Research of Ruby GRPC-implementations

We have been using GRUF framework to run GRPC-server in Ruby. But we faced a lack of query-backlog. When GRPC-server is overloaded and has no capacity to process request right now, it rejects a request end return ResourseExhausted error.

Here is a research on how to fix it. There are several ways:

• use revese-proxy with query-backlog to park requests until they could be processed
• move back query-backlog to ruby-grpc library (by reverting very old PR)
• migrate to griffin framework as alternative with support of backlog and multiprocessing

How it works

How to run stress-tests

Provisioning

Install all the dependencies and run all components:

• client - Rack-based HTTP-server serving /hello/gprc, /hello/gruf, /hello/nginx and /hello/griffin endpoints. Makes GRPC-request to hello-world implementation of GRPC-servers
• prometheus - a Prometheus instance configured to collect metrics from client
• grafana - a Grafana UI for visualize metrics, http://localhost:3000
• gruf - basic implementation based on GRUF framework
• grpc - grpc-ruby gem based implementation with moved back request backlog patch
• griffin - GRPC-server based on griffin gem
• nginx - NGINX GRPC reverse proxy with support of query backlog

To setup environment run:

# provision
dip provision

# get services up and running
dip compose up client grafana prometheus gruf griffin nginx grpc

Check everything is ok

# check all GRPC-servers are up and running
dip grpcurl gruf
dip grpcurl griffin
dip grpcurl nginx
dip grpcurl grpc

# check if http server started and is ready to use underlying GRPC
dip compose exec client curl http://localhost:9292/hello/gruf
dip compose exec client curl http://localhost:9292/hello/griffin
dip compose exec client curl http://localhost:9292/hello/grpc
dip compose exec client curl http://localhost:9292/hello/nginx

# check if metrics port is exposed
dip compose exec client curl http://localhost:9394/metrics

Runs stress-tests

# runs k6 stress test for any GRPC-implementation
dip k6 gruf_http
dip k6 griffin_http
dip k6 grpc_http
dip k6 nginx_http

Results

Navigate to http://localhost:3000/ admin/password to the "GRPC endpoint" dashboard