DragonflyDB vs Redis in production: what issues did you discover after switching?

[bensky111]

We’re considering migrating some Redis workloads to DragonflyDB because of the performance and lower memory usage claims.

For teams already using Dragonfly in production:

What unexpected limitations or compatibility issues did you hit?
Any problems with replication, persistence, or failover?
How stable is it under heavy concurrent workloads?
Did you need to change application patterns compared to Redis?

Would appreciate real production feedback before we test it at scale.

[alpha819beta]

We migrated a high-throughput caching service from Redis to DragonflyDB about 4 months ago.

Positives:

Much better memory efficiency compared to Redis
Lower CPU usage during peak traffic
Very good performance for cache-heavy workloads
Easy migration because most Redis commands worked without changes

Challenges we faced:

A few Redis modules/features we depended on were not fully compatible
Replication monitoring/documentation could be clearer
Some latency spikes appeared during snapshot persistence under very heavy write load
Fewer community examples compared to Redis, so troubleshooting took longer

Overall, for pure caching workloads DragonflyDB has been excellent, but I’d still carefully test edge-case Redis compatibility before a full production migration.

[romange]

@alpha819beta Thanks for the feedback!

Could you provide us with specific example where you think Replication monitoring/documentation could be clearer?
Also what kind of troubleshooting did you have to do?

[gheorghitamutu]

Hello,

Unfortunately.. this one: #7639.
I've explained the situation in the Bug Report as I'm not sure we should be tackling so much the allocator side. We never encountered this issue with Redis before. It can definitely might be some wrong configuration on my side, maybe, but it's not something very easily debuggable.
So any advices are welcomed on how to reduce/control RSS size or maybe to entire disable the custom allocator.

Other than this.. anything else just works. Nothing that we'd have been constrained by wasn't a deal breaker.

[gheorghitamutu]

I've actually dug deeper into this and found the real reason being some replication interaction between multiple instances.
Explained here: #7639 (comment).

[gheorghitamutu]

This has been solved and it has been a wrong assumption.
Thank you!

[kostasrim]

@gheorghitamutu don't worry about it, a system's configuration can sometimes be confusing 😄 have a good day!