adriangligor/octoplex

HTTP request multiplexer

Octoplex

HTTP request multiplexer

Octoplex is a service that accepts a request containing a batch of HTTP requests, executes them all in parallel and returns all results. It handles errors, DNS resolver caching, TCP connection pooling and HTTP keep-alive and enforces deadlines/timeouts.

The main design goal is to be as efficient as possible by having an async architecture, using all CPU cores, locking as little as possible and avoid unnecessary memory copies.

Get started

Just follow these simple steps:

1. Clone the repository, for example git clone https://github.com/adriangligor/octoplex.git
2. Start the Octoplex service via cargo run --release
3. Make a request

curl --location --request POST 'http://localhost:8080/multiplex' \
     --header 'Content-Type: application/json' \
     --data-raw '{
       "timeout_msec": 250,
       "requests": [
         {
           "uri": "https://www.google.com/",
           "headers": {
             "X-Some-Header": "some value"
           }
         },
         {
           "method": "POST",
           "uri": "https://reqres.in/api/users",
           "headers": {
             "Content-Type": "application/json"
           },
           "body": "{\"name\":\"Bruce Campbell\",\"job\":\"Fake Shemp\"}"
         }
       ]
     }'

Docker support

For building a release image containing the Octoplex service, run: docker build -t adriangligor/octoplex:latest .

For starting Octoplex, run: docker run -p 8080:8080 adriangligor/octoplex. Now requests can be sent to http://localhost:8080/multiplex.

Integration tests

Tests can be run as usual via cargo test. This includes integration tests, which require (and check for) mock services like WireMock (used as a target service for Octoplex) to be brought up and configured upfront. For this reason, a Docker Compose setup is provided as well (see below). When not using Docker Compose, follow these steps:

1. Start WireMock using the content of extra/wiremock as its root directory (see --root-dir in the WireMock documentation). The provided Docker Compose support can also be used to start WireMock only.
2. Start the Octoplex service via cargo run
3. Run the tests via cargo test

Integration tests are locating the Octoplex and WireMock services by using the following environment variables:

• TARGET_HOST (default: "localhost"), the host running the Octoplex service
• TARGET_PORT (default: "8080"), the port the Octoplex service is listening on
• WIREMOCK_HOST (default: "localhost"), the host running WireMock
• WIREMOCK_PORT (default: "18080"), the port the WireMock service is listening on
• WIREMOCK_PORT_SSL (default: "18443"), the SSL port the WireMock service is listening on

Docker Compose support

In order to simplify the development of integration tests, a Docker Compose setup is provided. By running docker-compose up -d octoplex the service starts up via "cargo run" using the default profile, with source mapped into the container. When the source changes, a restart is enough to pick up the change. A WireMock container is also started.
By running docker-compose run tests, the test suite is ran against the Octoplex container using WireMock.

Roadmap items

• Introduce CLI arguments
• Introduce logging of key information (config, incoming requests, deadline violations, ...)
• Support logging to a log file
• Implement more unit and integration tests
• Support detaching as daemon
• Ensure DNS resolving and connection establishment in the background even after reaching batch timeout
• Ensure async DNS resolving, caching, respecting TTL and distributing evenly across multiple resolved IPs
• Ensure proper caching of TCP connections to resolved IPs, max request count or TTL per connection, growing and shrinking of connection pool
• Ensure detection and cleanup of stale connections
• Create a good Dockerfile
• Support gRPC (or similar, with zero-copy capabilities) frontend in addition to HTTP frontend (split off frontends into modules or workspace crates)
• Collect statistics and make the available to clients
• Support TLS both on all frontends and outgoing
• Support outgoing gzip compression
• Ensure minimal overhead over actual outgoing requests
• Minimize incoming request size (e.g. when batched requests are very similar, like in OpenRTB auctions)
• Introduce config file
• Profile CPU and memory usage, allocations, etc., then optimise