gone

gone

A simple leadership election service for distributed systems.
This solves the problem of distributed schedulers executing at the same time, when only one execution is desired.

This is an HTTP wrapper over redis's global locks, with HTTP GET requests for ease of use.
gone itself is safe to be scaled up and still return a single leader, thanks to the guarantees of redsync.

Installation

Spin up a redis instance, listening at localhost:6379

docker run -p 6379:6379 --name rds -d redis

Run the code, the HTTP server runs on port 8080 by default.

go run main.go

Usage

No user set up required, keys are assumed to never collide.

1. Choose a system identifier. This self-chosen string is reused across your distributed service: e.g. if there are 5 redundant copies of a "cleaning scheduler" service, then this would be something like cleaning_scheduler.
2. Choose a round identifier. This self-chosen string is ephemeral, but must match across your distributed service: e.g. if a tmpfs cleaner runs hourly, this would be something like 2020-10-18-0800-tmpfscleaner. All copies of the redundant service must generate the same value, so coarse timestamps are recommended.
3. Whoever receives the 202 Accepted HTTP response wins! Losers receive 204 No Content.
4. Optional: upon completion, the winner sends a POST to the complete endpoint to indicate successful execution.

Am I the leader?

HTTP GET to /api/v1/elect/{system_identifier}/{round_identifier}

• Returns a 202 Accepted if the requester is the elected leader
• Returns a 204 No Content if the requester is not the elected leader. These requests will take longer (approx. 1.5s) due to mutex timeouts.
• In case of 500 Internal Server Error, an error occurred when recording or verifying leadership. It is safe to retry.

Who was the leader?

HTTP GET to /api/v1/elected/{system_identifier}/{round_identifier}

• Returns a 200 OK alongside the Request.RemoteAddr of the elected leader
• If no leader was elected (e.g. the round hasn't occurred yet), 204 No Content is returned.
• In case of 500 Internal Server Error, a redis communication error has occurred. It is safe to retry.

optional: Did I do my job?

HTTP GET /api/v1/complete/{system_identifier}/{round_identifier}

• Returns a 200 OK upon successful recording
• Returns a 400 Bad Request if another leader(!!!) has already recorded their completion. If this is encountered, the calling system is in a bad state.
• In case of 500 Internal Server Error, an error occurred when recording or verifying the completion. It is safe to retry.

optional: Did anyone do the job?

HTTP GET /api/v1/completed/{system_identifier}/{round_identifier}

• Returns a 204 No Content if no completion found for the given round
• Returns a 200 OK alongside the Request.RemoteAddr of the completer
• In case of 500 Internal Server Error, a redis communication error has occurred. It is safe to retry.