Skip to main content

State Backends

Meltano is capable of running pipelines that load data incrementally. In order for these pipelines to run efficiently and properly without losing historical data or loading duplicate records, Meltano needs to keep track of the state of a pipeline as part of each run. This state can be stored in Meltano's backend system database, but for Meltano projects running in ephemeral environments or in circumstances where administering a dedicated backend database is undesirable, Meltano also supports persisting state in remote state backends backed by cloud storage services.

Supported Backends​

Installation​

No extra work is needed to use the default system database or local filesystem as a state backend as they are already part of any base Meltano distribution.

To use a cloud storage backend, install Meltano using one of the following extras:

  • meltano[s3] to use the AWS S3 state backend.
  • meltano[azure] to use the Azure Blob Storage state backend.
  • meltano[gcs] to use the Google Cloud Storage state backend.

If the base Meltano distribution is already installed in your environment, install the relevant extra and pip will install only the missing requirements.

State backends are only available in Meltano version 2.10+. If you are using an earlier version, you'll need to upgrade your project via meltano upgrade before you using the state backends feature.

Configuration​

Default (System Database)​

The state backend settings for your Meltano project can be configured much the same as any other Meltano setting. The main setting is state_backend.uri which is set to the special keyword string systemdb by default. This will tell Meltano to store state in the same backend database as it stores other project data.

It can be changed by running meltano config meltano set state_backend.uri <URI for desired state backend> or by directly editing a project's meltano.yml to add the following:

state_backend:
    uri: <URI for desired state backend>

Local File System​

To store state on the local filesystem, set the state_backend.uri setting to file://<absolute path to the directory where you wish to store state>.

For example, to store state directly on the local filesystem at the path ${MELTANO_SYS_DIR_ROOT}/state, run:

meltano config meltano set state_backend.uri 'file:///${MELTANO_SYS_DIR_ROOT}/state'

Note the single quotes which prevent the early expansion of the environment variable.

Meltano will now store the state for a given state_id at the path file:///${MELTANO_SYS_DIR_ROOT}/state/<state_id>/state.json.

The local filesystem state backend will utilize the locking strategy described below.

Azure Blob Storage​

To store state remotely in Azure Blob Storage, set the state_backend.uri setting to azure://<your container_name>/<prefix for state JSON blobs>.

To authenticate to Azure, Meltano will also need a connection string. You can provide this via the state_backend.azure.connection_string setting. If no state_backend.azure.connection_string setting is configured, Meltano will use the value of the AZURE_STORAGE_CONNECTION_STRING environment variable. If the connection string is not provided via either of these methods, Meltano will not be able to authenticate to Azure and any state operations will fail.

AWS S3​

To store state remotely in S3, set the state_backend.uri setting to s3://<your bucket name>/<prefix for state JSON blobs>.

To authenticate to AWS, you can provide an AWS access key ID and AWS secret access key via either of the following methods:

  • Configure the state_backend.s3.aws_access_key_id and state_backend.s3.aws_secret_access_key settings.
  • Pass the access key ID and secret access key directly in the state_backend.uri by setting it to s3://<aws_access_key_id>:<aws_secret_access_key>@<your bucket name>/<prefix for state JSON blobs>.

If credentials are not provided via either of these methods, Meltano will use the values set in the AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY environment variables. If these environment variables are not set, it will use the credentials stored in the environment's AWS credentials file.

If AWS credentials are not found via any of the methods described above, Meltano will not be able to authenticate to S3 and state operations will fail.

Endpoint URL​

Meltano supports service-specific endpoint URLs via:

  • The endpoint_url setting in the shared AWS config file.
  • The AWS_ENDPOINT_URL environment variable.
  • The AWS_ENDPOINT_URL_S3 environment variable.
  • The Meltano state_backend.s3.endpoint_url setting.

The AWS_IGNORE_CONFIGURED_ENDPOINT_URLS environment variable won't have any effect if state_backend.s3.endpoint_url is not null.

For reference, read the AWS documentation on service-specific endpoints.

Google Cloud Storage​

To store state remotely in Google Cloud Storage, set the state_backend.uri setting to gs://<your bucket name>/<prefix for state JSON blobs>.

To authenticate to GCS, you must provide a path to a service account credentials file. These can be configured via the state_backend.gcs.application_credentials setting.

If credentials are not provided via these settings, Meltano will use the value the GOOGLE_APPLICATION_CREDENTIALS environment variable, if it is set.

If GCS credentials are not found via any of the methods described above, Meltano will not be able to authenticate to Google Cloud Storage and state operations will fail.

Locking​

Because the systemdb state backend utilizes a transactional database, it can rely on the database's transaction logic to prevent conflicts among multiple runs of the same pipeline.

For some other state backends, Meltano implements its own simple locking mechanism. This locking mechanism utilizes reasonable defaults but you may wish to configure the locking settings differently. You can do this via two state_backend settings:

  • state_backend.lock_timeout_seconds
  • state_backend.lock_retry_seconds

When Meltano tries to read or write state for a given state_id, it will try to acquire a lock on the state data for that state_id. For example, using the local filesystem state backend with state_backend.uri set to file:///${MELTANO_SYS_DIR_ROOT}/state, it will check to see if a file exists at the path file:///${MELTANO_SYS_DIR_ROOT}/state/<state_id>/lock. If no such file exists, Meltano will create it and write the current UTC timestamp into the file. If the file does exist, Meltano will check the UTC timestamp written in the file to determine when it was locked by another process. Meltano will add the configured value for state_backend.lock_timeout_seconds to the timestamp in the file to determine when the lock expires. If the lock is expired (i.e. if the expiration time is before the time at which Meltano attempts to acquire the new lock), then Meltano will overwrite the lock file with the current timestamp. If the lock is not expired, then Meltano will wait the number of seconds configured by the state_backend.lock_retry_seconds setting and then attempt to acquire the lock again.

In most deployments, it should be rare for the same pipeline to be running in parallel or for manual invocations of the meltano state command to take place during a pipeline's run. But if the default values for lock_timeout_seconds and lock_retry_seconds (10 seconds and 1 second, respectively) cause issues in your deployment, you can configure them to more appropriate values by running meltano config meltano state_backend.lock_timeout_seconds <new value> and meltano config meltano state_backend.lock_retry_seconds <new value> .

Migrating State​

You can migrate state from one backend to another backend using the meltano state get and meltano state set commands.

For example if you've been storing state in Meltano's system database and would like to migrate to S3, you'll need to first store state in a local json file, configure the S3 state backend, and then set state for your job using the local json file.

So to migrate state for a job with the state ID dev:tap-github-to-target-jsonl, you first need to ensure that your meltano project is configured to use the source state backend that currently holds the job state. For this example, we'll use systemdb as our source. To check the current configuration, run meltano config meltano list and then find the value for state_backend.uri:

$ meltano config meltano list | grep state_backend.uri

state_backend.uri [env: MELTANO_STATE_BACKEND_URI] current value: 'systemdb' (default)

Now that we've confirmed that our project is configured to use the correct source state backend, we can get the state for our job and write it to a local file:

$ meltano state get dev:tap-github-to-target-jsonl > dev:tap-github-to-target-jsonl.json

Now we need to reconfigure our meltano project to use the new destination state backend. For this example, we'll use an S3 bucket with the name meltano and store state under the prefix state.

$ meltano config meltano set state_backend.uri "s3://meltano/state"
Meltano setting 'state_backend.uri' was set in `.env`: 's3://meltano/state'

$ meltano config meltano set state_backend.s3.aws_access_key_id <AWS_ACCESS_KEY_ID>
Meltano setting 'state_backend.s3.aws_access_key_id' was set in `.env`: '(redacted)'

$ meltano config meltano set state_backend.s3.aws_secret_access_key <AWS_SECRET_ACCESS_KEY>
Meltano setting 'state_backend.s3.aws_secret_access_key' was set in `.env`: '(redacted)'

Now we can set the state in the new backend using the state file we just wrote:

$ meltano state set dev:tap-github-to-target-jsonl --input-file dev:tap-github-to-taget-jsonl.json

Meltano will prompt you to confirm:

This will overwrite the state's current value. Continue? [y/N]:

Enter y and then Meltano will write the job state to S3.

To do this quickly for multiple jobs, you can loop through the output of meltano state list and use the --force flag for meltano state set to prevent confirmation prompts:

for job_id in $(meltano state list); do meltano state get $job_id > $job_id-state.json; done

To migrate state for some subset of state IDs, use the --pattern flag with meltano state list:

for job_id in $(meltano state list --pattern 'dev:*'); do meltano state get $job_id > $job_id-state.json; done

Then configure your project to use the S3 state backend as we did above.

Now loop through the state json files and set state in S3 for each one:

for state_file in *-state.json; do meltano state set --force ${state_file%-state.json} --input-file $state_file; done
Github LogoEdit this page on Github