Hi all, I jsut finishing watching interesting show case video:
on 3 uses cases:
- Indigo Veronika use case.
- Tokiomarine hcc use case .
- Argenta bank use case.
Its on the 3rd use case on Argenta bank that by the minute 42 and 46 seconds the presenter mentions “snapshots” implementation, which that made me wondering about implementation details. As presenter says, for past snapshots asked on demand, they just load PITs and BRIDGEs tables - that I think serve star model views Presentation Layers.
But I wonder, bit more details:
- Do they load snapshots in independent new set of tables ? like dim_customer_<snapshot_date>, bridge_<snapshot_date>, pit_<snapshot_date> ?
- Or using snowflake Timetravel feature where we just postfix on SELECTS … at <snapshot_date> on the views ?
Would like to hear from the group ideas, either specifically the people form vaultspeed if they hear this post from me either anyone in the group could share your experience implementing snapshots, in how physically you doing ?
truncate/load new set of tables ?
and/or using some middle layer to help filter data as-it-was ?
I can think in snowflake we can create timetravel to both:
- zerocopy clone - to create tables as of some timestamp - but this would consume space and would require postfix something like _<snapshot_timestamp> to avoid naming conflicts with table/views names.
- SELECT … AS of <snapshot_timestamp> - this avoids duplicating data in new set of tables, seems good idea at least while using Snowflake platform right ?
So basically its that, I wonder:
- do you make copy of snaphot data ?
- how access layer works from the perspective of enduser ? do they need to use different set of dim/fact views ?
Thanks in advance ~