Skip to main content

Use a FIFO Queue to Ensure In-Order Processing

Prismatic's integration runner is designed to process requests in parallel. If you invoke an integration with multiple requests in quick succession, the runner will scale and process all of the requests at the same time.

If you have a workflow that requires requests to be processed in a specific order, or a workflow that requires you to process only one record at a time, you'll need to take some extra steps to ensure that the requests are processed sequentially.

Queueing systems, like Amazon SQS or Azure Service Bus often offer a FIFO (first-in, first-out) queue type which allow you to write messages to the queue, and retrieve them in the order that they were added. You can use a FIFO queue in an integration to queue up requests your integration receives and process them one by one.

First-in, first-out (FIFO) flows​

Regardless of which queueing system you use, the general flow of a FIFO integration is the same:

  • One flow recieves requests in parallel and quickly writes them to the queue.
  • Another flow runs on a regular schedule, reads one message from the queue at a time, and processes messages in serial.
Illustration of a FIFO queue

Additional flows can be added to the integration to handle other tasks, like configuring queues or cleaning up unprocessable requests from a dead letter queue.

FIFO queues in Amazon SQS​

You can use the built-in Amazon SQS component to ensure that your integration processes requests one at a time.

Example Integration

An Amazon SQS-based FIFO integration will generally have four flows:

  1. A "setup" flow that creates and configures the SQS queue. The flow is triggered by an instance deploy trigger so that it runs when a customer deploys an instance. It contains a single action - Create Queue - which is idempotent and can be run many times. We can use the instance's ID as the queue name to ensure that each instance has its own queue.
  2. A "write" flow that receives requests and writes them to the queue. This flow is triggered by a webhook request, and can run several executions in parallel. This flow contains two steps - one step that fetches the queue's URL based on its name, and another step that writes the trigger's payload to the queue. You can once again use the instance's ID as the group ID to ensure only one message is processed at a time.
  3. A "read" flow that reads messages from the queue and processes them. This flow is triggered on a schedule (as often as every minute). The flow enters a loop and requests a message from the queue. If the queue is empty, the flow exits. If a message is returned, the flow processes the message and deletes it from the queue.
  4. A "cleanup" flow that deletes the queue. This flow is triggered by an instance delete trigger so that it runs when a customer deletes an instance. It contains two actions - one that fetches the queue's URL based on its name, and another that deletes the queue.

Message deduplication in Amazon SQS​

When creating an Amazon SQS queue, you have the option to enable content-based deduplication. When enabled, if two messages with the same content are added to the queue within a 5-minute window, only one message will be added to the queue. This is handy if your integration receives duplicate requests.

If you have your own mechanism for determining whether a message is a duplicate, you can disable content-based deduplication and use the Message Deduplication ID in the "write" flow to determine whether a message is a duplicate.

Assured ordering in Amazon SQS​

One concern you may have with this approach is that if the "read" flow runs every minute, and one flow takes time to process a large batch of messages, another "read" flow may begin running. Will this cause messages to be processed out of order? No. Amazon SQS will not return additional messages to any reader until the current message has been processed / deleted (or the message runs past its visibility timeout). If one "read" flow is already processing a message, an additional "read" flow will be told that no new messages are available, and will exit.

Real-time processing in an Amazon SQS FIFO integration​

The "read" flow runs every minute to process messages. If you need to process messages in real-time, you can use a webhook trigger to trigger the "read" flow from the "write" flow after a message is written to the queue.

Handling AWS credentials in an Amazon SQS FIFO integration​

If you leverage Amazon SQS in your integration, you will need to provide AWS credentials to the integration runner. You can either:

  • Have your customer provide their own AWS credentials when they deploy an instance. This requires that your customer have an AWS account and create an IAM user with the appropriate permissions.
  • Use your own AWS account and credentials. You can provide default credentials in the config wizard designer, but mark the connection as organization-visible. Your customers' instances will then use your credentials to access AWS, but they will not be able to see the credentials in the config wizard or through the API.