Production use (or rather validation for adopting in production architecture)

[drmikecrowe]

I'm liking everything I see about Tile38, and it looks like it will solve a technical issue I'd have to implement (roaming geofences). Is there a list available of existing sites using Tile38?

I'm mainly seeking data to validate this choice:

• Who is using it in production?
• What level of traffic are they handling?
• What don't I know yet that I need to know before depending on Tile38 in production? (assuming I have no issues implementing Replication, failover, load-balance the reads, etc)

[dinapappor]

At work we're using tile38 with a master and 4 followers.

We're having about ~20 SETs at peak and about 1800 nearby at peak. We're running these on t2.mediums on aws and we're barely seeing any cpu usage at all. It's been 99-100% cpu idleness looking at cpu graphs for those instances. This goes for both master/leader and the followers.

We're doing loadbalancing by putting the followers behind an ELB that our API call. We're using redis interface and the amount of connections are fairly well spread over the followers.

Doing failover is not something I have done yet. I have found no viable option for this yet. :(

[tobilg]

@dinapappor Thanks for the insights! Do you happen to also use the hooks functionalities, and if so, in what setting? Tile38 directly doing http(s) requests, or some layer in between?

[tobilg]

@tidwall I'd be great to eventually also have the production recommendations either in the readme or the tile38.com website. Thanks!

[tidwall]

@tobilg I agree. I'd be happy to explore ideas around how a recommendations page might best be organized. I'm often in the weeds with the codebase and I'm sure I take for granted the challenges with deploying in production.

Perhaps be new topic page on the tile38.com site?

[dinapappor]

@tobilg No, I do not use any hooks, communication is purely done by means of the redis interface.

Changing back somewhat to the original topic: I am curious as how to achieve redundancy and failovering with tile38. If anyone could give insights on that I would very much appreciate it.

[tobilg]

@tidwall I think it would be great to have a dedicated page on tile38.com covering production deployments and operations, as well as some "recipes" on how to potentially scale out etc.

[KrishnaPG]

Came here while looking for tips on how to organize the collections for optimal performance in production. For example, if an object belongs to multiple categories (e.g. vehicle, and coldstorage categories), should

• these two categories be made into two collections and enter the object into those two collections separately, or
• just maintain one collection and set the categories as array field on one object?

Which one is more optimal for queries (e.g. select all vehicles in this radius, or select all cold-storage vehicles in my region etc.)

[trendstate]

Would be keen to see how this production deployments work. 😊 I'm working through some sensitivity testing myself to work out how to make it as fast as possible -but I'm sure I'm barely scratching the surface Thanks