Migrating Millions of Redis Keys Without Downtime

Migrating a Redis dataset sounds simple until the key count crosses a few million and the app can't go down. Here's how I approached it.

The problem

We needed to move millions of keys between Redis deployments without a maintenance window. A naive DUMP/RESTORE loop would've taken hours and hammered the source.

Using RIOT

RIOT handles the heavy lifting — it streams keys with backpressure instead of loading everything into memory.

riot replicate redis://old:6379 redis://new:6379 \
  --mode live \
  --read-batch 500 \
  --threads 4

The key choice is --mode live: RIOT takes an initial snapshot and tails keyspace notifications, so writes that land mid-migration still make it across. That's what lets you cut over with no downtime — a plain snapshot scan would miss everything written after it started.

A few things that mattered:

• --mode live over a one-shot snapshot — it keeps source and target in sync until you flip traffic.
• Tune --threads and --read-batch to push throughput without overwhelming the source.
• Verify with a checksum pass after, not just a key count.

Takeaways

• Live migrations are mostly about controlling throughput, not raw speed.
• Always reconcile after — a key count match isn't a value match.
• Keep a rollback path until you've cut traffic over for real.