Use prevent_destroy
to Protect Production Pipelines and Services
Tecton recreates repo objects when certain fields are changed. For example, changing the SQL transformation in a Feature View would by default cause Tecton to recreate that Feature View. Recreates are destructive, so for Feature Views with materialized tables, Tecton will rematerialize those tables and clean up the old tables. During this time, those Feature Views will not be available for online serving.
See Making Changes to Features for a guide on how to safely make changes to Tecton objects and avoid destructive changes.
If you would like a Tecton object to never be accidentally recreated or
destroyed (i.e. deleted), you can set the prevent_destroy
parameter on that
object. This will prevent that Tecton object from being recreated or destroyed.
Also, since destructive changes cascade to dependent objects, setting
prevent_destroy
will block destructive changes on all inputs. For example,
setting prevent_destroy
on a Feature Service will block destructive changes on
all Feature Views that that service depends on.
- Spark
- Rift
@batch_feature_view(
sources=[FilteredSource(transactions_batch)],
entities=[user],
mode="spark_sql",
online=True,
offline=True,
feature_start_time=datetime(2021, 5, 20),
batch_schedule=timedelta(days=1),
prevent_destroy=True,
)
def critical_feature_view(transactions):
pass
critical_feature_service = FeatureService(
name="my_critical_service",
features=[critical_feature_view],
prevent_destroy=True,
)
@batch_feature_view(
sources=[FilteredSource(transactions_batch)],
entities=[user],
mode="pandas",
online=True,
offline=True,
feature_start_time=datetime(2021, 5, 20),
batch_schedule=timedelta(days=1),
prevent_destroy=True,
)
def critical_feature_view(transactions):
pass
critical_feature_service = FeatureService(
name="my_critical_service",
features=[critical_feature_view],
prevent_destroy=True,
)
prevent_destroy
has no effect in development workspaces, as there are no
assets or processes to protect.
Typical Uses​
prevent_destroy
is most commonly used on Feature Views and Feature Services
for the following reasons:
- Prevent serving downtime by setting
prevent_destroy
on production Feature Services.- Feature Services with
prevent_destroy
set will prevent recreates of any Feature Views used in the Feature Service.
- Feature Services with
- Avoid accidentally rematerializing large Feature Views, even if they're not
used in production.
- If a Feature View is expensive to materialize (e.g. it transforms years of
data) consider enabling
prevent_destroy
for that Feature View even if it's not used in production applications.
- If a Feature View is expensive to materialize (e.g. it transforms years of
data) consider enabling
- Prevent schema changes to Feature Services or Feature Views.
- Schema changes to Feature Services and Views are always destructive. If a
downstream application depends on the exact schema of Feature Service or
View, then
prevent_destroy
may be used to ensure that the schema does not change accidentally.
- Schema changes to Feature Services and Views are always destructive. If a
downstream application depends on the exact schema of Feature Service or
View, then
Deleting Protected Objects​
If you want to delete a protected object, you must first set prevent_destroy
to False, apply that plan, and then apply a second plan to remove the object.