basic metrics

.gitignore

@@ -0,0 +1 @@
+result

modules/llama-swap-exporter.nix

@@ -0,0 +1,58 @@
+{
+  config,
+  lib,
+  pkgs,
+  ...
+}:
+
+let
+  cfg = config.services.llama-swap-exporter;
+  exporter = pkgs.callPackage ../servers/eisen/llama-swap-exporter/default.nix { };
+in
+
+{
+  options.services.llama-swap-exporter = {
+    enable = lib.mkEnableOption "llama-swap Prometheus exporter";
+
+    port = lib.mkOption {
+      type = lib.types.port;
+      default = 9409;
+      description = "Port for the Prometheus metrics endpoint.";
+    };
+
+    url = lib.mkOption {
+      type = lib.types.str;
+      default = "http://localhost:8080/api/metrics";
+      description = "llama-swap metrics endpoint URL.";
+    };
+
+    interval = lib.mkOption {
+      type = lib.types.int;
+      default = 15;
+      description = "Scrape interval in seconds.";
+    };
+  };
+
+  config = lib.mkIf cfg.enable {
+    systemd.services.llama-swap-exporter = {
+      description = "llama-swap Prometheus exporter";
+      wantedBy = [ "multi-user.target" ];
+      after = [ "network.target" ];
+
+      serviceConfig = {
+        Type = "simple";
+        ExecStart = "${exporter}/bin/exporter.py";
+        Restart = "on-failure";
+        RestartSec = "5s";
+        User = "root";
+        Group = "root";
+        Environment = [
+          "PROMETHEUS_PORT=${toString cfg.port}"
+          "LLAMA_SWAP_URL=${cfg.url}"
+          "SCRAPE_INTERVAL=${toString cfg.interval}"
+        ];
+        ReadWritePaths = [ "/tmp" ];
+      };
+    };
+  };
+}

servers/eisen/configuration.nix

@@ -21,6 +21,7 @@ let
     grafana = 3002;
     tolgee = 8200;
     # ntfy = 3003;
+    llama-swap = 8080;
   };
   internalPorts = {
     prometheus-node = 9000;
@@ -28,6 +29,7 @@ let
     prometheus-sonarr = 9101;
     prometheus-radarr = 9102;
     prometheus-prowlarr = 9103;
+    prometheus-llama-swap = 9409;
     prometheus = 9090;
   };
 in
@@ -42,6 +44,7 @@ in
   imports = [
     nix-index-database.nixosModules.nix-index
     ./hardware-configuration.nix
+    ../../modules/llama-swap-exporter.nix
   ];
 
   nix = {
@@ -105,6 +108,12 @@ in
       environmentFile = "/etc/secrets/karakeep.env";
     };
 
+    llama-swap-exporter = {
+      enable = true;
+      url = "http://100.120.15.10:${toString ports.llama-swap}/api/metrics";
+      port = internalPorts.prometheus-llama-swap;
+    };
+
     dnsmasq = {
       enable = true;
     };
@@ -185,6 +194,14 @@ in
             }
           ];
         }
+        {
+          job_name = "llama-swap";
+          static_configs = [
+            {
+              targets = [ "localhost:${toString internalPorts.prometheus-llama-swap}" ];
+            }
+          ];
+        }
       ];
     };
 

servers/eisen/llama-swap-exporter/default.nix

@@ -0,0 +1,23 @@
+{
+  lib,
+  pkgs ? import <nixpkgs> { },
+  ...
+}:
+
+pkgs.python3Packages.buildPythonApplication {
+  pname = "llama-swap-exporter";
+  version = "0.1.0";
+
+  src = ./src;
+  build-system = with pkgs.python3Packages; [
+    setuptools
+    setuptools-scm
+  ];
+  pyproject = true;
+  meta = {
+    description = "Prometheus exporter for llama-swap metrics";
+    license = lib.licenses.mit;
+    maintainers = [ ];
+    platforms = lib.platforms.linux;
+  };
+}

servers/eisen/llama-swap-exporter/src/exporter.py

@@ -0,0 +1,114 @@
+#!/usr/bin/env python
+
+"""Prometheus exporter for llama-swap metrics endpoint."""
+
+import json
+import os
+import time
+import urllib.request
+from http.server import BaseHTTPRequestHandler, HTTPServer
+
+LLAMA_SWAP_URL = os.environ.get("LLAMA_SWAP_URL", "http://localhost:8080/metrics")
+PROMETHEUS_PORT = int(os.environ.get("PROMETHEUS_PORT", "9409"))
+SCRAPE_INTERVAL = int(os.environ.get("SCRAPE_INTERVAL", "15"))
+
+last_metrics = {}
+last_scrape_time = 0
+
+
+def scrape_llama_swap():
+    try:
+        req = urllib.request.Request(LLAMA_SWAP_URL)
+        with urllib.request.urlopen(req, timeout=10) as resp:
+            data = json.loads(resp.read().decode())
+            if not data:
+                print("No data found when scraping (request successful, empty data)")
+                return {}
+            entry = data[-1]
+            return {
+                "llama_cache_tokens": float(entry.get("cache_tokens", 0)),
+                "llama_input_tokens": float(entry.get("input_tokens", 0)),
+                "llama_output_tokens": float(entry.get("output_tokens", 0)),
+                "llama_prompt_per_second": entry.get("prompt_per_second", 0),
+                "llama_tokens_per_second": entry.get("tokens_per_second", 0),
+                "llama_duration_ms": float(entry.get("duration_ms", 0)),
+                "llama_model": entry.get("model", "unknown"),
+                "llama_has_capture": 1 if entry.get("has_capture") else 0,
+            }
+    except Exception as e:
+        print(f"Error scraping llama-swap: {e}")
+        return {}
+
+
+def format_metrics(metrics):
+    if not metrics:
+        metrics = {}
+
+    lines = []
+    model = metrics.get("llama_model", "unknown")
+
+    for name, value in metrics.items():
+        if name == "llama_model":
+            continue
+        if "second" in name:
+            lines.append(f"# HELP {name} Rate from llama-swap")
+            lines.append(f"# TYPE {name} gauge")
+            lines.append(f'{name}{{model="{model}"}} {value}')
+        elif "tokens" in name:
+            lines.append(f"# HELP {name} Total tokens from llama-swap")
+            lines.append(f"# TYPE {name} gauge")
+            lines.append(f'{name}{{model="{model}"}} {value}')
+        elif "duration" in name:
+            value_s = value / 1000.0
+            lines.append("# HELP llama_duration_seconds Inference duration")
+            lines.append("# TYPE llama_duration_seconds gauge")
+            lines.append(f'llama_duration_seconds{{model="{model}"}} {value_s}')
+        elif name == "llama_has_capture":
+            lines.append("# HELP llama_has_capture Whether capture is available")
+            lines.append("# TYPE llama_has_capture gauge")
+            lines.append(f'llama_has_capture{{model="{model}"}} {value}')
+
+    lines.append(
+        "# HELP llama_last_scrape_timestamp_seconds When metrics were last scraped"
+    )
+    lines.append("# TYPE llama_last_scrape_timestamp_seconds gauge")
+    lines.append(f"llama_last_scrape_timestamp_seconds {time.time()}")
+
+    return "\n".join(lines) + "\n"
+
+
+last_metrics = None
+
+
+class MetricsHandler(BaseHTTPRequestHandler):
+    def do_GET(self):
+        global last_scrape_time, last_metrics
+        if time.time() - last_scrape_time > SCRAPE_INTERVAL:
+            last_metrics = scrape_llama_swap()
+            last_scrape_time = time.time()
+
+        if self.path == "/metrics":
+            self.send_response(200)
+            self.send_header("Content-Type", "text/plain; charset=utf-8")
+            self.end_headers()
+            self.wfile.write(format_metrics(last_metrics).encode())
+        elif self.path == "/health":
+            self.send_response(200)
+            self.send_header("Content-Type", "text/plain")
+            self.end_headers()
+            self.wfile.write(b"OK")
+        else:
+            self.send_response(404)
+            self.end_headers()
+
+    def log_message(self, format, *args):
+        pass
+
+
+if __name__ == "__main__":
+    print(f"Starting llama-swap exporter on port {PROMETHEUS_PORT}")
+    print(f"Scraping llama-swap from: {LLAMA_SWAP_URL}")
+    server = HTTPServer(("0.0.0.0", PROMETHEUS_PORT), MetricsHandler)
+    server.serve_forever()
+else:
+    print("Exporter loaded")

servers/eisen/llama-swap-exporter/src/setup.py

@@ -0,0 +1,10 @@
+from setuptools import find_packages, setup
+
+setup(
+    name="llama-swap-exporter",
+    version="1.0",
+    # Modules to import from other scripts:
+    packages=find_packages(),
+    # Executables
+    scripts=["exporter.py"],
+)