[question]: Ping message published to redis vs. direct SSE

[iankirk99]

Hi, I just started using this in production and have noticed a large number of publish commands being used even when not actively sending messages to clients (for my use case, I have a lot of concurrent users connected, but relatively few messages sent to each user).

It seems like this is mostly coming from the ping messages, which in my understanding gets sent to Redis every 10 seconds, then re-processed in the same handler in the onMessage function and forwarded to the client. I'm assuming that this is done to keep the client HTTP connection alive, but this creates an overhead of 6 publish commands per minute per active connection.

I'm wondering if this could be simplified by just sending ping messages directly to the client and avoiding Redis entirely. This should decrease idle load on Redis and the server (assuming there aren't other side effects that I'm missing here).

Happy to make the PR for this, but thought I'd check to make sure I'm not missing anything here first.

[joschan21]

Hi! This is mainly to prevent a timeout error that is automatically thrown from Redis-side after exactly 5 mins of inactivity (not sending any message via publish). The ping works very well to prevent this, what do you think about making the interval adjustable or letting the user disable entirely if wanted?

Testing the client connection and redis connection are kinda two different lifecycles

Appreciate the issue, lmk what you think of this proposal! Happy to make this work either way

[pacorroX]

Making the interval adjustable would be incredible! That’s literally why I came here, to see if it was possible.

It seems a bit unnecessary and not very cost-efficient for people using Vercel (even with Fluid!). If there were a way to adjust the timing, that would be perfect.

[iankirk99]

Got it, makes sense about the Redis timeout. An adjustable interval would be great.

One concern might be browser connection timeouts, since this would likely happen before the 5 minute Redis timeout. In this case, it might make sense to disconnect the Redis pings from the client pings.

For example:

• Every 10 seconds, send a ping directly from the handler to the client (without using publish)
• Every 2-4 minutes, publish a ping to Redis (this doesn't need to be processed in onMessage or forwarded to clients)

Either of these values could be configurable, but I could also see an argument for not exposing them to avoid overcomplicating the config options.

[joschan21]

cool let's make both configurable! i'm pretty busy with travelling this week, if it's urgent for one of you guys i'm happy to review PRs from the airport

otherwise i can also implement this when back in Germany :]

[iankirk99]

Just opened a PR for this! #9