Configuration as a service

In my organization we have a micro-services architecture. Up until now, it was fairly easy to manage and work with as the configurations are rather limited, and often independent of specific tenant (we have multi-tenant system).

But recently the number of tenant specific configuration has steadily increased and we need a central service to manage all the configurations specific to each client per service.

For example, a client can have a configuration for minimum number of tokens after which classification triggers. It might be 2 for tenant A or 4+ for tenant B.

What I am looking for is if there is already an available system, which people use in production, capable enough to handle generic types of configuration? The configurations can be complex JSON structures, YAML files or something similar (note: we are willing to write the layer for transformation), java properties files, ini files etc.

We have already considered zookeeper, etcd, consul etc. as an agent and a wrapper on top of it, but we might have to store some data as well, which would cause performance issue, and overall it is not useful in our case.

Hello After 4+ years could you please share your findings and opinions?

We are also looking for a solution for "Configuration as a service" for several microservices. Very new to this topic. What are the most modern solutions?

We have built something in-house for managing tenant configurations very similar to what you have described. It does layering and overrides, and there's a client that attaches to applications and picks up config changes dynamically. So no restarts are required when you change configurations.

I've been looking for out of the box solutions for the same and found Spring Cloud Config. Check it out. cloud.spring.io/spring-cloud-config

I think we too will have to go the same route. Can spring cloud store complex JSON or just simple KV.?

I will explore this further. Looks like a decent way to expose configuration to consumers.

Check out Tweek. We use it for feature flags/AB tests but it can be used for the scenario you are describing.

Ujjwal Kanth, did you get this to work?

Siddarthan Sarumathi Pandian Took me a while, but yeah, I got it to work. I basically wrote a custom microservice on top of a graph store dgraph.io, with large bulk store as s3.

Thanks Ujjwal Kanth :) Shall take a look.

I haven't tested everything so this is a biased answer, but I used Consul (and am slowly introducing it in my new company) to centralize and manage configs and it's always a pleasure to work with this. (In a Kubernetes world, I used Helm to inject config from templates and that was also really nice, but different flows).

I don't remember the max size for records of the Consul key value store, but it can store pretty much any config (I used it mostly for Json, ini and NGINX templates, which are just text files).

I think the limit is <2mb. We could have a blend of consul + some data backend to store large data files. But I am sceptical as we have to actually understand the config that we are going to store.

The thing is, we have a usecase where we allow a hook or a watcher on a particular entry in a configuration, which makes storing configuration as bytes (json or ini stored as simple bytes in consul) not feasible.

Although, thank you so much for giving a plus point to add in an internal discussion on zookeeper vs etcd vs consul.

The mention of hooks brings another feature from Consul: Consul events and watches, which lets you send custom events (basically anything, just a string, with some filtering data if you want to restrict the audience), which would be received by whatever process which registered for it using Consul watch APIs. It's really a black-box messaging leveraging Consul gossips for your own custom use cases (btw, Consul also work across region and datacenter, if you have any hybrid or multi-region setup). I never had much use for it, (which is why I forgot about it until I read your mention for watchers)

Spring Cloud Config could be a good fit.

Doesn't completely fit our use case, however looks interesting. We can use this as a library in the consumer of this configuration service. Thanks.

I am not 100% sure, but I have heard good things about Terraform. Do you think it will fit your needs?

AFAIK terraform is for "infra as code". We use terraform to replicate the environment from one service provider to another.

Example we have a setup definition of 150 machines which run our query stack. This incurs quite a bill on aws :)

We wrote a terraform script to move this entire setup to GCM (still testing) but essentially this is what terraform does. again AFAIK :)

Also, the configurations are dynamic in nature. It isn't just properties file or some yaml file which gets loaded. It can be parameters which determine a certain service will be added to mesh flow or not.

Ah, my bad. :) I am used to setting up servers for machines (the tenants running the microservices) using Terraform. I thought that's what you meant by config. Also, I overlooked the fact that the config file could change.

Thread

Configuration as a service

Responses(6)

Recent threads

Search Hashnode

Configuration as a service

Responses(6)

Recent threads