The HybridOnlineStore allows routing online feature operations to different online store backends based on a configurable tag (such as tribe, team, or project) on the FeatureView. This enables a single Feast deployment to support multiple online store backends, each configured independently and selected dynamically at runtime.
To use the HybridOnlineStore, install Feast with all required online store dependencies (e.g., Bigtable, Cassandra, etc.) for the stores you plan to use. For example:
pip install 'feast[gcp,cassandra]'
{% code title="feature_store.yaml" %}
project: my_feature_repo
registry: data/registry.db
provider: local
online_store:
type: hybrid
routing_tag: team # or any tag name you want to use in FeatureView's for routing
online_stores:
- type: bigtable
conf:
project_id: my_gcp_project
instance: my_bigtable_instance
- type: cassandra
conf:
hosts:
- cassandra1.example.com
- cassandra2.example.com
keyspace: feast_keyspace
username: feast_user
password: feast_password{% endcode %}
To enable routing, add a tag to your FeatureView that matches the routing_tag specified in your feature_store.yaml. For example, if your routing_tag is team, add a team tag to your FeatureView:
tags:
team: bigtable # This tag determines which online store is usedThe value of this tag (e.g., bigtable) should match the type or identifier of the online store you want to use for this FeatureView. The HybridOnlineStore will route all online operations for this FeatureView to the corresponding backend.
{% code title="feature_view" %}
name: user_features
entities:
- name: user_id
join_keys: ["user_id"]
ttl: null
schema:
- name: age
dtype: int64
- name: country
dtype: string
online: true
source:
path: data/user_features.parquet
event_timestamp_column: event_timestamp
created_timestamp_column: created_timestamp
tags:
team: bigtable # This tag determines which online store is used{% endcode %}
The team tag in the FeatureView's tags field determines which online store backend is used for this FeatureView. In this example, all online operations for user_features will be routed to the Bigtable online store, as specified by the tag value and the routing_tag in your feature_store.yaml.
The HybridOnlineStore will route requests to the correct online store based on the value of the tag specified by routing_tag.
The full set of configuration options for each online store is available in their respective documentation:
For a full explanation of configuration options, please refer to the documentation for each online store backend you configure in the online_stores list.
Storage specifications can be found at docs/specs/online_store_format.md.
The set of functionality supported by online stores is described in detail here. Below is a matrix indicating which functionality is supported by the HybridOnlineStore.
| HybridOnlineStore | |
|---|---|
| write feature values to the online store | yes |
| read feature values from the online store | yes |
| update infrastructure (e.g. tables) in the online store | yes |
| teardown infrastructure (e.g. tables) in the online store | yes |
| generate a plan of infrastructure changes | no |
| support for on-demand transforms | yes |
| readable by Python SDK | yes |
| readable by Java | no |
| readable by Go | no |
| support for entityless feature views | yes |
| support for concurrent writing to the same key | yes |
| support for ttl (time to live) at retrieval | no |
| support for deleting expired data | no |
| collocated by feature view | yes |
| collocated by feature service | no |
| collocated by entity key | yes |
To compare this set of functionality against other online stores, please see the full functionality matrix.