag. The entry is created with the primary key and child entry of ../VERSION; the use of a relative path guarantees that it will not be resolved. Table tag key is named s3guard_version.

When the table is initialized by S3Guard, the table will be tagged during the creating and the version marker entry will be created in the table. If the table lacks the version marker entry or tag, S3Guard will try to create it according to the following rules:

1. If the table lacks both version markers AND it’s empty, both markers will be added. If the table is not empty the check throws IOException
2. If there’s no version marker ITEM, the compatibility with the TAG will be checked, and the version marker ITEM will be added if the TAG version is compatible. If the TAG version is not compatible, the check throws OException
3. If there’s no version marker TAG, the compatibility with the ITEM version marker will be checked, and the version marker ITEM will be added if the ITEM version is compatible. If the ITEM version is not compatible, the check throws IOException
4. If the TAG and ITEM versions are both present then both will be checked for compatibility. If the ITEM or TAG version marker is not compatible, the check throws IOException

Note: If the user does not have sufficient rights to tag the table the initialization of S3Guard will not fail, but there will be no version marker tag on the dynamo table.

Versioning policy

1. The version number of an S3Guard table will only be incremented when an incompatible change is made to the table structure —that is, the structure has changed so that it is no longer readable by older versions, or because it has added new mandatory fields which older versions do not create.
2. The version number of S3Guard tables will only be changed by incrementing the value.
3. Updated versions of S3Guard MAY continue to support older version tables.
4. If an incompatible change is made such that existing tables are not compatible, then a means shall be provided to update existing tables. For example: an option in the Command Line Interface, or an option to upgrade tables during S3Guard initialization.

Note: this policy does not indicate any intent to upgrade table structures in an incompatible manner. The version marker in tables exists to support such an option if it ever becomes necessary, by ensuring that all S3Guard client can recognise any version mismatch.

On-Demand Dynamo Capacity

Amazon DynamoDB On-Demand removes the need to pre-allocate I/O capacity for S3Guard tables. Instead the caller is only charged per I/O Operation.

• There are no SLA capacity guarantees. This is generally not an issue for S3Guard applications.
• There’s no explicit limit on I/O capacity, so operations which make heavy use of S3Guard tables (for example: SQL query planning) do not get throttled.
• You are charged more per DynamoDB API call, in exchange for paying nothing when you are not interacting with DynamoDB.
• There’s no way put a limit on the I/O; you may unintentionally run up large bills through sustained heavy load.
• The s3guard set-capacity command fails: it does not make sense any more.

When idle, S3Guard tables are only billed for the data stored, not for any unused capacity. For this reason, there is no performance benefit from sharing a single S3Guard table across multiple buckets.

Creating a S3Guard Table with On-Demand Tables

The default settings for S3Guard are to create on-demand tables; this can also be done explicitly in the s3guard init command by setting the read and write capacities to zero.

hadoop s3guard init -meta dynamodb://ireland-team -write 0 -read 0 s3a://ireland-1

Enabling DynamoDB On-Demand for an existing S3Guard table

You cannot currently convert an existing S3Guard table to being an on-demand table through the s3guard command.

It can be done through the AWS console or the CLI. From the Web console or the command line, switch the billing to pay-per-request.

Once enabled, the read and write capacities of the table listed in the hadoop s3guard bucket-info command become “0”, and the “billing-mode” attribute changes to “per-request”:

> hadoop s3guard bucket-info s3a://example-bucket/

Filesystem s3a://example-bucket
Location: eu-west-1
Filesystem s3a://example-bucket is using S3Guard with store
  DynamoDBMetadataStore{region=eu-west-1, tableName=example-bucket,
  tableArn=arn:aws:dynamodb:eu-west-1:11111122223333:table/example-bucket}
Authoritative S3Guard: fs.s3a.metadatastore.authoritative=false
Metadata Store Diagnostics:
  ARN=arn:aws:dynamodb:eu-west-1:11111122223333:table/example-bucket
  billing-mode=per-request
  description=S3Guard metadata store in DynamoDB
  name=example-bucket
  persist.authoritative.bit=true
  read-capacity=0
  region=eu-west-1
  retryPolicy=ExponentialBackoffRetry(maxRetries=9, sleepTime=250 MILLISECONDS)
  size=66797
  status=ACTIVE
  table={AttributeDefinitions:
    [{AttributeName: child,AttributeType: S},
     {AttributeName: parent,AttributeType: S}],
     TableName: example-bucket,
     KeySchema: [{
       AttributeName: parent,KeyType: HASH},
       {AttributeName: child,KeyType: RANGE}],
     TableStatus: ACTIVE,
     CreationDateTime: Thu Oct 11 18:51:14 BST 2018,
     ProvisionedThroughput: {
       LastIncreaseDateTime: Tue Oct 30 16:48:45 GMT 2018,
       LastDecreaseDateTime: Tue Oct 30 18:00:03 GMT 2018,
       NumberOfDecreasesToday: 0,
       ReadCapacityUnits: 0,
       WriteCapacityUnits: 0},
     TableSizeBytes: 66797,
     ItemCount: 415,
     TableArn: arn:aws:dynamodb:eu-west-1:11111122223333:table/example-bucket,
     TableId: a7b0728a-f008-4260-b2a0-aaaaabbbbb,}
  write-capacity=0
The "magic" committer is supported

Autoscaling (Provisioned Capacity) S3Guard tables.

DynamoDB Auto Scaling can automatically increase and decrease the allocated capacity.

Before DynamoDB On-Demand was introduced, autoscaling was the sole form of dynamic scaling.

Experiments with S3Guard and DynamoDB Auto Scaling have shown that any Auto Scaling operation will only take place after callers have been throttled for a period of time. The clients will still need to be configured to retry when overloaded until any extra capacity is allocated. Furthermore, as this retrying will block the threads from performing other operations -including more I/O, the the autoscale may not scale fast enough.

This is why the DynamoDB On-Demand appears is a better option for workloads with Hadoop, Spark, Hive and other applications.

If autoscaling is to be used, we recommend experimenting with the option, based on usage information collected from previous days, and choosing a combination of retry counts and an interval which allow for the clients to cope with some throttling, but not to time-out other applications.

No Versioning Metadata Available

When the first S3AFileSystem clients are upgraded to a version of S3AFileSystem that contains these change tracking features, any existing S3Guard metadata will not contain ETags or object version IDs. Reads of files tracked in such S3Guard metadata will access whatever version of the file is available in S3 at the time of read. Only if the file is subsequently updated will S3Guard start tracking ETag and object version ID and as such generating RemoteFileChangedException if an inconsistency is detected.

Similarly, when S3Guard metadata is pruned, S3Guard will no longer be able to detect an inconsistent read. S3Guard metadata should be retained for at least as long as the perceived possible read-after-overwrite temporary inconsistency window. That window is expected to be short, but there are no guarantees so it is at the administrator’s discretion to weigh the risk.

Known Limitations

hadoop s3guard import [-meta URI] s3a://my-bucket/file-with-bad-metadata

Error: S3Guard table lacks version marker.

The table which was intended to be used as a S3guard metadata store does not have any version marker indicating that it is a S3Guard table.

It may be that this is not a S3Guard table.

• Make sure that this is the correct table name.
• Delete the table, so it can be rebuilt.

Error: Database table is from an incompatible S3Guard version

This indicates that the version of S3Guard which created (or possibly updated) the database table is from a different version that that expected by the S3A client.

This error will also include the expected and actual version numbers.

If the expected version is lower than the actual version, then the version of the S3A client library is too old to interact with this S3Guard-managed bucket. Upgrade the application/library.

If the expected version is higher than the actual version, then the table itself will need upgrading.

Error "DynamoDB table TABLE does not exist in region REGION; auto-creation is turned off"

S3Guard could not find the DynamoDB table for the Metadata Store, and it was not configured to create it. Either the table was missing, or the configuration is preventing S3Guard from finding the table.

1. Verify that the value of fs.s3a.s3guard.ddb.table is correct.
2. If the region for an existing table has been set in fs.s3a.s3guard.ddb.region, verify that the value is correct.
3. If the region is not set, verify that the table exists in the same region as the bucket being used.
4. Create the table if necessary.

Error "The level of configured provisioned throughput for the table was exceeded"

org.apache.hadoop.fs.s3a.AWSServiceThrottledException: listFiles on s3a://bucket/10/d1/d2/d3:
com.amazonaws.services.dynamodbv2.model.ProvisionedThroughputExceededException:
The level of configured provisioned throughput for the table was exceeded.
Consider increasing your provisioning level with the UpdateTable API.
(Service: AmazonDynamoDBv2; Status Code: 400;
Error Code: ProvisionedThroughputExceededException;

The I/O load of clients of the (shared) DynamoDB table was exceeded.

1. Switch to On-Demand Dynamo DB tables (AWS console)
2. Increase the capacity of the DynamoDB table (AWS console or s3guard set-capacity)/
3. Increase the retry count and/or sleep time of S3Guard on throttle events (Hadoop configuration).

Error Max retries exceeded

The I/O load of clients of the (shared) DynamoDB table was exceeded, and the number of attempts to retry the operation exceeded the configured amount.

1. Switch to On-Demand Dynamo DB tables (AWS console).
2. Increase the capacity of the DynamoDB table.
3. Increase the retry count and/or sleep time of S3Guard on throttle events.

Error when running set-capacity: org.apache.hadoop.fs.s3a.AWSServiceThrottledException: ProvisionTable

org.apache.hadoop.fs.s3a.AWSServiceThrottledException: ProvisionTable on s3guard-example:
com.amazonaws.services.dynamodbv2.model.LimitExceededException:
Subscriber limit exceeded: Provisioned throughput decreases are limited within a given UTC day.
After the first 4 decreases, each subsequent decrease in the same UTC day can be performed at most once every 3600 seconds.
Number of decreases today: 6.
Last decrease at Wednesday, July 25, 2018 8:48:14 PM UTC.
Next decrease can be made at Wednesday, July 25, 2018 9:48:14 PM UTC

There’s are limit on how often you can change the capacity of an DynamoDB table; if you call set-capacity too often, it fails. Wait until the after the time indicated and try again.

Error Invalid region specified

java.io.IOException: Invalid region specified "iceland-2":
  Region can be configured with fs.s3a.s3guard.ddb.region:
  us-gov-west-1, us-east-1, us-east-2, us-west-1, us-west-2,
  eu-west-1, eu-west-2, eu-west-3, eu-central-1, ap-south-1,
  ap-southeast-1, ap-southeast-2, ap-northeast-1, ap-northeast-2,
  sa-east-1, cn-north-1, cn-northwest-1, ca-central-1
  at org.apache.hadoop.fs.s3a.s3guard.DynamoDBClientFactory$DefaultDynamoDBClientFactory.getRegion
  at org.apache.hadoop.fs.s3a.s3guard.DynamoDBClientFactory$DefaultDynamoDBClientFactory.createDynamoDBClient

The region specified in fs.s3a.s3guard.ddb.region is invalid.

“Neither ReadCapacityUnits nor WriteCapacityUnits can be specified when BillingMode is PAY_PER_REQUEST”

ValidationException; One or more parameter values were invalid:
  Neither ReadCapacityUnits nor WriteCapacityUnits can be specified when
  BillingMode is PAY_PER_REQUEST
  (Service: AmazonDynamoDBv2; Status Code: 400; Error Code: ValidationException)

On-Demand DynamoDB tables do not have any fixed capacity -it is an error to try to change it with the set-capacity command.

MetadataPersistenceException

A filesystem write operation failed to persist metadata to S3Guard. The file was successfully written to S3 and now the S3Guard metadata is likely to be out of sync.

See Fail on Error for more detail.

Error RemoteFileChangedException

An exception like the following could occur for a couple of reasons:

• the S3Guard metadata is out of sync with the true S3 metadata. For example, the S3Guard DynamoDB table is tracking a different ETag than the ETag shown in the exception. This may suggest the object was updated in S3 without involvement from S3Guard or there was a transient failure when S3Guard tried to write to DynamoDB.

• S3 is exhibiting read-after-overwrite temporary inconsistency. The S3Guard metadata was updated with a new ETag during a recent write, but the current read is not seeing that ETag due to S3 eventual consistency. This exception prevents the reader from an inconsistent read where the reader sees an older version of the file.

org.apache.hadoop.fs.s3a.RemoteFileChangedException: open 's3a://my-bucket/test/file.txt':
  Change reported by S3 while reading at position 0.
  ETag 4e886e26c072fef250cfaf8037675405 was unavailable
  at org.apache.hadoop.fs.s3a.impl.ChangeTracker.processResponse(ChangeTracker.java:167)
  at org.apache.hadoop.fs.s3a.S3AInputStream.reopen(S3AInputStream.java:207)
  at org.apache.hadoop.fs.s3a.S3AInputStream.lambda$lazySeek$1(S3AInputStream.java:355)
  at org.apache.hadoop.fs.s3a.Invoker.lambda$retry$2(Invoker.java:195)
  at org.apache.hadoop.fs.s3a.Invoker.once(Invoker.java:109)
  at org.apache.hadoop.fs.s3a.Invoker.lambda$retry$3(Invoker.java:265)
  at org.apache.hadoop.fs.s3a.Invoker.retryUntranslated(Invoker.java:322)
  at org.apache.hadoop.fs.s3a.Invoker.retry(Invoker.java:261)
  at org.apache.hadoop.fs.s3a.Invoker.retry(Invoker.java:193)
  at org.apache.hadoop.fs.s3a.Invoker.retry(Invoker.java:215)
  at org.apache.hadoop.fs.s3a.S3AInputStream.lazySeek(S3AInputStream.java:348)
  at org.apache.hadoop.fs.s3a.S3AInputStream.read(S3AInputStream.java:381)
  at java.io.FilterInputStream.read(FilterInputStream.java:83)

Error AWSClientIOException: copyFile caused by NullPointerException

The AWS SDK has an issue where it will throw a relatively generic AmazonClientException caused by NullPointerException when copying a file and specifying a precondition that cannot be met. This can bubble up from S3AFileSystem.rename(). It suggests that the file in S3 is inconsistent with the metadata in S3Guard.

org.apache.hadoop.fs.s3a.AWSClientIOException: copyFile(test/rename-eventually2.dat, test/dest2.dat) on test/rename-eventually2.dat: com.amazonaws.AmazonClientException: Unable to complete transfer: null: Unable to complete transfer: null
  at org.apache.hadoop.fs.s3a.S3AUtils.translateException(S3AUtils.java:201)
  at org.apache.hadoop.fs.s3a.Invoker.once(Invoker.java:111)
  at org.apache.hadoop.fs.s3a.Invoker.lambda$retry$4(Invoker.java:314)
  at org.apache.hadoop.fs.s3a.Invoker.retryUntranslated(Invoker.java:406)
  at org.apache.hadoop.fs.s3a.Invoker.retry(Invoker.java:310)
  at org.apache.hadoop.fs.s3a.Invoker.retry(Invoker.java:285)
  at org.apache.hadoop.fs.s3a.S3AFileSystem.copyFile(S3AFileSystem.java:3034)
  at org.apache.hadoop.fs.s3a.S3AFileSystem.innerRename(S3AFileSystem.java:1258)
  at org.apache.hadoop.fs.s3a.S3AFileSystem.rename(S3AFileSystem.java:1119)
  at org.apache.hadoop.fs.s3a.ITestS3ARemoteFileChanged.lambda$testRenameEventuallyConsistentFile2$6(ITestS3ARemoteFileChanged.java:556)
  at org.apache.hadoop.test.LambdaTestUtils.intercept(LambdaTestUtils.java:498)
  at org.apache.hadoop.fs.s3a.ITestS3ARemoteFileChanged.testRenameEventuallyConsistentFile2(ITestS3ARemoteFileChanged.java:554)
  at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
  at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
  at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
  at java.lang.reflect.Method.invoke(Method.java:498)
  at org.junit.runners.model.FrameworkMethod$1.runReflectiveCall(FrameworkMethod.java:50)
  at org.junit.internal.runners.model.ReflectiveCallable.run(ReflectiveCallable.java:12)
  at org.junit.runners.model.FrameworkMethod.invokeExplosively(FrameworkMethod.java:47)
  at org.junit.internal.runners.statements.InvokeMethod.evaluate(InvokeMethod.java:17)
  at org.junit.internal.runners.statements.RunBefores.evaluate(RunBefores.java:26)
  at org.junit.internal.runners.statements.RunAfters.evaluate(RunAfters.java:27)
  at org.junit.rules.TestWatcher$1.evaluate(TestWatcher.java:55)
  at org.junit.internal.runners.statements.FailOnTimeout$CallableStatement.call(FailOnTimeout.java:298)
  at org.junit.internal.runners.statements.FailOnTimeout$CallableStatement.call(FailOnTimeout.java:292)
  at java.util.concurrent.FutureTask.run(FutureTask.java:266)
  at java.lang.Thread.run(Thread.java:748)
Caused by: com.amazonaws.AmazonClientException: Unable to complete transfer: null
  at com.amazonaws.services.s3.transfer.internal.AbstractTransfer.unwrapExecutionException(AbstractTransfer.java:286)
  at com.amazonaws.services.s3.transfer.internal.AbstractTransfer.rethrowExecutionException(AbstractTransfer.java:265)
  at com.amazonaws.services.s3.transfer.internal.CopyImpl.waitForCopyResult(CopyImpl.java:67)
  at org.apache.hadoop.fs.s3a.impl.CopyOutcome.waitForCopy(CopyOutcome.java:72)
  at org.apache.hadoop.fs.s3a.S3AFileSystem.lambda$copyFile$14(S3AFileSystem.java:3047)
  at org.apache.hadoop.fs.s3a.Invoker.once(Invoker.java:109)
  ... 25 more
Caused by: java.lang.NullPointerException
  at com.amazonaws.services.s3.transfer.internal.CopyCallable.copyInOneChunk(CopyCallable.java:154)
  at com.amazonaws.services.s3.transfer.internal.CopyCallable.call(CopyCallable.java:134)
  at com.amazonaws.services.s3.transfer.internal.CopyMonitor.call(CopyMonitor.java:132)
  at com.amazonaws.services.s3.transfer.internal.CopyMonitor.call(CopyMonitor.java:43)
  at java.util.concurrent.FutureTask.run(FutureTask.java:266)
  at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
  at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
  ... 1 more

Error Attempt to change a resource which is still in use: Table is being deleted

com.amazonaws.services.dynamodbv2.model.ResourceInUseException:
  Attempt to change a resource which is still in use: Table is being deleted:
   s3guard.test.testDynamoDBInitDestroy351245027
    (Service: AmazonDynamoDBv2; Status Code: 400; Error Code: ResourceInUseException;)

You have attempted to call hadoop s3guard destroy on a table which is already being destroyed.