Add alpine virtual repositories (#164)
ci/woodpecker/tag/docker Pipeline was successful

Virtual repos cover rpm, deb, npm and others but not alpine, so a local alpine repo and github_alpine remotes cannot be served as one apk repository.

- merge members' `<dir>/APKINDEX.tar.gz` per directory; member order wins duplicate (P, V, A); a member 404 is skipped
- 302 `<dir>/<P>-<V>.apk` downloads to the member whose index listed the package
- serve the merged index unsigned, like local and github_alpine indexes
- cache merges per directory on the shared virtual cache; member errors fail with 502
- add docs/alpine/virtual.md, unit tests and an apk e2e with abuild-built packages

Requires unkin/artifactapi#163.

Reviewed-on: #164
Co-authored-by: unkin-agent <unkin-agent@unkin.net>
Co-committed-by: unkin-agent <unkin-agent@unkin.net>
This commit was merged in pull request #164.
This commit is contained in:
2026-10-24 18:35:52 +11:00
committed by BenVincent
parent fe9de81afa
commit 9294686ea6
6 changed files with 448 additions and 4 deletions
+90
View File
@@ -0,0 +1,90 @@
package virtual
import (
"archive/tar"
"bytes"
"compress/gzip"
"errors"
"fmt"
"io"
"strings"
"time"
)
// AlpineMember is one member's decompressed APKINDEX for a directory; Index is
// nil when the member does not serve that directory.
type AlpineMember struct {
RemoteName string
Index []byte
}
// AlpineRepo is a merged directory: its APKINDEX.tar.gz plus, per
// "<name>-<version>.apk" apk will request, the member that owns it.
type AlpineRepo struct {
Index []byte
Owners map[string][]byte
}
func (r *AlpineRepo) index() []byte { return r.Index }
func (r *AlpineRepo) files() map[string][]byte { return r.Owners }
// MergeAlpine merges members' APKINDEX records. Members are in priority order:
// the first to carry a (P, V, A) wins. The index is unsigned, like the local
// and github_alpine indexes it merges.
func MergeAlpine(members []AlpineMember) *AlpineRepo {
repo := &AlpineRepo{Owners: map[string][]byte{}}
seen := map[string]bool{}
var b bytes.Buffer
for _, m := range members {
for _, rec := range splitStanzas(m.Index) {
p, v, a := stanzaField(rec, "P"), stanzaField(rec, "V"), stanzaField(rec, "A")
key := p + "\x00" + v + "\x00" + a
if p == "" || v == "" || seen[key] {
continue
}
seen[key] = true
if _, ok := repo.Owners[p+"-"+v+".apk"]; !ok {
repo.Owners[p+"-"+v+".apk"] = []byte(m.RemoteName)
}
b.WriteString(rec)
b.WriteString("\n\n")
}
}
var t bytes.Buffer
tw := tar.NewWriter(&t)
_ = tw.WriteHeader(&tar.Header{Name: "APKINDEX", Mode: 0o644, Size: int64(b.Len()), Typeflag: tar.TypeReg, ModTime: time.Unix(0, 0)})
_, _ = tw.Write(b.Bytes())
_ = tw.Close()
repo.Index = gzipDeterministic(t.Bytes())
return repo
}
// apkIndexText extracts the APKINDEX entry from an APKINDEX.tar.gz, skipping a
// leading signature stream if present.
func apkIndexText(body []byte) ([]byte, error) {
zr, err := gzip.NewReader(bytes.NewReader(body))
if err != nil {
return nil, err
}
tr := tar.NewReader(zr)
for {
hdr, err := tr.Next()
if errors.Is(err, io.EOF) {
return nil, fmt.Errorf("no APKINDEX entry")
}
if err != nil {
return nil, err
}
if hdr.Name == "APKINDEX" {
return io.ReadAll(tr)
}
}
}
// alpinePath splits a request into its directory and file, or ok is false for
// anything but an APKINDEX.tar.gz or .apk.
func alpinePath(p string) (dir, file string, ok bool) {
i := strings.LastIndex(p, "/")
dir, file = p[:i+1], p[i+1:]
return dir, file, dir != "" && (file == "APKINDEX.tar.gz" || strings.HasSuffix(file, ".apk"))
}
+140
View File
@@ -0,0 +1,140 @@
package virtual
import (
"archive/tar"
"bytes"
"compress/gzip"
"context"
"errors"
"strings"
"testing"
"git.unkin.net/unkin/artifactapi/pkg/models"
)
func apkRec(p, v, a string) string {
return "C:Q1" + p + v + "\nP:" + p + "\nV:" + v + "\nA:" + a + "\nS:10\n"
}
func apkTarGz(t *testing.T, name string, body []byte, closeTar bool) []byte {
t.Helper()
var tb bytes.Buffer
tw := tar.NewWriter(&tb)
if err := tw.WriteHeader(&tar.Header{Name: name, Mode: 0o644, Size: int64(len(body)), Typeflag: tar.TypeReg}); err != nil {
t.Fatal(err)
}
tw.Write(body)
if closeTar {
tw.Close()
} else {
tw.Flush()
}
var gb bytes.Buffer
gz := gzip.NewWriter(&gb)
gz.Write(tb.Bytes())
gz.Close()
return gb.Bytes()
}
func TestMergeAlpine(t *testing.T) {
repo := MergeAlpine([]AlpineMember{
{RemoteName: "a", Index: []byte(apkRec("foo", "1.0-r0", "x86_64") + "\n" + apkRec("bar", "1-r0", "noarch") + "\nC:Q1x\nV:9\n")},
{RemoteName: "b"},
{RemoteName: "c", Index: []byte(apkRec("foo", "1.0-r0", "x86_64") + "\n" + apkRec("foo", "2.0-r0", "x86_64") + "\n" + apkRec("bar", "1-r0", "x86_64"))},
})
idx, err := apkIndexText(repo.Index)
if err != nil {
t.Fatal(err)
}
recs := splitStanzas(idx)
if len(recs) != 4 || strings.Count(string(idx), "P:foo\nV:1.0-r0\n") != 1 {
t.Fatalf("want foo 1.0 once, foo 2.0 and bar for both arches, record without P dropped:\n%s", idx)
}
for file, want := range map[string]string{"foo-1.0-r0.apk": "a", "foo-2.0-r0.apk": "c", "bar-1-r0.apk": "a"} {
if got := string(repo.Owners[file]); got != want {
t.Errorf("%s owned by %q, want %q", file, got, want)
}
}
if again := MergeAlpine([]AlpineMember{{RemoteName: "a", Index: idx}}); !bytes.Equal(again.Index, MergeAlpine([]AlpineMember{{RemoteName: "a", Index: idx}}).Index) {
t.Fatal("merged index must be byte-stable across merges")
}
}
func TestApkIndexTextSkipsSignature(t *testing.T) {
signed := append(apkTarGz(t, ".SIGN.RSA.alpine.rsa.pub", []byte("sig"), false), apkTarGz(t, "APKINDEX", []byte(apkRec("foo", "1", "x86_64")), true)...)
idx, err := apkIndexText(signed)
if err != nil || !strings.Contains(string(idx), "P:foo\n") {
t.Fatalf("signed index: %v\n%s", err, idx)
}
if _, err := apkIndexText(apkTarGz(t, "DESCRIPTION", []byte("x"), true)); err == nil {
t.Fatal("index without APKINDEX entry should fail")
}
}
func alpineEngine(members map[string]string) *Engine {
e := fakeEngine(nil)
e.apkMember = func(_ context.Context, name, dir string) (*AlpineMember, error) {
idx, ok := members[name]
if !ok {
return nil, errors.New("upstream down")
}
if dir != "x86_64/" {
return &AlpineMember{RemoteName: name}, nil
}
return &AlpineMember{RemoteName: name, Index: []byte(idx)}, nil
}
return e
}
func alpineVirt(members ...string) models.Virtual {
return models.Virtual{Name: "v", PackageType: models.PackageAlpine, Members: members}
}
func TestFetchAlpineIndexAndRedirect(t *testing.T) {
e := alpineEngine(map[string]string{"loc": apkRec("foo", "1-r0", "x86_64"), "gh": apkRec("foo", "1-r0", "x86_64") + "\n" + apkRec("tool", "2-r0", "x86_64")})
virt := alpineVirt("loc", "gh")
body, ct, err := e.Fetch(context.Background(), virt, "./x86_64/APKINDEX.tar.gz", "")
if err != nil || ct != "application/gzip" {
t.Fatalf("fetch index: %v %q", err, ct)
}
if idx, _ := apkIndexText(body); strings.Count(string(idx), "P:") != 2 {
t.Fatalf("merged index:\n%s", idx)
}
for _, tc := range []struct {
path, want string
ok bool
}{
{path: "x86_64/foo-1-r0.apk", ok: true, want: "http://h/api/v1/remote/loc/x86_64/foo-1-r0.apk"},
{path: "./x86_64/tool-2-r0.apk", ok: true, want: "http://h/api/v1/remote/gh/x86_64/tool-2-r0.apk"},
{path: "x86_64/../x86_64/tool-2-r0.apk", ok: true, want: "http://h/api/v1/remote/gh/x86_64/tool-2-r0.apk"},
{path: "x86_64/missing-1-r0.apk"},
{path: "aarch64/foo-1-r0.apk"},
{path: "x86_64/APKINDEX.tar.gz"},
{path: "foo-1-r0.apk"},
} {
got, ok, err := e.MemberRedirect(context.Background(), virt, tc.path, "", "http://h/")
if got != tc.want || ok != tc.ok || err != nil {
t.Errorf("%s: got (%q, %v, %v), want (%q, %v)", tc.path, got, ok, err, tc.want, tc.ok)
}
}
for _, p := range []string{"x86_64/missing-1-r0.apk", "x86_64/DESCRIPTION", "APKINDEX.tar.gz"} {
if _, _, err := e.Fetch(context.Background(), virt, p, ""); !errors.Is(err, ErrNotFound) {
t.Errorf("%s: want not found, got %v", p, err)
}
}
}
func TestFetchAlpineFailsClosed(t *testing.T) {
e := alpineEngine(map[string]string{"a": apkRec("foo", "1-r0", "x86_64")})
virt := alpineVirt("a", "down")
if _, _, err := e.Fetch(context.Background(), virt, "x86_64/APKINDEX.tar.gz", ""); err == nil || errors.Is(err, ErrNotFound) {
t.Fatalf("index with a member down: want upstream error, got %v", err)
}
if _, ok, err := e.MemberRedirect(context.Background(), virt, "x86_64/foo-1-r0.apk", "", "http://h/"); ok || err != nil {
t.Fatalf("redirect with a member down must defer to Fetch: ok %v err %v", ok, err)
}
if _, _, err := e.Fetch(context.Background(), virt, "x86_64/foo-1-r0.apk", ""); err == nil || errors.Is(err, ErrNotFound) {
t.Fatalf("download with a member down: want upstream error, got %v", err)
}
}
+82 -3
View File
@@ -28,6 +28,7 @@ type Engine struct {
getRemote func(context.Context, string) (*models.Remote, error)
rpmMember func(context.Context, string) (*RPMMember, error)
debMember func(ctx context.Context, name, base string) (*DebMember, error)
apkMember func(ctx context.Context, name, dir string) (*AlpineMember, error)
npmMember func(ctx context.Context, name, pkg string) (*NPMMember, error)
// NPMLocal returns a local npm repo's packument with tarball paths relative
@@ -76,13 +77,15 @@ type repoGen struct {
const rpmMergeTTL = 60 * time.Second
// maxDebSuites bounds the client-chosen suites cached per deb virtual.
// maxDebSuites bounds the client-chosen deb suites or alpine directories
// cached per virtual.
const maxDebSuites = 32
func NewEngine(db *database.DB, proxyEngine *proxy.Engine) *Engine {
e := &Engine{db: db, proxyEngine: proxyEngine, getRemote: db.GetRemote, mergeTTL: rpmMergeTTL}
e.rpmMember = e.fetchRPMMember
e.debMember = e.fetchDebMember
e.apkMember = e.fetchAlpineMember
e.npmMember = e.fetchNPMMember
e.gemsMember = e.fetchGemsMember
e.ansibleDoc = e.ansibleMemberDoc
@@ -97,6 +100,8 @@ func (e *Engine) Fetch(ctx context.Context, virt models.Virtual, path string, pr
return e.fetchRPM(ctx, virt, path)
case models.PackageDeb:
return e.fetchDeb(ctx, virt, path)
case models.PackageAlpine:
return e.fetchAlpine(ctx, virt, path)
}
merger, err := GetMerger(virt.PackageType)
@@ -216,8 +221,8 @@ var (
)
// MemberRedirect maps an rpm, deb, npm or ansible virtual package path
// (<member>/<href>, as written by MergeRPM/MergeDeb/MergeNPM/AnsibleVersion)
// to the owning member's route. ok is false for any other path; ErrBadPath
// (<member>/<href>, as written by MergeRPM/MergeDeb/MergeNPM/AnsibleVersion),
// or an alpine <dir>/<name>-<version>.apk, to the owning member's route. ok is false for any other path; ErrBadPath
// means the path is absolute or escapes the member.
func (e *Engine) MemberRedirect(ctx context.Context, virt models.Virtual, path, rawQuery, proxyBaseURL string) (string, bool, error) {
switch virt.PackageType {
@@ -231,6 +236,8 @@ func (e *Engine) MemberRedirect(ctx context.Context, virt models.Virtual, path,
if strings.HasPrefix(path, "dists/") {
return "", false, nil
}
case models.PackageAlpine:
return e.alpineRedirect(ctx, virt, path, rawQuery, proxyBaseURL)
case models.PackageNPM, models.PackageAnsible:
default:
return "", false, nil
@@ -445,6 +452,78 @@ func (e *Engine) fetchDebMember(ctx context.Context, name, base string) (*DebMem
return m, nil
}
func (e *Engine) mergedAlpine(ctx context.Context, virt models.Virtual, dir string) (mergedRepo, error) {
repo, err := e.merged(ctx, "alpine\x00"+virt.Name+"\x00"+dir, func(ctx context.Context) (mergedRepo, error) {
members, err := fetchMembers(virt, func(name string) (*AlpineMember, error) { return e.apkMember(ctx, name, dir) })
if err != nil {
return nil, err
}
return MergeAlpine(members), nil
})
if err == nil {
e.evictLRU("alpine\x00"+virt.Name+"\x00", maxDebSuites)
}
return repo, err
}
func (e *Engine) fetchAlpine(ctx context.Context, virt models.Virtual, reqPath string) ([]byte, string, error) {
dir, file, ok := alpinePath(strings.TrimPrefix(path.Clean("/"+reqPath), "/"))
if !ok {
return nil, "", ErrNotFound
}
repo, err := e.mergedAlpine(ctx, virt, dir)
if err != nil {
return nil, "", err
}
if file != "APKINDEX.tar.gz" {
return nil, "", ErrNotFound
}
return repo.index(), "application/gzip", nil
}
// alpineRedirect finds the member whose index lists the requested .apk; apk
// builds the download path itself, so the index cannot carry a member prefix.
// A failed merge returns ok=false so Fetch re-merges and fails with 502.
func (e *Engine) alpineRedirect(ctx context.Context, virt models.Virtual, reqPath, rawQuery, proxyBaseURL string) (string, bool, error) {
dir, file, ok := alpinePath(strings.TrimPrefix(path.Clean("/"+reqPath), "/"))
if !ok || file == "APKINDEX.tar.gz" {
return "", false, nil
}
owner, ok := e.cachedFile("alpine\x00"+virt.Name+"\x00"+dir, file)
if !ok {
repo, err := e.mergedAlpine(ctx, virt, dir)
if err != nil {
return "", false, nil
}
if owner, ok = repo.files()[file]; !ok {
return "", false, nil
}
}
return e.memberLocation(ctx, string(owner), dir+file, rawQuery, proxyBaseURL)
}
// fetchAlpineMember reads a member's APKINDEX under dir; a member without that
// directory contributes nothing.
func (e *Engine) fetchAlpineMember(ctx context.Context, name, dir string) (*AlpineMember, error) {
remote, err := e.getRemote(ctx, name)
if err != nil {
return nil, fmt.Errorf("remote %q: %w", name, err)
}
raw, err := e.fetchMemberPath(ctx, *remote, dir+"APKINDEX.tar.gz")
var pe *proxy.ProxyError
if errors.Is(err, ErrNotFound) || errors.As(err, &pe) && pe.Status == http.StatusNotFound {
return &AlpineMember{RemoteName: name}, nil
}
if err != nil {
return nil, err
}
idx, err := apkIndexText(raw)
if err != nil {
return nil, fmt.Errorf("%s/%sAPKINDEX.tar.gz: %w", name, dir, err)
}
return &AlpineMember{RemoteName: name, Index: idx}, nil
}
// evictLRU drops the least recently used merges under prefix beyond keep.
func (e *Engine) evictLRU(prefix string, keep int) {
e.mu.Lock()