How to use large dynamic data in wasm proxy rust filter?

[leandrocurioso]

Title: How to use large dynamic data in wasm proxy rust filter?

Description:

I have develop several filters with wasm proxy using rust and now there is a requirement I'm not sure how to tackle.

Scenario:

I have a large file with a list of clients more than 1000000+ registers and I need to route to specific endpoints based on a header request header. Ex: X-Client-Id: 123456

Have have some options:

1. Create a route for each one of them: (Bad idea since the routes are evaluated top/down this would increase a lot my downstream time)
2. Create a wasm proxy filter and inject this as a configuration sent by the go-control-plane (xds delta snapshot grpc) and build a hash map inside the filter configuration. (Tried this but the snapshot request gets too big and result in a timeout, and the envoy memory increases a lot)
3. Do not send this data by xds delta but load that file content on the on_configure hook of the wasm proxy filter. (Tried that but the sandbox rust vm doesn't allow io:filesystem operations)

I also noticed that if I keep sending snapshots to envoy they keep increasing the memory, I looked into the os process and if the envoy container keeps receiving snapshots it puts all in memory without any recycling.

IMPORTANT POINT:

Envoy will run in kubernetes, when the pod starts up it should have this data ready to be used and ocasionally new updates in this list will happen and I need to update the "cache".

Is there a way anyone can recommend me to go? I 'm kinda lost at this point.

Thanks in advance for your help and if you need more details please comment! =D

[yanavlasov]

You can try using universal matcher https://www.envoyproxy.io/docs/envoy/latest/api-v3/config/route/v3/route_components.proto for matching routes. It supports MatcherTree with the exact or prefix matchers that are very efficient.

[leandrocurioso]

@yanavlasov

Do you know if there is a limit of the tree nodes?

[agrawroh]

Is it 1:1 mapping for route to endpoint?

[leandrocurioso]

Is it 1:1 mapping for route to endpoint?

@agrawroh

No.

The routing will happen when I check for a header key: "X-Client-Id: 123", then I match against a key paired list or hashmap. The key is the client id and if the match successfully happens then go to a specific cluster endpoint.

So basically and endpoint can serve a list/hashmap of clients.

[yanavlasov]

There is no limit, other than resources available to Envoy process.