mesh-controller/internal/builder/disk_1834_test.go

package builder

import (
	"context"
	"errors"
	"os"
	"path/filepath"
	"strings"
	"testing"
)

// Issue 1834: a build node's root disk filled and every check there ended "the store never became ready". A node
// frees what each build and check leaves, refuses to take work under a floor of free space, and a store that does
// not come up says why.

func init() {
	// The disk the tests run on is none of their business; the tests of the floor say their own.
	FreeBytes = func(string) (uint64, error) { return 500 << 30, nil }
}

func withFree(t *testing.T, bytes uint64) {
	t.Helper()
	was := FreeBytes
	FreeBytes = func(string) (uint64, error) { return bytes, nil }
	t.Cleanup(func() { FreeBytes = was })
}

// 1834/AC2: given 3 GiB free, when a check or a build is asked, then it refuses at once, saying so, running nothing.
func TestACheckUnderTheFloorOfFreeSpaceRefusesSayingSo(t *testing.T) {
	withFree(t, 3<<30)
	var ran []string
	run := func(_ context.Context, _ string, name string, args ...string) (string, error) {
		ran = append(ran, name+" "+strings.Join(args, " "))
		return "", nil
	}
	_, err := Check(context.Background(), run, CheckSpec{ID: "build-1"}, t.TempDir(), "registry.invalid", GitCredential{}, nil)
	if err == nil || !strings.Contains(err.Error(), "the build node has 3 GiB free on /, under the floor of 20 GiB") {
		t.Fatalf("a check on a full disk said %v", err)
	}
	if len(ran) != 0 {
		t.Errorf("a refused check ran %v", ran)
	}
}

func TestABuildUnderTheFloorOfFreeSpaceRefusesSayingSo(t *testing.T) {
	withFree(t, 3<<30)
	r := &recorded{}
	_, err := Build(context.Background(), r.run, r, "https://forge.invalid/x.git", "", "", t.TempDir(), nil, Npmrc{}, GitCredential{}, nil)
	if err == nil || !strings.Contains(err.Error(), "under the floor of 20 GiB") {
		t.Fatalf("a build on a full disk said %v", err)
	}
	if len(r.ran) != 0 {
		t.Errorf("a refused build ran %v", r.ran)
	}
}

// The workspace's filesystem is read too, and a workspace not made yet is read at its nearest folder.
func TestTheWorkspacesFilesystemIsReadToo(t *testing.T) {
	was := FreeBytes
	t.Cleanup(func() { FreeBytes = was })
	missing := filepath.Join(t.TempDir(), "not", "yet")
	FreeBytes = func(p string) (uint64, error) {
		if p == "/" {
			return 100 << 30, nil
		}
		if _, err := os.Stat(p); err != nil {
			t.Errorf("read the free space of %s, which is not there", p)
		}
		return 1 << 30, nil
	}
	err := RoomToWork("/", missing)
	if err == nil || !strings.Contains(err.Error(), "1 GiB free on "+missing) {
		t.Errorf("the workspace's disk, under the floor, said %v", err)
	}
	FreeBytes = func(string) (uint64, error) { return 0, errors.New("no") }
	if RoomToWork("/") == nil {
		t.Error("a disk that cannot be read was taken as having room")
	}
}

// 1834/AC3: given a store that stops at once, when the check waits for it, then its exit code and last log lines are
// in the error and the container is removed afterwards, not by --rm before anyone could read it.
func TestAStoreThatDiesSaysWhyItDied(t *testing.T) {
	var ran []string
	run := func(_ context.Context, _ string, name string, args ...string) (string, error) {
		line := name + " " + strings.Join(args, " ")
		ran = append(ran, line)
		switch {
		case args[0] == "exec":
			return "", errors.New("Error response from daemon: container is not running")
		case args[0] == "inspect":
			return "exited exit code 1\n", nil
		case args[0] == "logs":
			return "initdb: error: could not create directory: No space left on device\n", nil
		}
		return "", nil
	}
	err := waitReady(context.Background(), run, "build-1-store", []string{"pg_isready"}, 1)
	for _, want := range []string{"build-1-store never became ready", "exit code 1", "No space left on device"} {
		if err == nil || !strings.Contains(err.Error(), want) {
			t.Errorf("the verdict %q lacks %q", err, want)
		}
	}
	logs := ""
	for _, l := range ran {
		if strings.HasPrefix(l, "docker logs") {
			logs = l
		}
	}
	if logs != "docker logs --tail 20 build-1-store" {
		t.Errorf("the log was read as %q", logs)
	}
}

func TestTheThrowawayStoreIsNotStartedWithRm(t *testing.T) {
	var started []string
	run := func(_ context.Context, _ string, name string, args ...string) (string, error) {
		if args[0] == "run" {
			started = args
		}
		return "127.0.0.1:40000\n", nil
	}
	if _, err := throwaway(context.Background(), run, run, "build-1-store", "postgres:17-alpine", 5432, nil, nil); err != nil {
		t.Fatal(err)
	}
	for _, a := range started {
		if a == "--rm" {
			t.Errorf("started with --rm, which erases why it died: %v", started)
		}
	}
}

// 1834/AC1: given an image built for a check or a walk and published to the registry, when the build ends, then the
// local image tag, the narrowed source and the context clone it left are gone; an image published nowhere (a mesh
// being raised) is kept, since it is the artifact.
// pushedNamed is a publisher that tags the image as <registry>/<repository> to push it, as the registry one does.
type pushedNamed struct{ *recorded }

func (pushedNamed) PushedNames(repository string) []string {
	return []string{"registry.invalid/" + repository}
}

func TestABuildRemovesWhatItLeftOnTheNode(t *testing.T) {
	r := &recorded{
		contents:  map[string]string{"modules/m/" + ManifestName: anImageWithAContext, "modules/m/Dockerfile": "FROM scratch\n"},
		secondary: map[string]map[string]string{"https://forge.invalid/ctl.git": {"go.mod": "module x\n"}},
	}
	workspace := t.TempDir()
	if _, err := Build(context.Background(), r.run, pushedNamed{r}, "https://forge.invalid/c.git", "modules/m", "", workspace, nil, Npmrc{},
		GitCredential{}, nil); err != nil {
		t.Fatal(err)
	}
	// Both names go in one rmi: the layers stay while either is left (the push tagged the image with the registry's).
	if !strings.Contains(strings.Join(r.ran, "\n"), "docker rmi m-server:c0ffeec0 registry.invalid/m/server") {
		t.Errorf("the image built for it and its pushed name were not removed: %v", r.ran)
	}
	left, _ := filepath.Glob(filepath.Join(workspace, "*"))
	if len(left) != 0 {
		t.Errorf("the build left %v in its workspace", left)
	}
}

const anImageWithAContext = `{"module":"m","version":"1",
  "build":{"artifacts":[{"name":"server","kind":"image","from":"Dockerfile",
    "context":{"repository":"https://forge.invalid/ctl.git","ref":"main"}}]}}`

func TestAnImageKeptLocallyIsNotRemoved(t *testing.T) {
	r := &recorded{contents: map[string]string{"modules/m/" + ManifestName: anImageWithAContext, "modules/m/Dockerfile": "FROM scratch\n"}}
	local := Local{Run: func(ctx context.Context, dir, name string, args ...string) (string, error) {
		if args[0] == "image" {
			return "sha256:" + strings.Repeat("a", 64) + "\n", nil
		}
		return r.run(ctx, dir, name, args...)
	}}
	if _, err := Build(context.Background(), local.Run, local, "https://forge.invalid/c.git", "modules/m", "", t.TempDir(), nil, Npmrc{},
		GitCredential{}, nil); err != nil {
		t.Fatal(err)
	}
	if strings.Contains(strings.Join(r.ran, "\n"), "docker rmi") {
		t.Errorf("the artifact of a mesh being raised was removed: %v", r.ran)
	}
}

// The held image the comparison pulls to read is let go of once read.
func TestTheHeldImageAComparisonPullsIsRemovedAfterReading(t *testing.T) {
	var ran []string
	run := func(_ context.Context, _ string, name string, args ...string) (string, error) {
		ran = append(ran, name+" "+strings.Join(args, " "))
		if args[0] == "create" {
			return "abcdef0123456789\n", nil
		}
		if args[0] == "image" {
			return "", errors.New("stop here")
		}
		return "", nil
	}
	_, _ = imageContent(context.Background(), run, t.TempDir(), "reg.invalid/m/server@sha256:"+strings.Repeat("a", 64))
	if got := ran[len(ran)-1]; !strings.HasPrefix(got, "docker rmi reg.invalid/m/server@sha256:") {
		t.Errorf("the last command was %q, not the removal of the pulled image: %v", got, ran)
	}
}

func TestAnImageThatCannotBeRemovedIsNotSaidRemoved(t *testing.T) {
	var said []string
	run := func(context.Context, string, string, ...string) (string, error) {
		return "", errors.New("image is being used")
	}
	releaseImage(context.Background(), run, []string{"a:1"}, func(step, format string, args ...any) {
		said = append(said, format)
	})
	for _, s := range said {
		if strings.Contains(s, "is removed") {
			t.Errorf("said removed: %q", s)
		}
	}
}