Skip to content

README.md

Latest commit

 

History

History
61 lines (41 loc) · 3.19 KB

README.md

File metadata and controls

61 lines (41 loc) · 3.19 KB

Warning

This repository is still WIP, things are subject to change, you SHOULD NOT use it in production. Any contributions are welcome

Replay Storage

Replay files are uploaded to an HTTP-accessible blob store. The server is fully pluggable — anything that speaks the contract below works (S3 with a presigned-URL fronting service, MinIO, a tiny custom server, etc.).

A reference local server lives at tools/ReplayStorageServer; useful for development and as a minimal implementation reference.

Configuration

In sharp/configs/timer.jsonc:

"replay": {
  "storage_base_url": "https://replays.example.com",
  "upload_non_personal_best": false
}
  • Omit the replay section (or leave storage_base_url empty) to disable remote uploading entirely. Timer.RequestManager will skip registering IReplayProvider, and no upload attempts will be made.
  • upload_non_personal_best — false (default) uploads only new PB/WR replays. true uploads every finished run.

HTTP contract

The timer drives three operations against your endpoint. The {key} segment contains forward slashes — your routing must treat it as a path, not a single segment.

Method URL Request body Success Notes
PUT {base_url}/{key} raw bytes any 2xx Body is a compressed replay file. No Content-Type is set.
GET {url} — 200 + bytes {url} is whatever the PUT returned/wrote — see below.
DELETE {url} — 2xx or 404 Best-effort cleanup after a failed DB write. Errors are swallowed.

The full URL stored in the DB and used for GET/DELETE is whatever IReplayStorage.UploadAsync returns. The bundled HttpReplayStorage simply returns {base_url}/{key}, so PUT and later GET hit the same path. A custom backend (e.g. S3 + CloudFront) can return a different download URL — it just has to be reachable for GET later.

Key layout

{map}/style_{style}/{track}/{steamId}_{runId}.replay                    main replay
{map}/style_{style}/{track}/stage_{stage}/{steamId}_{runId}.replay      stage replay

{map} is always lowercase. {steamId} is the 64-bit SteamID. {runId} is the run's primary key from surf_runs.

Failure handling

  • Upload fails → the run still exists in the DB without a replay row; logged as Failed to upload replay remotely.
  • Upload succeeds but DB write fails → the timer attempts a DELETE on the just-uploaded URL to avoid orphans. Don't treat a single DELETE as authoritative — periodic reconciliation against surf_runs_replay.Replay is recommended for long-lived deployments.

Running the reference server

cd tools/ReplayStorageServer
dotnet run

Listens on http://0.0.0.0:5080, stores files under ./replays/. Both are configurable via appsettings.json. Then point storage_base_url at it:

"replay": { "storage_base_url": "http://127.0.0.1:5080" }