Cluster configuration & time out behaviour

[rwahyudi]

I'm trying to setup replication where each wordpress node has its own redis copy and have a master somewhere else.
predis this could be achieved with :
$parameters = ['tcp://master.example.com?role=master', 'tcp://localhost '];

How do we do this in milliCache ?

In regards to timeout behaviour - what happen to the page if redis is not accessible ? Will it stop user loading the page and erroring ?

[ouun]

Hi @rwahyudi, thanks for the questions!

If Redis is unavailable: MilliCache degrades gracefull. It logs the error and serves pages as uncached WordPress. Caching resumes when the connection recovers.

Replication/cluster: Not supported yet. We connect to a single endpoint today.
But can add it with the next release. We add features when there’s a real need rather than shipping config that sits unused, and your case is a good reason to build it.

Working option today: run KeyDB active-active and point each node at its local instance (MC_STORAGE_HOST=127.0.0.1). That gives you a local instance per node, replicated to its peer, with no client-side changes.

I’m happy to address client-side replication for v1.7.0 (next couple of weeks). I’ll follow up here when it lands.

[rwahyudi]

Thank you so much for the quick response and pointer to KeyDB!
Sad that KeyDB seems to be dead and no longer maintained.

In regards to failure behavior I think there should be timeout configuration and the default should be < 1 second.
The default predis timeout is 5 second which is way too long for caching system.

Looking at the diagram - unresponsive storage backend will cause a double whammy effect :

• Timeout for the 'Check Cache' which will produce a MISS
• Another timeout when trying to Store.

Ideally we should also have a flag for failed backend so that it can be avoided for the next X minutes.

Maybe something like :
rwahyudi@a13e91b ?

( vibe code result )

[ouun]

Agreed on both. v1.7.0 will ship a configurable timeout with a low default (~0.5s) plus a circuit breaker that trips on failure and skips the backend for a cooldown. Thanks for the sharp input!

[ouun]

@rwahyudi, could you please try v1.7.0-beta.4 with your setup? It adds Replication and Sentinel support. Please note that you will configure it using the MC_STORAGE_HOST constant in MilliCache, as this is an advanced feature for which only MilliCache Pro provides a UI. You can find the updated documentation on setting up a high availability connection here.

For the new constants, MC_STORAGE_TIMEOUT and MC_STORAGE_READ_TIMEOUT, please refer to this updated doc.

[rwahyudi]

Thanks for this ! Quick test seems fine - will do more troubleshooting over the weekend.

[ouun]

@rwahyudi I hope you're caching happily now! I just wanted to mention that MilliCache Pro has been released in the meantime. It offers full replication and Sentinel support for Redis-compatible backends, as well as a dedicated UI.