DynamoDB Session Handler

Introduction

The DynamoDB Session Handler is a custom session handler for PHP that allows developers to use Amazon DynamoDB as a session store. Using DynamoDB for session storage alleviates issues that occur with session handling in a distributed web application by moving sessions off of the local file system and into a shared location. DynamoDB is fast, scalable, easy to setup, and handles replication of your data automatically.

The DynamoDB Session Handler uses the session_set_save_handler() function to hook DynamoDB operations into PHP's native session functions to allow for a true drop in replacement. This includes support for features like session locking and garbage collection which are a part of PHP's default session handler.

For more information on the Amazon DynamoDB service, please visit the Amazon DynamoDB homepage.

Basic Usage

1. Register the handler

The first step is to instantiate the Amazon DynamoDB client and register the session handler.

require 'vendor/autoload.php';

use Aws\DynamoDb\DynamoDbClient;

$dynamoDb = DynamoDbClient::factory(array('region' => '<region name>'));

$sessionHandler = $dynamoDb->registerSessionHandler(array(
    'table_name' => 'sessions'
));

You can also instantiate the SessionHandler object directly using it's factory method.

require 'vendor/autoload.php';

use Aws\DynamoDb\DynamoDbClient;
use Aws\DynamoDb\Session\SessionHandler;

$dynamoDb = DynamoDbClient::factory(array(
    'region' => '<region name>',
));

$sessionHandler = SessionHandler::factory(array(
    'dynamodb_client' => $dynamoDb,
    'table_name'      => 'sessions',
));
$sessionHandler->register();

2. Create a table for storing your sessions

Before you can actually use the session handler, you need to create a table in which to store the sessions. This can be done ahead of time through the AWS Console for Amazon DynamoDB, or you can use the session handler object (which you've already configured with the table name) by doing the following:

$sessionHandler->createSessionsTable(5, 5);

The two parameters for this function are used to specify the read and write provisioned throughput for the table, respectively.

Note

The createSessionsTable function uses the TableExists waiter internally, so this function call will block until the table exists and is ready to be used.

3. Use PHP sessions like normal

Once the session handler is registered and the table exists, you can write to and read from the session using the $_SESSION superglobal, just like you normally do with PHP's default session handler. The DynamoDB Session Handler encapsulates and abstracts the interactions with Amazon DynamoDB and enables you to simply use PHP's native session functions and interface.

// Start the session
session_start();

// Alter the session data
$_SESSION['user.name'] = 'jeremy';
$_SESSION['user.role'] = 'admin';

// Close the session (optional, but recommended)
session_write_close();

Configuration

You may configure the behavior of the session handler using the following options. All options are optional, but you should make sure to understand what the defaults are.

table_name	The name of the DynamoDB table in which to store the sessions. This defaults to sessions.
hash_key	The name of the hash key in the DynamoDB sessions table. This defaults to id.
session_lifetime	The lifetime of an inactive session before it should be garbage collected. If it is not provided, then the actual lifetime value that will be used is ini_get('session.gc_maxlifetime').
consistent_read	Whether or not the session handler should use consistent reads for the GetItem operation. This defaults to true.
locking_strategy	The strategy used for doing session locking. By default the handler uses the NullLockingStrategy, which means that session locking is not enabled (see the Session Locking section for more information). Valid values for this option include null, 'null', 'pessemistic', or an instance of NullLockingStrategy or PessimisticLockingStrategy.
automatic_gc	Whether or not to use PHP's session auto garbage collection. This defaults to the value of (bool) ini_get('session.gc_probability'), but the recommended value is false. (see the Garbage Collection section for more information).
gc_batch_size	The batch size used for removing expired sessions during garbage collection. This defaults to 25, which is the maximum size of a single BatchWriteItem operation. This value should also take your provisioned throughput into account as well as the timing of your garbage collection.
gc_operation_delay	The delay (in seconds) between service operations performed during garbage collection. This defaults to 0. Increasing this value allows you to throttle your own requests in an attempt to stay within your provisioned throughput capacity during garbage collection.
max_lock_wait_time	Maximum time (in seconds) that the session handler should wait to acquire a lock before giving up. This defaults to 10 and is only used with the PessimisticLockingStrategy.
min_lock_retry_microtime	Minimum time (in microseconds) that the session handler should wait between attempts to acquire a lock. This defaults to 10000 and is only used with the PessimisticLockingStrategy.
max_lock_retry_microtime	Maximum time (in microseconds) that the session handler should wait between attempts to acquire a lock. This defaults to 50000 and is only used with the PessimisticLockingStrategy.
dynamodb_client	The DynamoDbClient object that should be used for performing DynamoDB operations. If you register the session handler from a client object using the registerSessionHandler() method, this will default to the client you are registering it from. If using the SessionHandler::factory() method, you are required to provide an instance of DynamoDbClient.

To configure the Session Handler, you must specify the configuration options when you instantiate the handler. The following code is an example with all of the configuration options specified.

$sessionHandler = $dynamoDb->registerSessionHandler(array(
    'table_name'               => 'sessions',
    'hash_key'                 => 'id',
    'session_lifetime'         => 3600,
    'consistent_read'          => true,
    'locking_strategy'         => null,
    'automatic_gc'             => 0,
    'gc_batch_size'            => 50,
    'max_lock_wait_time'       => 15,
    'min_lock_retry_microtime' => 5000,
    'max_lock_retry_microtime' => 50000,
));

Pricing

Aside from data storage and data transfer fees, the costs associated with using Amazon DynamoDB are calculated based on the provisioned throughput capacity of your table (see the Amazon DynamoDB pricing details). Throughput is measured in units of Write Capacity and Read Capacity. The Amazon DynamoDB homepage says:

A unit of Write Capacity enables you to perform one write per second for items of up to 1KB in size. Similarly, a unit of Read Capacity enables you to perform one strongly consistent read per second (or two eventually consistent reads per second) of items of up to 1KB in size. Larger items will require more capacity. You can calculate the number of units of read and write capacity you need by estimating the number of reads or writes you need to do per second and multiplying by the size of your items (rounded up to the nearest KB).

Ultimately, the throughput and the costs required for your sessions table is going to correlate with your expected traffic and session size. The following table explains the amount of read and write operations that are performed on your DynamoDB table for each of the session functions.

Read via session_start() (Using NullLockingStrategy)	1 read operation (only 0.5 if consistent_read is false). (Conditional) 1 write operation to delete the session if it is expired.
Read via session_start() (Using PessimisticLockingStrategy)	A minimum of 1 write operation. (Conditional) Additional write operations for each attempt at acquiring a lock on the session. Based on configured lock wait time and retry options. (Conditional) 1 write operation to delete the session if it is expired.
Write via session_write_close()	1 write operation.
Delete via session_destroy()	1 write operation.
Garbage Collection	0.5 read operations per KB of data in the table to scan for expired sessions. 1 write operation per expired item to delete it.

Session Locking

The DynamoDB Session Handler supports pessimistic session locking in order to mimic the behavior of PHP's default session handler. By default the DynamoDB Session Handler has this feature turned off since it can become a performance bottleneck and drive up costs, especially when an application accesses the session when using ajax requests or iframes. You should carefully consider whether or not your application requires session locking or not before enabling it.

By default the session handler uses the NullLockingStrategy which does not do any session locking. To enable session locking, you should use the PessimisticLockingStrategy, which can be specified when the session handler is created.

$sessionHandler = $dynamoDb->registerSessionHandler(array(
    'table_name'       => 'sessions',
    'locking_strategy' => 'pessimistic',
));

Garbage Collection

The DynamoDB Session Handler supports session garbage collection by using a series of Scan and BatchWriteItem operations. Due to the nature of how the Scan operation works and in order to find all of the expired sessions and delete them, the garbage collection process can require a lot of provisioned throughput.

For this reason it is discouraged to rely on the PHP's normal session garbage collection triggers (i.e., the session.gc_probability and session.gc_divisor ini settings). A better practice is to set session.gc_probability to 0 and schedule the garbage collection to occur during an off-peak time where a burst of consumed throughput will not disrupt the rest of the application.

For example, you could have a nightly cron job trigger a script to run the garbage collection. This script might look something like the following:

require 'vendor/autoload.php';

use Aws\DynamoDb\DynamoDbClient;
use Aws\DynamoDb\Session\SessionHandler;

$dynamoDb = DynamoDbClient::factory(array(
    'region' => '<region name>',
));

$sessionHandler = SessionHandler::factory(array(
    'dynamodb_client' => $dynamoDb,
    'table_name'      => 'sessions',
));

$sessionHandler->garbageCollect();

You can also use the gc_operation_delay configuration option on the session handler to introduce delays in between the Scan and BatchWriteItem operations that are performed by the garbage collection process. This will increase the amount of time it takes the garbage collection to complete, but it can help you spread out the requests made by the session handler in order to help you stay close to or within your provisioned throughput capacity during garbage collection.

Best Practices

1. Create your sessions table in a region that is geographically closest to or in the same region as your application servers. This will ensure the lowest latency between your application and DynamoDB database.
2. Choose the provisioned throughput capacity of your sessions table carefully, taking into account the expected traffic to your application and the expected size of your sessions.
3. Monitor your consumed throughput through the AWS Management Console or with Amazon CloudWatch and adjust your throughput settings as needed to meet the demands of your application.
4. Keep the size of your sessions small. Sessions that are less than 1KB will perform better and require less provisioned throughput capacity.
5. Do not use session locking unless your application requires it.
6. Instead of using PHP's built-in session garbage collection triggers, schedule your garbage collection via a cron job, or another scheduling mechanism, to run during off-peak hours. Use the gc_operation_delay option to add delays in between the requests performed for the garbage collection process.