Add docs and support for Render.com deployments (khanon/oai-reverse-proxy!9)

.env.example

@@ -2,12 +2,13 @@
 # have sensible defaults.  See config.ts for more details.
 
 # PORT=7860
+# SERVER_TITLE=Coom Tunnel
 # MODEL_RATE_LIMIT=4
 # MAX_OUTPUT_TOKENS=300
 # LOG_LEVEL=info
 # REJECT_DISALLOWED=false
 # REJECT_MESSAGE="This content violates /aicg/'s acceptable use policy."
-# CHECK_KEYS=false
+# CHECK_KEYS=true
 # QUOTA_DISPLAY_MODE=full
 # QUEUE_MODE=fair
 
@@ -23,10 +24,9 @@
 # PROMPT_LOGGING=false
 
 # ------------------------------------------------------------------------------
-# !!! EVERYTHING BELOW IS FOR LOCAL DEVELOPMENT ONLY !!!
-# If you are deploying this to Huggingface Spaces, do not set any keys here as
-# the .env file is public and your keys will be exposed.
-# Sensitive keys should be set in the Secrets UI instead.
+# The values below are secret -- make sure they are set securely.
+# For Huggingface, set them via the Secrets section in your Space's config UI.
+# For Render, create a "secret file" called .env using the Environment tab.
 
 # You can add multiple keys by separating them with a comma.
 OPENAI_KEY=sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

.replit

@@ -1,27 +0,0 @@
-run = "npm run start:replit"
-hidden = [".build", ".config"]
-
-[packager]
-language = "nodejs"
-
-[packager.features]
-enabledForHosting = false
-packageSearch = true
-guessImports = true
-
-[nix]
-channel = "stable-22_11"
-
-[gitHubImport]
-requiredFiles = [".replit", "replit.nix", ".config"]
-
-[languages]
-
-[languages.typescript]
-pattern = "**/{*.ts,*.js,*.tsx,*.jsx}"
-
-[languages.typescript.languageServer]
-start = "typescript-language-server --stdio"
-
-[deployment]
-run = ["sh", "-c", "npm run start:replit"]

README.md

@@ -26,14 +26,10 @@ This proxy only forwards text generation requests to the downstream service and
 If you'd like to run your own instance of this proxy, you'll need to deploy it somewhere and configure it with your API keys. A few easy options are provided below, though you can also deploy it to any other service you'd like.
 
 ### Deploy to Huggingface (Recommended)
-[See here for instructions on how to deploy to a Huggingface Space.](./docs/huggingface.md)
+[See here for instructions on how to deploy to a Huggingface Space.](./docs/deploy-huggingface.md)
 
-### Deploy to Repl.it (WIP)
-Still working on this. It's a bit more technical than the Huggingface option; you can give it a shot by clicking on the button below.
-
-[![Run on Repl.it](https://replit.com/badge/github/nai-degen/oai-reverse-proxy)](https://replit.com/new/github/nai-degen/oai-reverse-proxy)
-
-You'll need to set your secrets in Repl.it similar to the Huggingface instructions above.  Currently .env files don't work properly so it only uses the default configuration.
+### Deploy to Render
+[See here for instructions on how to deploy to Render.com.](./docs/deploy-render.md)
 
 ## Local Development
 To run the proxy locally for development or testing, install Node.js >= 18.0.0 and follow the steps below.

docker/Dockerfile

@@ -1,13 +0,0 @@
-FROM node:18-bullseye-slim
-
-WORKDIR /app
-
-COPY package*.json ./
-RUN npm install
-
-COPY . .
-RUN npm run build
-
-EXPOSE 7860
-
-CMD [ "npm", "start" ]

docker/huggingface/Dockerfile

@@ -0,0 +1,11 @@
+FROM node:18-bullseye-slim
+RUN apt-get update && \
+    apt-get install -y git
+RUN git clone https://gitgud.io/khanon/oai-reverse-proxy.git /app
+WORKDIR /app
+RUN npm install
+COPY Dockerfile greeting.md* .env* ./
+RUN npm run build
+EXPOSE 7860
+ENV NODE_ENV=production
+CMD [ "npm", "start" ]

docker/render/Dockerfile

@@ -0,0 +1,27 @@
+# syntax = docker/dockerfile:1.2
+
+FROM node:18-bullseye-slim
+RUN apt-get update && \
+    apt-get install -y curl
+    
+# Unlike Huggingface, Render can only deploy straight from a git repo and
+# doesn't allow you to create or modify arbitrary files via the web UI.
+# To use a greeting file, set `GREETING_URL` to a URL that points to a raw
+# text file containing your greeting, such as a GitHub Gist.
+
+# You may need to clear the build cache if you change the greeting, otherwise
+# Render will use the cached layer from the previous build.
+
+WORKDIR /app
+ARG GREETING_URL
+RUN if [ -n "$GREETING_URL" ]; then \
+    curl -sL "$GREETING_URL" > greeting.md; \
+    fi
+COPY package*.json greeting.md* ./
+RUN npm install
+COPY . .
+RUN npm run build
+RUN --mount=type=secret,id=_env,dst=/etc/secrets/.env cat /etc/secrets/.env >> .env
+EXPOSE 10000
+ENV NODE_ENV=production
+CMD [ "npm", "start" ]

docs/deploy-render.md

@@ -0,0 +1,49 @@
+# Deploy to Render.com
+Render.com offers a free tier that includes 750 hours of compute time per month.  This is enough to run a single proxy instance 24/7.  Instances shut down after 15 minutes without traffic but start up again automatically when a request is received.
+
+### 1. Create account
+- [Sign up for Render.com](https://render.com/) to create an account and access the dashboard.
+
+### 2. Create a service using a Blueprint
+Render allows you to deploy and auutomatically configure a repository containing a [render.yaml](../render.yaml) file using its Blueprints feature.  This is the easiest way to get started.
+
+- Click the **Blueprints** tab at the top of the dashboard.
+- Click **New Blueprint Instance**.
+- Under **Public Git repository**, enter `https://gitlab.com/khanon/oai-proxy`.
+  - Note that this is not the GitGud repository, but a mirror on GitLab.
+- Click **Continue**.
+- Under **Blueprint Name**, enter a name.
+- Under **Branch**, enter `main`.
+- Click **Create New Resources**.
+
+The service will be created according to the instructions in the `render.yaml` file.  This will take a few minutes.  It may show that the build failed, which is normal as you have not yet set the required environment variables.
+
+### 3. Set environment variables
+- Return to the **Dashboard** tab.
+- Click the name of the service you just created, which may show as "Deploy failed".
+- Click the **Environment** tab.
+- Click **Add Secret File**.
+- Under **Filename**, enter `.env`.
+- Under **Contents**, enter all of your environment variables, one per line, in the format `NAME=value`.
+  - For example, `OPENAI_KEY=sk-abc123`.
+- Click **Save Changes**.
+
+The service will automatically rebuild and deploy with the new environment variables.  This will take a few minutes.  The link to your deployed proxy will appear at the top of the page.
+
+# Optional
+
+## Updating the server
+
+To update your server, go to the page for your Web Service and click **Manual Deploy** > **Deploy latest commit**.  This will pull the latest version of the code and redeploy the server.
+
+_If you have trouble with this, you can also try selecting **Clear build cache & deploy** instead from the same menu._
+
+## Adding a greeting message
+
+To show a greeting message on the Server Info page, set the `GREETING_URL` environment variable within Render to the URL of a Markdown file.  This URL should point to a raw text file, not an HTML page. You can use a public GitHub Gist or GitLab Snippet for this.  For example: `GREETING_URL=https://gitlab.com/-/snippets/2542011/raw/main/greeting.md`.  You can change the title of the page by setting the `SERVER_TITLE` environment variable.
+
+Don't set `GREETING_URL` in the `.env` secret file you created earlier; it must be set in Render's environment variables section for it to work correctly.
+
+## Customizing the server
+
+You can customize the server by editing the `.env` configuration you created earlier. Refer to [.env.example](../.env.example) for a list of all available configuration options. Further information can be found in the [config.ts](../src/config.ts) file.

render.yaml

@@ -0,0 +1,10 @@
+services:
+  - type: web
+    name: oai-proxy
+    env: docker
+    repo: https://gitlab.com/khanon/oai-proxy.git
+    region: oregon
+    plan: free
+    branch: main
+    healthCheckPath: /health
+    dockerfilePath: ./docker/render/Dockerfile   

replit.nix

@@ -1,9 +0,0 @@
-{ pkgs }: {
-    deps = [
-        pkgs.nodejs-18_x
-        pkgs.yarn
-        pkgs.esbuild
-        pkgs.nodePackages.typescript
-        pkgs.nodePackages.typescript-language-server
-    ];
-}

src/info-page.ts

@@ -16,7 +16,8 @@ export const handleInfoPage = (req: Request, res: Response) => {
     return;
   }
 
-  // Huggingface puts spaces behind some cloudflare ssl proxy, so `req.protocol` is `http` but the correct URL is actually `https`
+  // Some load balancers/reverse proxies don't give us the right protocol in
+  // the host header. Huggingface works this way, Cloudflare does not.
   const host = req.get("host");
   const isHuggingface = host?.includes("hf.space");
   const protocol = isHuggingface ? "https" : req.protocol;
@@ -80,12 +81,10 @@ function cacheInfoPageHtml(host: string) {
     ...getQueueInformation(),
     keys: keyInfo,
     config: listConfig(),
-    commitSha: process.env.COMMIT_SHA || "dev",
+    build: process.env.COMMIT_SHA || "dev",
   };
 
-  const title = process.env.SPACE_ID
-    ? `${process.env.SPACE_AUTHOR_NAME} / ${process.env.SPACE_TITLE}`
-    : "OAI Reverse Proxy";
+  const title = getServerTitle();
   const headerHtml = buildInfoPageHeader(new showdown.Converter(), title);
 
   const pageBody = `<!DOCTYPE html>
@@ -160,3 +159,22 @@ function getQueueInformation() {
     estimatedQueueTime: waitMs > 2000 ? waitTime : "no wait",
   };
 }
+
+function getServerTitle() {
+  // Use manually set title if available
+  if (process.env.SERVER_TITLE) {
+    return process.env.SERVER_TITLE;
+  }
+
+  // Huggingface
+  if (process.env.SPACE_ID) {
+    return `${process.env.SPACE_AUTHOR_NAME} / ${process.env.SPACE_TITLE}`;
+  }
+
+  // Render
+  if (process.env.RENDER) {
+    return `Render / ${process.env.RENDER_SERVICE_NAME}`;
+  }
+
+  return "OAI Reverse Proxy";
+}

src/server.ts

@@ -22,8 +22,12 @@ app.use(
   pinoHttp({
     quietReqLogger: true,
     logger,
-    // SillyTavern spams the hell out of this endpoint so don't log it
-    autoLogging: { ignore: (req) => req.url === "/proxy/kobold/api/v1/model" },
+    autoLogging: {
+      ignore: (req) => {
+        const ignored = ["/proxy/kobold/api/v1/model", "/health"];
+        return ignored.includes(req.url as string);
+      },
+    },
     redact: {
       paths: [
         "req.headers.cookie",
@@ -36,6 +40,8 @@ app.use(
     },
   })
 );
+
+app.get("/health", (_req, res) => res.sendStatus(200));
 app.use((req, _res, next) => {
   req.startTime = Date.now();
   req.retryCount = 0;
@@ -46,9 +52,10 @@ app.use(
   express.json({ limit: "10mb" }),
   express.urlencoded({ extended: true, limit: "10mb" })
 );
-// TODO: this works if we're always being deployed to Huggingface but if users
-// deploy this somewhere without a load balancer then incoming requests can
-// spoof the X-Forwarded-For header and bypass the rate limiting.
+
+// TODO: Detect (or support manual configuration of) whether the app is behind
+// a load balancer/reverse proxy, which is necessary to determine request IP
+// addresses correctly.
 app.set("trust proxy", true);
 
 // routes
@@ -126,6 +133,17 @@ function registerUncaughtExceptionHandler() {
 }
 
 function setGitSha() {
+  // On Render, the .git directory isn't available in the docker build context
+  // so we can't get the SHA directly, but they expose it as an env variable.
+  if (process.env.RENDER) {
+    const shaString = `${process.env.RENDER_GIT_COMMIT?.slice(0, 7)} (${
+      process.env.RENDER_GIT_REPO_SLUG
+    })`;
+    process.env.COMMIT_SHA = shaString;
+    logger.info({ sha: shaString }, "Got commit SHA via Render config.");
+    return;
+  }
+
   try {
     // Huggingface seems to have changed something about how they deploy Spaces
     // and git commands fail because of some ownership issue with the .git

src/types/custom.d.ts

@@ -7,7 +7,7 @@ declare global {
     interface Request {
       key?: Key;
       api: "kobold" | "openai" | "anthropic";
-      user: User;
+      user?: User;
       isStreaming?: boolean;
       startTime: number;
       retryCount: number;