Skip to main content

Persistence

By default, LocalStack is an ephemeral environment, meaning that, once you terminate your LocalStack instance, all state will be discarded. Persistence is a LocalStack Pro feature that can save and restore the state of LocalStack including all AWS resources and their data.

LocalStack has two distinct persistence mechanisms

  • Snapshot-based persistence: essentially a "pause and resume" feature for your LocalStack container, which takes a snapshot of your LocalStack instance and dumps the data to disk.
  • Cloud Pods: a way to store, distribute, inspect, and version snapshots.

Save strategies​

LocalStack takes point-in-time snapshot of its state and dumps them to disk. There are four strategies that you can choose from that govern when these snapshots are taken. You can select a particular save strategy by setting SNAPSHOT_SAVE_STRATEGY=<strategy>.

  • ON_REQUEST: on every AWS API call that potentially modifies the state of a service, LocalStack will save the state of that service. This strategy minimizes the chance for data loss, but also has significant performance implications. The service has to be locked during snapshotting, meaning that any requests to the particular AWS service will be blocked until the snapshot is complete. In many cases this is just a few milliseconds, but can become significant in some services.
  • ON_SHUTDOWN: the state of all services are saved during the shutdown phase of LocalStack. This strategy has zero performance impact, but is not good when you want to minimize the chance for data loss. Should LocalStack for some reason not shut down properly or is terminated before it can finalize the snapshot, you may be left with an incomplete state on disk.
  • SCHEDULED (default): saves at regular intervals the state of all the services that have been modified since the last snapshot. By default, the flush interval is 15 seconds. It can be configured via the SNAPSHOT_FLUSH_INTERVAL configuration variable. This is a compromise between ON_REQUEST and ON_SHUTDOWN in terms of performance and reliability.
  • MANUAL: turns off automatic snapshotting and gives you control through the internal state endpoints.

Load Strategies​

You can also configure when LocalStack should restore the state snapshots.

  • ON_REQUEST: (default) the state is loaded lazily when the service is requested. This maintains LocalStack's lazy-loading behavior for AWS services.
  • ON_STARTUP: the state of all services in the snapshot is restored when LocalStack starts up. This means that services that have stored state are also started on LocalStack start, which will increase the startup time, but also give you immediate feedback whether the state was restored correctly.
  • MANUAL: turns off automatic loading of snapshots and gives you control through the internal state endpoints.

Endpoints​

As mentioned, with the MANUAL save or load strategy you can trigger snapshotting manually when it best suits your application flow.

  • POST /_localstack/state/<service>/save take a snapshot the given service
  • POST /_localstack/state/<service>/load load the most recent snapshot of the given service

For example, a snapshot for a particular service (e.g., s3) can be triggered by running the following command. The service name refers to the AWS service code.

curl -X POST http://localhost:4566/_localstack/state/s3/save

It is also possible to take and load a snapshot of all the services at once. We provide the following endpoints:

  • POST /_localstack/state/save
  • POST /_localstack/state/load

The response streams line by line the service that has been saved/loaded and the status of the operation.

curl -X POST localhost:4566/_localstack/state/save
{"service": "sqs", "status": "ok"}
{"service": "s3", "status": "ok"}

Service coverage​

Although we are working to support both snapshot-based persistence and Cloud pods for all AWS services, there are some common issues, known limitations, and also services that are not well tested for persistence support. Please help us improve persistence support by reporting bugs on our GitHub issue tracker.

Here is a list of currently supported services and known issues. Persistence for services that are not listed here may work correctly, but are untested and unsupported.

Supported & tested​

  • ACM
  • Amplify
  • API Gateway
  • AppConfig
  • AppSync
  • CloudWatch
  • Cognito
  • DynamodDB
  • IAM
  • Kinesis
  • KMS
  • Lambda
  • RDS: Postgres, MariaDB, MySQL
  • Route53
  • S3
  • SecresManager
  • SNS
  • SQS
  • SSM
  • Stepfunctions

Known limitations​

  • ElastiCache: Redis instances are not restored
  • MSK: Kafka brokers are not restored
  • EC2: works for most resources, but emulated VM data is not restored
  • Firehose: Kinesis delivery streams are not restored
  • RDS: MSSQL database is not restored
  • Neptune: database is not restored
  • DocDB: database is not restored

Not Implemented​

  • MQ

Technical Details​

State persistence in LocalStack works on a per-service basis and uses a custom state serialization protocol based on Python's pickle mechanism. Some services also store application-specific data, which we call assets. For example, when you start an RDS PostgreSQL database, LocalStack not only stores the RDS resource information, but also the PostgreSQL data. Another example is Kinesis, which persists some data in form of JSON objects per account, or DynamoDB that serializes its stat into an SQLite database per account and region.

The current LocalStack snapshot is stored into /var/lib/localstack/state, and separated into api_states (LocalStack internal state), and assets (one directory per service). Here is what this looks like:

/var/lib/localstack/state    # state directory
├── api_states               # serialized LocalStack stores
│   ├── dynamodb
│   │   └── store.state
│   ├── ec2
│   │   └── backend.state
│   ├── iot
│   │   └── store.state
│   └── lambda
│       └── store.state
├── dynamodb                # dynamodb assets
│   ├── 000000000000_eu-central-1.db
│   └── 886002141588_us-east-1.db
└── kinesis                 # kinesis assets
    ├── 000000000000.json
    └── 886002141588.json

To load a snapshot, LocalStack traverses the state directory and deserializes state files to loads them into the memory. When we restore server backends (like an RDS server or DynamoDB server), we make sure that they are configured to use the state stored in the respective asset directory.

When LocalStack saves snapshots, it has to lock the particular service to avoid state pollution. That means that, while a snapshot for a particular service is created, all requests to the service are blocked. Depending on what you are building, you may find this behavior is slowing down your application. In most cases, the ON_SHUTDOWN save strategy should solve this problem.