You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
model the metadata separate from the checks. moving check configurations from the central metadata.
Goal is to create a separation for the metadata that should go to data discovery from the unit testing responsibility.
The validity configurations in contracts now reside in the checks
There is the alternative to separate the meta data from the check configuration. For example:
dataset: dim_customer
columns:
- name: id
data_type: character varying
duplicates: not allowed
missing: not allowed
valid_regex: '^(?:[A-Z])$'
checks:
- type: no_missing_values
- type: no_invalid_values
The above contains the metadata:
duplicates: not allowed
missing: not allowed
valid_regex: '^(?:[A-Z])$'
And this metadata is in this alternative separated from the checks that configure Soda.
This way, metadata can be pushed to a catalog without it being tied to checks. So you can indicate that a column is unique without actually checking for it. Those are 2 separate concerns.
In similar style, we could also consider the schema check to be configured as a check. Eg
But in the schema case, currently I think the default should be that the schema is checked. I think many users will forget to include the check and assume it's checked. And then later be surprised if the check is not performed.
Improve variable declarations for better tool support eg
There are 2 potential contract YAML syntax updates to consider:
Goal is to create a separation for the metadata that should go to data discovery from the unit testing responsibility.
The validity configurations in contracts now reside in the checks
There is the alternative to separate the meta data from the check configuration. For example:
The above contains the metadata:
And this metadata is in this alternative separated from the checks that configure Soda.
This way, metadata can be pushed to a catalog without it being tied to checks. So you can indicate that a column is unique without actually checking for it. Those are 2 separate concerns.
In similar style, we could also consider the schema check to be configured as a check. Eg
But in the schema case, currently I think the default should be that the schema is checked. I think many users will forget to include the check and assume it's checked. And then later be surprised if the check is not performed.
The text was updated successfully, but these errors were encountered: