• Be pragmatic. We advise our clients not to aim at implementing the “perfect” data mesh but to be guided by addressing their specific pain points and objectives. For example, polyglot storage and multi-modal access are useful concepts, but companies should focus on their actual requirements to maximize impact. (View Highlight)
Producers can also publish and share external tables, which are “views” over files stored outside of Snowflake, and which can optionally include Delta Lake and Iceberg formats (View Highlight)
users can easily instantiate and scale their own compute clusters without support from an IT infrastructure team. Cloning dev and test environments is equally straightforward. A change data capture mechanism can be set up with a 1-line SQL DDL statement. (View Highlight)
Some governance is decided centrally and is applied to all databases with a DevOps process. This can be facilitated by features such as object tags to keep an easy overview of the different objects owned by the domains (View Highlight)
Naming conventions should be planned carefully, as there can be a lot of objects, considering that every domain might require DT(A)P (Development, Test, Acceptance, Production) environments, which they can easily create with Zero-Copy Cloning. (View Highlight)
Every domain can have multiple schemas, with one serving as a layer to make products available to other domains. Another approach would be using a common share database where every domain will have a schema to publish its data products as views (no copies). (View Highlight)
