internal/autogen/extramodels_test.go
Rama main · réplica de lectura del repositorio original.
package autogen
import (
"os"
"path/filepath"
"slices"
"strings"
"testing"
"github.com/quartermaster-labs/quartermaster/internal/config"
"gopkg.in/yaml.v3"
)
// generateFromYAML writes a generate file with the given settings tail and
// runs the full LoadGenerateFile -> Generate trip over an empty models root.
func generateFromYAML(t *testing.T, settingsTail string) string {
t.Helper()
dir := t.TempDir()
models := filepath.Join(dir, "models")
if err := os.MkdirAll(models, 0o755); err != nil {
t.Fatal(err)
}
gen := filepath.Join(dir, "quartermaster-generate.yaml")
yml := "settings:\n modelsRoot: " + filepath.ToSlash(models) + "\n" + settingsTail
if err := os.WriteFile(gen, []byte(yml), 0o644); err != nil {
t.Fatal(err)
}
gf, err := LoadGenerateFile(gen, "")
if err != nil {
t.Fatal(err)
}
out, err := Generate(gf, "test")
if err != nil {
t.Fatal(err)
}
return out
}
// The point of extraModels: a hand-written block survives a regen, loads
// through the real config loader, and lands in the exclusive swap group so the
// scheduler evicts around it like any discovered model.
func TestGenerate_ExtraModelEmittedAsWritten(t *testing.T) {
out := generateFromYAML(t, ""+
" extraModels:\n"+
" strata-qwen:\n"+
" cmd: >\n"+
" strata-server --port ${PORT}\n"+
" --model D:/models/qwen\n"+
" proxy: http://127.0.0.1:${PORT}\n"+
" checkEndpoint: /v1/models\n"+
" estVramGB: 18.5\n"+
" ttl: 600\n")
if !slices.Contains(modelKeys(out), "strata-qwen") {
t.Fatalf("extra model not emitted:\n%s", out)
}
cfg, err := config.LoadConfigFromReader(strings.NewReader(out))
if err != nil {
t.Fatalf("generated config does not load: %v\n%s", err, out)
}
m, ok := cfg.Models["strata-qwen"]
if !ok {
t.Fatalf("loader dropped the extra model:\n%s", out)
}
if m.EstVramGB != 18.5 || m.CheckEndpoint != "/v1/models" || m.UnloadAfter != 600 {
t.Fatalf("fields did not round-trip: %+v", m)
}
if !strings.Contains(m.Cmd, "strata-server --port") || !strings.Contains(m.Cmd, "--model D:/models/qwen") {
t.Fatalf("cmd did not round-trip: %q", m.Cmd)
}
g, ok := cfg.Groups["exclusive"]
if !ok || !slices.Contains(g.Members, "strata-qwen") || !g.Exclusive {
t.Fatalf("extra model is not in the exclusive swap group: %+v", cfg.Groups)
}
}
// A broken entry is skipped with a reason and never takes the rest of the
// config (or the good entry beside it) down with it.
func TestGenerate_ExtraModelInvalidSkipped(t *testing.T) {
cases := map[string]struct{ block, reason string }{
"typo": {
" cmd: strata-server --port ${PORT}\n estVramGB: 4\n chekEndpoint: /health\n",
"chekEndpoint",
},
"no-cmd": {
" proxy: http://127.0.0.1:${PORT}\n estVramGB: 4\n",
"cmd is required",
},
"no-estimate": {
" cmd: strata-server --port ${PORT}\n",
"set estVramGB",
},
}
for name, c := range cases {
t.Run(name, func(t *testing.T) {
out := generateFromYAML(t, ""+
" extraModels:\n"+
" bad:\n"+c.block+
" good:\n"+
" cmd: other-server --port ${PORT}\n"+
" estRamGB: 2\n")
if slices.Contains(modelKeys(out), "bad") {
t.Fatalf("invalid extra model was emitted:\n%s", out)
}
if !strings.Contains(out, `# SKIPPED extra model "bad": `) || !strings.Contains(out, c.reason) {
t.Fatalf("no in-band reason containing %q:\n%s", c.reason, out)
}
cfg, err := config.LoadConfigFromReader(strings.NewReader(out))
if err != nil {
t.Fatalf("generated config does not load: %v\n%s", err, out)
}
if _, ok := cfg.Models["good"]; !ok {
t.Fatalf("valid sibling entry was lost:\n%s", out)
}
})
}
}
// A name a discovered model already serves keeps its owner.
func TestGenerate_ExtraModelNameClash(t *testing.T) {
root := t.TempDir()
seededGguf(t, root, "pub/repo/Model-7B-Q4_K_M.gguf")
gf := GenerateFile{Settings: Settings{ModelsRoot: root}}
gf.Settings.ExtraModels = mustNode(t, "model-7b-q4_k_m:\n cmd: strata-server --port ${PORT}\n estVramGB: 4\n")
out, err := Generate(gf, "test")
if err != nil {
t.Fatal(err)
}
assertUniqueKeys(t, out)
if strings.Contains(out, "strata-server") {
t.Fatalf("clashing extra model was emitted:\n%s", out)
}
if !strings.Contains(out, `# SKIPPED extra model "model-7b-q4_k_m": that name is already served`) {
t.Fatalf("clash left no in-band reason:\n%s", out)
}
}
func mustNode(t *testing.T, src string) yaml.Node {
t.Helper()
var doc yaml.Node
if err := yaml.Unmarshal([]byte(src), &doc); err != nil {
t.Fatal(err)
}
return *doc.Content[0]
}