Zero-downtime Platform Updates

Tecton is committed to providing the most reliable Feature Platform while also delivering ever-increasing value through continual innovation.

To meet this commitment, we release updates to the Tecton Platform each week, which allow us to regularly deliver service improvements in the form of new features, enhancements, and fixes. Releases are zero-downtime, meaning they have no impact on platform performance and availability.

Pre-release testing and validation

At Tecton, release quality is a top priority. Before each update is released, it goes through a full suite of validation tests, including full performance and reliability regression testing on internal accounts for all supported cloud and data platforms.

Tecton has developed processes to verify the correctness of a release on sample customer workloads. Tecton has also developed a canary process to verify the correctness of feature values served on sample real-time get-feature requests. See our Canary process blog post for more background on how we approach pre-release validation.

Staged release schedule

Updates are staged to Tecton accounts in two waves over three days. This staged approach enables Tecton to monitor activity as the update is released and quickly respond in the unlikely case that any issues occur.

• Tuesday, 2pm - 4pm Pacific Time: Wave 1 provides early access for designated accounts. Customers may designate development/testing accounts to participate in early access to test their production workloads against Tecton Platform updates before they are released to production accounts.
• Wednesday, 2pm - 4pm Pacific Time: Wave 1 re-release. In the unlikely occasion that an issue is detected on Day 1, then Wave 1 accounts will receive the patched release on Day 2.
• Thursday, 2pm - 4pm Pacific Time: Wave 2 release for all remaining accounts. This final release may be canceled if any regression is detected during the previous release waves. In the case that this release is canceled, then any previous releases during the week will be rolled back.

By default, all Tecton accounts are part of the Wave 2 release. Contact Tecton Support to opt-in your development/testing account to the Wave 1 release.

Versioned Materialization Runtime

To ensure stability in production, Tecton versions the runtime deployed to customers' materialization clusters.

In 0.8+, tecton_materialization_runtime is a required parameter for Batch Feature Views, Stream Feature Views, and Feature Tables with materialization enabled and running on Spark. Users can also use the Repo Config to set a default version for all Feature Views and Feature Tables in a Feature Repository.

Tecton provides +1 support -- a Feature View applied using Tecton 0.8 will be compatible with a tecton_materialization_runtime of 0.8 or 0.9. This allows Tecton users to first upgrade their materialization runtimes before upgrading their local SDK version.

Generally, users should set tecton_materialization_runtime to the most up-to-date compatible version of tecton. Users can update their Feature Views & Tables when upgrading their SDK version to a new minor version or when recommended by Tecton Support (e.g. when a version is released with a new feature or bugfix). Updates are also documented in the Changelog.

A Stream Feature View running in Spark with a pinned materialization runtime version will only restart when a user updates its tecton_materialization_runtime and runs tecton apply. This will be a zero-downtime restart since Tecton practices a blue-green release process (as described in the following section).

In 0.9+, environment is a required parameter for Batch and Stream Feature Views running on the Rift compute platform. For jobs running on the Rift, environment encapsulates the runtime Python environment. Please see the Environments in Rift page for more details.

See the Python dependencies for tecton here.

Zero-downtime releases

For Feature Views and Feature Tables without a pinned Tecton materialization runtime version (e.g. those on SDK <=0.7), Tecton practices a blue-green release process to provide zero-downtime weekly updates. For example, for Stream Feature Views with Spark, Tecton will verify that the updated stream process is online before stopping the existing process.