Question on Memory Usage, maxmemory and cache_mode settings

[kaitalledo]

Hi Team,

We have successfully switched from Redis to DragonflyDB (1.4.0) on our production web application after testing it in our pre-production environment for sometime. So far we are very happy with it but I just have one thing that we have noticed. Previously, with Redis, memory usage was around 25GB (with maxmemory setting of 64GB and key eviction set to "volatile-lru") after we run a cache priming script.

With dragonflydb, memory usage hovers around 1GB-1.5GB after priming. Having said this, I'm not entirely sure if it's actually caching stuff or the cache_mode setting is doing something with it. But as I understand it, based on the documentation, cache_mode should only kick in when the maxmemory is reached (also set to 64gb on dragonflydb).

Can someone help enlighten me? Thank you in advance! Config of dragonflydb is below:

--pidfile=/var/run/dragonfly/dragonfly.pid
--log_dir=/var/log/dragonfly
--dir=/var/lib/dragonfly
--version_check=true
--bind=localhost
--maxmemory=64g
--cache_mode=true
--dbnum=16
--dbfilename=dump.rdb
--df_snapshot_format=false

[chakaz]

It's awesome that you've switched to Dragonfly, thanks for letting us know!
The fact that Dragonfly uses 25x less memory than Redis tells me that something is off. Do you use it in the exact same way, with roughly the same load? Indeed, --cache_mode should not have an effect when so much below --maxmemory.
Can you please share the output of info? (issue it inside redis-cli)

Also, a side note: I see that you're binding to localhost, does that mean you're running other services on the same machine? The reason I'm asking is because if that's the case, you may want to limit the number of threads used by Dragonfly, because by default it will use all, which could degrade other services using the same machine.

[chakaz]

I see that the DB has 49468 keys (+1 in db4), with 0 expired. Also, net received bytes is ~90mb, aligned with the current perceived consumption of 73mb. So it looks like this is just the data.. can you tell how many keys and other stats on the previous Redis deployment?

Re/ limiting threads - you could use --proactor_threads=X to limit Dragonfly to X threads

[kaitalledo]

Unfortunately, we have already decommissioned the Redis instance, so I am not able to get the details you are asing.

As for limiting the threads, I have now set it to 250 and the memory usage jumped to 6.5GB.

Are there any other tips you can give me? :)

Cheers,

[chakaz]

I would actually suggest to lower that number:

• If you machine is dedicated to running Dragonfly, don't specify the number, which will have the effect of having 1 thread per hardware CPU. This is the ideal case.
• If you do run other services on that machine, calculate how many CPUs you need for them, and use the remainder for Dragonfly. For example, if your machine is running on a machine with 32 CPUs, and you need 10 for the other services, run dragonfly with --proactor_threads=22 to let Dragonfly utilize only 22 CPUs

Running Dragonfly with more threads than physical CPUs will just reduce efficiency

[chakaz]

Oh, and the additional memory usage that you've witnessed is likely not due to more values stored in Dragonfly, but just the per-thread overhead x 250..

[kaitalledo]

All good now. Thank you so much!