llama-swap: 249 -> 262 (#559567)

nixos/modules/services/networking/llama-swap.nix

@@ -7,7 +7,29 @@
 let
   cfg = config.services.llama-swap;
   settingsFormat = pkgs.formats.yaml { };
-  configFile = settingsFormat.generate "config.yaml" cfg.settings;
+  unvalidatedConfigFile = settingsFormat.generate "llama-swap-config.yaml" cfg.settings;
+
+  # similar to writeNginxConfig
+  validateLlamaSwapConfig =
+    file:
+    pkgs.runCommandLocal "llama-swap-validated-config.yaml"
+      {
+        inherit file;
+        __structuredAttrs = true;
+        nativeBuildInputs = [
+          cfg.package
+        ];
+      } # sh
+      ''
+        llama-swap -validate -config $file
+        cp $file $out
+      '';
+
+  configFile =
+    if cfg.validateConfigFile then
+      validateLlamaSwapConfig unvalidatedConfigFile
+    else
+      unvalidatedConfigFile;
 in
 {
   options.services.llama-swap = {
@@ -15,6 +37,10 @@ in
 
     package = lib.mkPackageOption pkgs "llama-swap" { };
 
+    validateConfigFile = lib.mkEnableOption "validating configuration with `llama-swap -validate`" // {
+      default = true;
+    };
+
     listenAddress = lib.mkOption {
       type = lib.types.str;
       default = "localhost";

nixos/tests/web-servers/llama-swap.nix

@@ -54,6 +54,7 @@ let
           machine =
             { pkgs, ... }:
             {
+              environment.systemPackages = [ pkgs.jq ];
               # running models can be memory intensive but
               # default `virtualisation.memorySize` is fine
 
@@ -118,6 +119,7 @@ let
           ''
             # core tests
             import json
+            import datetime as dt
 
             def get_json(route):
               args = [
@@ -144,7 +146,7 @@ let
           ''
           + lib.optionalString withUI ''
             with subtest('check is serving ui'):
-              machine.succeed('curl --fail http:/localhost:8080/ui/')
+              machine.succeed('curl --fail http://localhost:8080/ui/')
 
           ''
           + ''
@@ -156,7 +158,9 @@ let
           + lib.optionalString useSmollm2-135m ''
             # extended tests using SmolLM2
             with subtest('check `/running` for preloaded smollm2'):
-              machine.wait_until_succeeds('curl --silent --fail http://localhost:8080/running | grep "smollm2"')
+              # wait until quick jq check passes
+              machine.wait_until_succeeds('curl --silent --fail http://localhost:8080/running | jq -e ".running[] | select(.model == \\"smollm2\\" and .state == \\"ready\\")"')
+              # now check granular details
               running_response = get_json('/running')
               assert len(running_response['running']) == 1, f'running doesn\'t have one entry as expected: {running_response}'
               running_model = running_response['running'][0]
@@ -199,7 +203,7 @@ let
                 assert running_model['state'] == 'ready', f'running smollm2 is not ready: {running_response}'
 
             with subtest('check `/running` for smollm2 to timeout'):
-              machine.wait_until_succeeds('curl --silent --fail http://localhost:8080/running | grep -v "smollm2"', timeout=11)
+              machine.wait_until_succeeds('curl --silent --fail http://localhost:8080/running | grep -v "smollm2"', timeout=dt.timedelta(seconds=11))
               running_response = get_json('/running')
               assert len(running_response['running']) == 0, "smollm2 was still running after timeout"
 

pkgs/by-name/ll/llama-swap/package.nix

@@ -2,7 +2,7 @@
   lib,
   stdenv,
 
-  buildGoModule,
+  buildGo127Module,
   fetchFromGitHub,
   versionCheckHook,
 
@@ -19,20 +19,23 @@
 let
   canExecute = stdenv.buildPlatform.canExecute stdenv.hostPlatform;
 in
-buildGoModule (finalAttrs: {
+buildGo127Module (finalAttrs: {
   pname = "llama-swap";
-  version = "249";
+  version = "262";
 
   outputs = [
     "out"
+    "kubeswap"
     "wol" # wake on lan proxy
+    "vllm_wrapper" # helper for using `vllm` with `--enable-sleep-mode`
+    "utils"
   ];
 
   src = fetchFromGitHub {
     owner = "mostlygeek";
     repo = "llama-swap";
     tag = "v${finalAttrs.version}";
-    hash = "sha256-7wXOL8XtcKV6Abdxar25C85ODQ34RYOAGYCTaCXxPpY=";
+    hash = "sha256-DAYaR+N7alGbdLsGlaiDlsHiqxOMg65069i/vHOkLKk=";
     # populate values that require us to use git. By doing this in postFetch we
     # can delete .git afterwards and maintain better reproducibility of the src.
     leaveDotGit = true;
@@ -45,7 +48,7 @@ buildGoModule (finalAttrs: {
     '';
   };
 
-  vendorHash = "sha256-MhR8B2+Yb/xqrTlIxaVHLoQf1eTOO49c65l72IAuZyU=";
+  vendorHash = "sha256-yelob7FlaGymASUP0DAUkALQm5vnXZnN5ThbnSkH2Ak=";
 
   # Upstream only embeds the UI when this build tag is set.
   tags = lib.optionals withUI [ "embed_ui" ];
@@ -68,6 +71,8 @@ buildGoModule (finalAttrs: {
   postPatch = ''
     substituteInPlace internal/process/process_command_forking_test.go \
       --replace-fail "#!/bin/bash" "#!${lib.getExe bash}"
+    substituteInPlace cmd/vllm-wrapper/main_test.go \
+      --replace-fail "#!/bin/bash" "#!${lib.getExe bash}"
   '';
 
   preBuild = ''
@@ -86,8 +91,10 @@ buildGoModule (finalAttrs: {
     "misc/process-cmd-test"
     # benchmark/regression testing tool
     "misc/benchmark-chatcompletion"
+    # testing tool for the performance monitoring code
+    "cmd/monitor-test"
   ]
-  ++ lib.optionals (!canExecute) [
+  ++ lib.optionals finalAttrs.doCheck [
     # some tests expect to execute `simple-something`; if it can't be executed
     # it's unneeded
     "misc/simple-responder"
@@ -124,14 +131,44 @@ buildGoModule (finalAttrs: {
     mkdir build
     ln -s "$GOPATH/bin/simple-responder" "./build/simple-responder_''${GOOS}_''${GOARCH}"
   '';
+  # doesn't need to be cleaned up outside of the check path as we don't build it
+  # due to `excludedPackages`
   postCheck = ''
     rm "$GOPATH/bin/simple-responder"
   '';
 
-  postInstall = ''
+  # project regularly adds small utility binaries
+  # this `installPhase` means we need to sort them
+  # into outputs or skip building them
+  installPhase = ''
+    runHook preInstall
+
     install -Dm444 -t "$out/share/llama-swap" config.example.yaml
+
+    mkdir -p "$out/bin"
+    mv "$GOPATH/bin/llama-swap" "$out/bin/"
+
+    mkdir -p "$kubeswap/bin"
+    mv "$GOPATH/bin/kubeswap" "$kubeswap/bin/"
+
     mkdir -p "$wol/bin"
-    mv "$out/bin/wol-proxy" "$wol/bin/"
+    mv "$GOPATH/bin/wol-proxy" "$wol/bin/"
+
+    mkdir -p "$vllm_wrapper/bin"
+    mv "$GOPATH/bin/vllm-wrapper" "$vllm_wrapper/bin/"
+
+    mkdir -p "$utils/bin"
+    mv "$GOPATH/bin/"{fake-model,test-concurrency} "$utils/bin/"
+
+    # log if we don't handle a new executable
+    EXECUTABLES=$(find "$GOPATH/bin" -maxdepth 1 -type f -executable -print -quit)
+    if [ -n "$EXECUTABLES" ]; then
+      echo "Error: Directory '$GOPATH/bin' still contains executables." >&2
+      echo "  $EXECUTABLES" >&2
+      exit 1
+    fi
+
+    runHook postInstall
   '';
 
   doInstallCheck = true;

pkgs/by-name/ll/llama-swap/ui.nix

@@ -7,14 +7,14 @@
 buildNpmPackage (finalAttrs: {
   pname = "${llama-swap.pname}-ui";
   inherit (llama-swap) version src;
-  npmDepsHash = "sha256-6MPXQtmaz97D9PUU2Nn5DH/2HZNP/rnAWVSck/FiCyk=";
+  npmDepsHash = "sha256-lmhRJ8275PIQ+7vHdr9aZ31lYeXUkXrWnlvuwOadjRQ=";
 
   postPatch = ''
     substituteInPlace vite.config.ts \
       --replace-fail "../internal/server/ui_dist" "${placeholder "out"}/ui_dist"
   '';
 
-  sourceRoot = "${finalAttrs.src.name}/ui-svelte";
+  sourceRoot = "${finalAttrs.src.name}/ui";
 
   # bundled "ui_dist" doesn't need node_modules
   postInstall = ''