Discussion on: How we use "ship small" to rapidly build new features at GitHub

[Mike Coutermarsh]

That really depends on the migration + feature.

We can never take downtime at GitHub. So when making any DB change, it has to work with both the old and new code.

A recent example, we just had a feature that required us to move data to a new table. The process was like this.

• Create the new table (no code changes)
• Change writes to both the old and new table
• Update the new feature to read from the new table (feature is behind a flag, so OK if data is incomplete)
• Run backfill, transferring old data to the new table (slowly over a week)
• Change old feature to read from the new table