llmgw/internal/cli/sourcecheck_test.go

package cli

import (
	"context"
	"net/http"
	"net/http/httptest"
	"strings"
	"testing"
)

func TestOnlyURLsAreChecked(t *testing.T) {
	// A path is checked by the node that will use it, which is the only
	// machine whose answer means anything — the control plane may not even
	// have that filesystem. A bare model id is the engine's business.
	for _, src := range []string{
		"/var/lib/llmgw-models/x.gguf", "./x.gguf", "Qwen/Qwen2.5-7B-Instruct",
	} {
		if _, ok := httpSource(src); ok {
			t.Errorf("%q was treated as a URL to check", src)
		}
		if _, err := checkSource(context.Background(), src); err != nil {
			t.Errorf("checking %q returned %v; it should be skipped", src, err)
		}
	}
}

func TestAnHFFileIsCheckedAtTheURLTheFetcherWouldUse(t *testing.T) {
	// Asking about a string that merely looks similar would be worse than not
	// asking: it would pass while the download later fails.
	got, ok := httpSource("hf:Qwen/Qwen2.5-Coder-3B-Instruct-GGUF/q4.gguf")
	if !ok {
		t.Fatal("an hf: file was not recognised")
	}
	want := "https://huggingface.co/Qwen/Qwen2.5-Coder-3B-Instruct-GGUF/resolve/main/q4.gguf"
	if got != want {
		t.Errorf("hf: resolved to %q, want %q", got, want)
	}
}

func TestAnHFRepositoryIsNotChecked(t *testing.T) {
	// vLLM and SGLang download whole repositories themselves; there is no one
	// file to ask about.
	if _, ok := httpSource("hf:Qwen/Qwen2.5-7B-Instruct"); ok {
		t.Error("a repository was treated as a single file")
	}
}

func TestAReachableSourceReportsItsSize(t *testing.T) {
	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		if r.Method != http.MethodHead {
			t.Errorf("the check used %s, not HEAD — it must not download the file", r.Method)
		}
		w.Header().Set("Content-Length", "1073741824")
		w.WriteHeader(http.StatusOK)
	}))
	defer srv.Close()

	size, err := checkSource(context.Background(), srv.URL+"/model.gguf")
	if err != nil {
		t.Fatal(err)
	}
	if size != 1<<30 {
		t.Errorf("size = %d", size)
	}
}

func TestAMissingFileSaysSoWithTheURL(t *testing.T) {
	// The whole point: a typo becomes a sentence while the operator still has
	// the file open, rather than a replica failing minutes later on a node
	// with an error that names the node.
	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
		http.Error(w, "not found", http.StatusNotFound)
	}))
	defer srv.Close()

	_, err := checkSource(context.Background(), srv.URL+"/typo.gguf")
	if err == nil {
		t.Fatal("a 404 was accepted")
	}
	if !strings.Contains(err.Error(), "typo.gguf") {
		t.Errorf("the error does not name the URL: %v", err)
	}
}

func TestAServerThatSendsNoLengthIsStillFine(t *testing.T) {
	// A resumable fetch does not need the length in advance, and refusing here
	// would reject a mirror that is merely terse.
	// A server that does not know the length omits the header rather than
	// sending a negative one, which is what a real mirror streaming from a
	// pipe does.
	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
		w.WriteHeader(http.StatusOK)
	}))
	defer srv.Close()

	size, err := checkSource(context.Background(), srv.URL+"/x.gguf")
	if err != nil {
		t.Errorf("a terse server was rejected: %v", err)
	}
	if size != 0 {
		t.Errorf("an unknown length came back as %d, want 0", size)
	}
}

func TestTheNoteSaysWhereTheAnswerCameFrom(t *testing.T) {
	// The control plane's network is not the nodes' network. The check catches
	// typos and a mirror that is down; it is not proof the fetch will work,
	// and the wording must not suggest it is.
	note := sourceNote("http://models.internal/x.gguf", 1<<30)
	if !strings.Contains(note, "from here") {
		t.Errorf("the note overstates what was checked: %s", note)
	}
}