bendnorman
commented
2 years ago Proposal
Have two BigQuery instances, dbcp-dev and dbcp-prod. Only a service account should have permission to load data to these instances. All Catalyst users will have permission to read from these Big Query instances so they can authenticate using their own accounts when creating new data sources in Tableau.
The dbcp-prod and dbcp-dev data will correspond with the outputs of the ETL on our main and dev branches. We will not be pushing data from other branches. Code should be merged into dev before dashboarding in Tableau.
In theory, we will have stable dashboards that are being shown to clients. These dashboards should be linked to dbcp-prod. These dashboards can live in a Production project in tableau. When a production dashboard needs to be updated or tested on dbcp-data, it can be copied into the Development project, changed, tested and then copied back to Production.
All draft dashboards can live in a Development project in Tableau and will use dbcp-dev as a datasource. When a draft dashboard is ready for production, it can be copied over to the Production project in tableau.
Questions
- Am I overcomplicating this? My main objective is to have stable dashboards we know won't break if we update the data.
- @aesharpe what is your desired Tableau versioning workflow?
Other information
- You can restore previous dashboard versions. Unfortunately "Revisions are saved for .TDE files or hyper files with a direct connection. Revisions are not saved for extracts." This feature would allow us to revert to previous dashboard versions but does not retain the previous extract.
- We could use BigQuery tags and labels to version our data.
- We can also use table snapshots to track changes to our data.
- If folks need to prototype a tableau dashboard with non dev or prod data, they can connect tableau to their local postgres instance.
TrentonBush
Our rapid changes to the ETL and dashboards highlighted the need for a more standard development and production cycle for this project.
Currently, each dashboard is pointing at one BigQuery instance with a live connection. Ideally, data is only loaded to BQ from our main branch but it seems like folks want to make smaller changes to the data as they are dashboarding.
I'm concerned we will have a couple of dashboards working well / in a production state, then break them all by updating the BQ data from an untested branch.
What are best practices for dev/prod with Tableau? We have 3 things we need to version control: our code, data and dashboards.