Don't override DMR endpoint if we're running in a docker container

closes #118

Signed-off-by: Christopher Petito <chrisjpetito@gmail.com>

Author: krissetto

pkg/model/provider/dmr/client.go

@@ -7,6 +7,7 @@ import (
 	"errors"
 	"fmt"
 	"log/slog"
+	"os"
 	"os/exec"
 	"strconv"
 	"strings"
@@ -501,10 +502,16 @@ func getDockerModelEndpointAndEngine() (endpoint, engine string, err error) {
 		return "", "", err
 	}
 	endpoint = strings.TrimSpace(st.Endpoint)
-	// TODO(krissetto): temporary override for the internal dmr endpoint that `docker model status --json` currently returns
-	if endpoint == "http://model-runner.docker.internal/engines/v1/" {
-		endpoint = "http://localhost:12434/engines/llama.cpp/v1"
+
+	inDockerContainer := false
+	finfo, err := os.Stat("/.dockerenv")
+	if err == nil && finfo.Mode().IsRegular() {
+		inDockerContainer = true
 	}
+
+	// normalize endpoint considering container environment
+	endpoint = normalizeDMREndpoint(endpoint, inDockerContainer)
+
 	engine = strings.TrimSpace(st.Engine)
 	if engine == "" {
 		if st.Backends != nil {
@@ -524,6 +531,22 @@ func getDockerModelEndpointAndEngine() (endpoint, engine string, err error) {
 	return endpoint, engine, nil
 }
 
+// normalizeDMREndpoint applies an override to the endpoint reported by
+// `docker model status --json` to ensure the DMR client uses a reachable address
+// from the current environment.
+func normalizeDMREndpoint(endpoint string, inDockerContainer bool) string {
+	// This env overriding might need to be updated if we end up having multiple separate DMR
+	// engines with different endpoints running at the same time
+	if hostEnvVar := os.Getenv("MODEL_RUNNER_HOST"); hostEnvVar != "" {
+		return hostEnvVar
+	}
+	// Only override if not running in a docker container
+	if endpoint == "http://model-runner.docker.internal/engines/v1/" && !inDockerContainer {
+		return "http://localhost:12434/engines/llama.cpp/v1"
+	}
+	return endpoint
+}
+
 // buildRuntimeFlagsFromModelConfig converts standard ModelConfig fields into backend-specific
 // runtime flags that the model-runner understands when launching the engine.
 // Currently supports the default engine "llama.cpp". Unknown/unsupported fields are ignored.

pkg/model/provider/dmr/client_test.go

@@ -1,6 +1,7 @@
 package dmr
 
 import (
+	"context"
 	"reflect"
 	"testing"
 
@@ -15,7 +16,7 @@ func TestNewClientWithDefaultBaseURL(t *testing.T) {
 		// BaseURL is empty, should use default
 	}
 
-	client, err := NewClient(t.Context(), cfg)
+	client, err := NewClient(context.Background(), cfg)
 	if err != nil {
 		t.Fatalf("Expected no error, got %v", err)
 	}
@@ -34,7 +35,7 @@ func TestNewClientWithExplicitBaseURL(t *testing.T) {
 		BaseURL:  customURL,
 	}
 
-	client, err := NewClient(t.Context(), cfg)
+	client, err := NewClient(context.Background(), cfg)
 	if err != nil {
 		t.Fatalf("Expected no error, got %v", err)
 	}
@@ -51,7 +52,7 @@ func TestNewClientWithWrongType(t *testing.T) {
 		Model:    "gpt-4",
 	}
 
-	_, err := NewClient(t.Context(), cfg)
+	_, err := NewClient(context.Background(), cfg)
 	if err == nil {
 		t.Fatal("Expected error for wrong model type, got nil")
 	}
@@ -120,3 +121,85 @@ func TestMergeRuntimeFlagsPreferUser_WarnsAndPrefersUser(t *testing.T) {
 		t.Fatalf("unexpected merged flags.\nexpected: %#v\nactual:   %#v", expected, merged)
 	}
 }
+
+func TestNormalizeDMREndpoint_NoEnvOverride(t *testing.T) {
+	tests := []struct {
+		name              string
+		endpoint          string
+		inDockerContainer bool
+		want              string
+	}{
+		{
+			name:              "override when not in docker",
+			endpoint:          "http://model-runner.docker.internal/engines/v1/",
+			inDockerContainer: false,
+			want:              "http://localhost:12434/engines/llama.cpp/v1",
+		},
+		{
+			name:              "no override when in docker",
+			endpoint:          "http://model-runner.docker.internal/engines/v1/",
+			inDockerContainer: true,
+			want:              "http://model-runner.docker.internal/engines/v1/",
+		},
+		{
+			name:              "other endpoint unchanged",
+			endpoint:          "http://example/engines/v1/",
+			inDockerContainer: false,
+			want:              "http://example/engines/v1/",
+		},
+		{
+			name:              "empty endpoint unchanged",
+			endpoint:          "",
+			inDockerContainer: false,
+			want:              "",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := normalizeDMREndpoint(tt.endpoint, tt.inDockerContainer)
+			if got != tt.want {
+				t.Fatalf("normalizeDMREndpoint(%q, %v) = %q, want %q", tt.endpoint, tt.inDockerContainer, got, tt.want)
+			}
+		})
+	}
+}
+
+func TestNormalizeDMREndpoint_EnvOverride(t *testing.T) {
+	t.Setenv("MODEL_RUNNER_HOST", "http://myhost:9999/custom/v1")
+
+	tests := []struct {
+		name              string
+		endpoint          string
+		inDockerContainer bool
+		want              string
+	}{
+		{
+			name:              "env overrides non-container default endpoint",
+			endpoint:          "http://model-runner.docker.internal/engines/v1/",
+			inDockerContainer: false,
+			want:              "http://myhost:9999/custom/v1",
+		},
+		{
+			name:              "env overrides in-container default endpoint",
+			endpoint:          "http://model-runner.docker.internal/engines/v1/",
+			inDockerContainer: true,
+			want:              "http://myhost:9999/custom/v1",
+		},
+		{
+			name:              "env overrides arbitrary endpoint",
+			endpoint:          "http://example/engines/v1/",
+			inDockerContainer: false,
+			want:              "http://myhost:9999/custom/v1",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := normalizeDMREndpoint(tt.endpoint, tt.inDockerContainer)
+			if got != tt.want {
+				t.Fatalf("normalizeDMREndpoint should prefer env var: got %q, want %q", got, tt.want)
+			}
+		})
+	}
+}