Add ansible virtual repositories (#158)
ci/woodpecker/tag/docker Pipeline was successful

Clients need one Galaxy server to install both internal collections and galaxy.ansible.com ones; ansible has no virtual type.

- merge member collection version lists in member order, the earlier member winning a duplicate version, cached with the shared virtual merge cache
- paginate limit/offset over the merged list, following remote member pages
- serve version detail from the owning member with href and download_url via the virtual; downloads 302 to the member through the member-redirect validation
- fail closed with 502 on a member error or a member versions list shorter than its count
- add docs/ansible/virtual.md and an e2e-docker case driving real ansible-galaxy

requires unkin/artifactapi#157

Reviewed-on: #158
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 #158.
This commit is contained in:
2026-10-18 18:09:08 +11:00
committed by BenVincent
parent a00e45c3a7
commit 0cfde41612
11 changed files with 622 additions and 6 deletions
+182
View File
@@ -0,0 +1,182 @@
package virtual
import (
"context"
"encoding/json"
"errors"
"fmt"
"net/http"
"net/url"
"strconv"
"strings"
"git.unkin.net/unkin/artifactapi/internal/provider/ansible"
"git.unkin.net/unkin/artifactapi/internal/proxy"
"git.unkin.net/unkin/artifactapi/pkg/models"
)
const (
maxAnsibleCollections = 1024
ansibleMemberPage = 100
maxAnsibleMemberPages = 100
)
// AnsibleMember is one member's versions of a collection; Found is false when
// the member does not have it.
type AnsibleMember struct {
RemoteName string
Found bool
Refs []ansible.VersionRef
}
// AnsibleDoc is a collection's merged versions; Owners maps each version to
// the member serving it.
type AnsibleDoc struct {
Refs []ansible.VersionRef
Owners map[string]string
body []byte
}
func (d *AnsibleDoc) index() []byte { return d.body }
func (d *AnsibleDoc) files() map[string][]byte { return nil }
// MergeAnsibleVersions concatenates member versions in member order; the
// first member wins a duplicate version. ErrNotFound means no member has the
// collection.
func MergeAnsibleVersions(members []AnsibleMember) (*AnsibleDoc, error) {
d := &AnsibleDoc{Refs: []ansible.VersionRef{}, Owners: map[string]string{}}
found := false
for _, m := range members {
found = found || m.Found
for _, r := range m.Refs {
if _, dup := d.Owners[r.Version]; dup {
continue
}
d.Owners[r.Version] = m.RemoteName
d.Refs = append(d.Refs, r)
}
}
if !found {
return nil, ErrNotFound
}
var err error
d.body, err = json.Marshal(d.Refs)
return d, err
}
// AnsibleVersions returns the merged versions of ns.name across members.
func (e *Engine) AnsibleVersions(ctx context.Context, virt models.Virtual, ns, name string) (*AnsibleDoc, error) {
prefix := "ansible\x00" + virt.Name + "\x00"
inserted := false
repo, err := e.merged(ctx, prefix+ns+"/"+name, func(ctx context.Context) (mergedRepo, error) {
inserted = true
members, err := fetchMembers(virt, func(m string) (*AnsibleMember, error) { return e.ansibleMember(ctx, m, ns, name) })
if err != nil {
return nil, err
}
return MergeAnsibleVersions(members)
})
if err != nil {
return nil, err
}
if inserted {
e.evictOldest(prefix, maxAnsibleCollections)
}
return repo.(*AnsibleDoc), nil
}
// AnsibleVersion serves the owning member's version detail with href and
// download_url pointing at the virtual; the download 302s to the member.
// Member documents carry root-relative URLs, so anything else is rejected.
func (e *Engine) AnsibleVersion(ctx context.Context, virt models.Virtual, ns, name, version, proxyBaseURL string) ([]byte, error) {
d, err := e.AnsibleVersions(ctx, virt, ns, name)
if err != nil {
return nil, err
}
owner, ok := d.Owners[version]
if !ok {
return nil, ErrNotFound
}
remote, err := e.getRemote(ctx, owner)
if err != nil {
return nil, fmt.Errorf("remote %q: %w", owner, err)
}
body, err := e.ansibleDoc(ctx, *remote, ansible.VersionPath(ns, name, version)[1:])
if err != nil {
return nil, fmt.Errorf("member %q: %v", owner, err)
}
var detail map[string]any
if err := json.Unmarshal(body, &detail); err != nil {
return nil, fmt.Errorf("member %q version detail: %v", owner, err)
}
dl, _ := detail["download_url"].(string)
u, err := url.Parse(dl)
if err != nil {
return nil, fmt.Errorf("member %q download_url %q: %v", owner, dl, err)
}
rest, ok := strings.CutPrefix(u.EscapedPath(), "/api/v1/"+string(remote.RepoType)+"/"+url.PathEscape(owner)+"/")
if !ok || u.Scheme != "" || u.Host != "" {
return nil, fmt.Errorf("member %q download_url %q is not served by the member", owner, dl)
}
base := strings.TrimRight(proxyBaseURL, "/") + "/api/v1/virtual/" + url.PathEscape(virt.Name)
detail["download_url"] = base + "/" + url.PathEscape(owner) + "/" + rest
detail["href"] = base + ansible.VersionPath(ns, name, version)
return json.Marshal(detail)
}
// ansibleMember reads every version a member lists for ns.name, following
// the member's pagination for remotes.
func (e *Engine) ansibleMember(ctx context.Context, member, ns, name string) (*AnsibleMember, error) {
remote, err := e.getRemote(ctx, member)
if err != nil {
return nil, fmt.Errorf("remote %q: %w", member, err)
}
m := &AnsibleMember{RemoteName: member}
list := ansible.CollectionPath(ns, name)[1:] + "versions/"
for page := 0; page < maxAnsibleMemberPages; page++ {
// Locals list every version at once; remote pages match the keys the
// remote caches client requests under.
path := list
if remote.RepoType != models.RepoTypeLocal {
path += "?limit=" + strconv.Itoa(ansibleMemberPage)
if page > 0 {
path += "&offset=" + strconv.Itoa(len(m.Refs))
}
}
body, err := e.ansibleDoc(ctx, *remote, path)
var pe *proxy.ProxyError
if errors.Is(err, ErrNotFound) || errors.As(err, &pe) && pe.Status == http.StatusNotFound {
return m, nil
}
if err != nil {
return nil, err
}
var doc struct {
Meta struct{ Count int }
Data []ansible.VersionRef
}
if err := json.Unmarshal(body, &doc); err != nil {
return nil, fmt.Errorf("%s/%s: %v", member, path, err)
}
m.Found = true
m.Refs = append(m.Refs, doc.Data...)
if remote.RepoType == models.RepoTypeLocal || len(m.Refs) >= doc.Meta.Count {
return m, nil
}
if len(doc.Data) == 0 {
return nil, fmt.Errorf("%s/%s: empty page after %d of %d versions", member, path, len(m.Refs), doc.Meta.Count)
}
}
return nil, fmt.Errorf("%s: more than %d pages of %s.%s versions", member, maxAnsibleMemberPages, ns, name)
}
// ansibleMemberDoc reads a Galaxy API document from a member.
func (e *Engine) ansibleMemberDoc(ctx context.Context, remote models.Remote, path string) ([]byte, error) {
if remote.RepoType != models.RepoTypeLocal {
return e.fetchMemberPath(ctx, remote, path)
}
if e.AnsibleLocal == nil {
return nil, errors.New("local ansible indexes are not wired")
}
return e.AnsibleLocal(ctx, remote.Name, path)
}
+206
View File
@@ -0,0 +1,206 @@
package virtual
import (
"context"
"encoding/json"
"errors"
"fmt"
"net/http"
"strings"
"sync"
"testing"
"time"
"git.unkin.net/unkin/artifactapi/internal/provider/ansible"
"git.unkin.net/unkin/artifactapi/internal/proxy"
"git.unkin.net/unkin/artifactapi/pkg/models"
)
func refs(vs ...string) []ansible.VersionRef {
out := make([]ansible.VersionRef, len(vs))
for i, v := range vs {
out[i] = ansible.VersionRef{Version: v}
}
return out
}
func TestMergeAnsibleVersionsMemberOrderWins(t *testing.T) {
d, err := MergeAnsibleVersions([]AnsibleMember{
{RemoteName: "internal", Found: true, Refs: refs("2.0.0", "1.0.0")},
{RemoteName: "missing"},
{RemoteName: "galaxy", Found: true, Refs: refs("3.0.0", "2.0.0", "0.9.0")},
})
if err != nil {
t.Fatal(err)
}
var got []string
for _, r := range d.Refs {
got = append(got, r.Version)
}
if strings.Join(got, ",") != "2.0.0,1.0.0,3.0.0,0.9.0" {
t.Fatalf("merged = %v", got)
}
for v, owner := range map[string]string{"2.0.0": "internal", "1.0.0": "internal", "3.0.0": "galaxy", "0.9.0": "galaxy"} {
if d.Owners[v] != owner {
t.Errorf("owner of %s = %q, want %q", v, d.Owners[v], owner)
}
}
if _, err := MergeAnsibleVersions([]AnsibleMember{{RemoteName: "a"}, {RemoteName: "b"}}); !errors.Is(err, ErrNotFound) {
t.Fatalf("no member has it: want ErrNotFound, got %v", err)
}
if d, err := MergeAnsibleVersions([]AnsibleMember{{RemoteName: "a", Found: true}}); err != nil || len(d.Refs) != 0 {
t.Fatalf("found with no versions = %v %v", d, err)
}
}
// galaxyPages answers a remote's versions list like Galaxy, two per page.
func galaxyPages(versions []string, path string) ([]byte, error) {
_, q, _ := strings.Cut(path, "?")
offset := 0
if _, off, ok := strings.Cut(q, "offset="); ok {
_, _ = fmt.Sscan(off, &offset)
}
end := min(offset+2, len(versions))
return json.Marshal(map[string]any{"meta": map[string]int{"count": len(versions)}, "data": refs(versions[min(offset, end):end]...)})
}
func ansibleEngine(doc func(remote models.Remote, path string) ([]byte, error)) *Engine {
return &Engine{
mergeTTL: time.Minute,
getRemote: func(_ context.Context, name string) (*models.Remote, error) {
rt := models.RepoTypeRemote
if name == "internal" {
rt = models.RepoTypeLocal
}
return &models.Remote{Name: name, RepoType: rt}, nil
},
ansibleDoc: func(_ context.Context, remote models.Remote, path string) ([]byte, error) { return doc(remote, path) },
}
}
func ansibleVirt(members ...string) models.Virtual {
return models.Virtual{Name: "ansible", PackageType: models.PackageAnsible, Members: members}
}
func TestAnsibleVersionsFollowsRemotePages(t *testing.T) {
var mu sync.Mutex
var asked []string
e := ansibleEngine(func(remote models.Remote, path string) ([]byte, error) {
mu.Lock()
asked = append(asked, remote.Name+" "+path)
mu.Unlock()
if remote.Name == "internal" {
return galaxyPages([]string{"1.0.0", "5.0.0"}, "")
}
return galaxyPages([]string{"5.0.0", "4.0.0", "3.0.0", "2.0.0", "1.5.0"}, path)
})
virt := ansibleVirt("internal", "galaxy")
d, err := e.AnsibleVersions(context.Background(), virt, "community", "general")
if err != nil {
t.Fatal(err)
}
if len(d.Refs) != 6 || d.Owners["5.0.0"] != "internal" || d.Owners["1.5.0"] != "galaxy" {
t.Fatalf("merged = %+v", d)
}
want := map[string]bool{
"internal api/v3/collections/community/general/versions/": true,
"galaxy api/v3/collections/community/general/versions/?limit=100": true,
"galaxy api/v3/collections/community/general/versions/?limit=100&offset=2": true,
"galaxy api/v3/collections/community/general/versions/?limit=100&offset=4": true,
}
if len(asked) != len(want) {
t.Fatalf("asked %v", asked)
}
for _, a := range asked {
if !want[a] {
t.Errorf("unexpected fetch %q", a)
}
}
if _, err := e.AnsibleVersions(context.Background(), virt, "community", "general"); err != nil || len(asked) != len(want) {
t.Fatalf("merge not cached: %v, %d fetches", err, len(asked))
}
}
func TestAnsibleVersionsFailsClosed(t *testing.T) {
e := ansibleEngine(func(remote models.Remote, path string) ([]byte, error) {
switch remote.Name {
case "internal":
return nil, ErrNotFound
case "gone":
return nil, &proxy.ProxyError{Status: http.StatusNotFound}
}
return nil, errors.New("upstream down")
})
if _, err := e.AnsibleVersions(context.Background(), ansibleVirt("internal", "gone"), "a", "b"); !errors.Is(err, ErrNotFound) {
t.Fatalf("all members lack it: want ErrNotFound, got %v", err)
}
if _, err := e.AnsibleVersions(context.Background(), ansibleVirt("internal", "galaxy"), "a", "b"); err == nil || errors.Is(err, ErrNotFound) {
t.Fatalf("member error: want a non-404 error, got %v", err)
}
}
func TestAnsibleVersionsShortRemote(t *testing.T) {
e := ansibleEngine(func(_ models.Remote, path string) ([]byte, error) {
if strings.Contains(path, "offset=") {
return json.Marshal(map[string]any{"meta": map[string]int{"count": 3}, "data": []any{}})
}
return galaxyPages([]string{"3.0.0", "2.0.0", "1.0.0"}, path)
})
if _, err := e.AnsibleVersions(context.Background(), ansibleVirt("galaxy"), "a", "b"); err == nil || errors.Is(err, ErrNotFound) {
t.Fatalf("empty page before count: want a non-404 error, got %v", err)
}
}
func TestAnsibleVersionRoutesViaVirtual(t *testing.T) {
detail := func(dl string) []byte {
b, _ := json.Marshal(map[string]any{"version": "1.0.0", "download_url": dl, "artifact": map[string]string{"sha256": "ab"}})
return b
}
dls := map[string]string{
"internal": "/api/v1/local/internal/artifacts/a-b-1.0.0.tar.gz",
"galaxy": "/api/v1/remote/galaxy/api/v3/plugin/ansible/content/published/collections/artifacts/a-b-2.0.0.tar.gz",
"evil": "https://evil.example/api/v1/remote/evil/x.tar.gz",
"other": "/api/v1/remote/someoneelse/x.tar.gz",
}
e := ansibleEngine(func(remote models.Remote, path string) ([]byte, error) {
if strings.Contains(path, "/versions/?") || strings.HasSuffix(path, "/versions/") {
v := map[string]string{"internal": "1.0.0", "galaxy": "2.0.0", "evil": "3.0.0", "other": "4.0.0"}[remote.Name]
return galaxyPages([]string{v}, path)
}
return detail(dls[remote.Name]), nil
})
virt := ansibleVirt("internal", "galaxy", "evil", "other")
for v, want := range map[string]string{
"1.0.0": "https://h/api/v1/virtual/ansible/internal/artifacts/a-b-1.0.0.tar.gz",
"2.0.0": "https://h/api/v1/virtual/ansible/galaxy/api/v3/plugin/ansible/content/published/collections/artifacts/a-b-2.0.0.tar.gz",
} {
body, err := e.AnsibleVersion(context.Background(), virt, "a", "b", v, "https://h")
if err != nil {
t.Fatal(err)
}
var got struct {
Href string
DownloadURL string `json:"download_url"`
Artifact struct{ Sha256 string }
}
_ = json.Unmarshal(body, &got)
if got.DownloadURL != want || got.Href != "https://h/api/v1/virtual/ansible/api/v3/collections/a/b/versions/"+v+"/" || got.Artifact.Sha256 != "ab" {
t.Errorf("%s detail = %+v", v, got)
}
loc, ok, err := e.MemberRedirect(context.Background(), virt, strings.TrimPrefix(want, "https://h/api/v1/virtual/ansible/"), "", "https://h")
if err != nil || !ok || loc != "https://h"+dls[map[string]string{"1.0.0": "internal", "2.0.0": "galaxy"}[v]] {
t.Errorf("%s redirect = %q %v %v", v, loc, ok, err)
}
}
for _, v := range []string{"3.0.0", "4.0.0"} {
if _, err := e.AnsibleVersion(context.Background(), virt, "a", "b", v, "https://h"); err == nil || errors.Is(err, ErrNotFound) {
t.Errorf("%s: foreign download_url accepted: %v", v, err)
}
}
if _, err := e.AnsibleVersion(context.Background(), virt, "a", "b", "9.9.9", "https://h"); !errors.Is(err, ErrNotFound) {
t.Errorf("unknown version: want ErrNotFound, got %v", err)
}
if _, _, err := e.MemberRedirect(context.Background(), virt, "galaxy/../internal/x", "", "https://h"); !errors.Is(err, ErrBadPath) {
t.Errorf("escaping redirect: want ErrBadPath, got %v", err)
}
}
+11 -4
View File
@@ -38,6 +38,11 @@ type Engine struct {
// ErrNotFound.
GemsLocal func(ctx context.Context, repo, path string) ([]byte, error)
// AnsibleLocal returns a local ansible repo's collection, full versions list
// or version detail with root-relative URLs, or ErrNotFound.
AnsibleLocal func(ctx context.Context, repo, path string) ([]byte, error)
ansibleDoc func(ctx context.Context, remote models.Remote, path string) ([]byte, error)
mergeTTL time.Duration
sf singleflight.Group
@@ -68,6 +73,7 @@ func NewEngine(db *database.DB, proxyEngine *proxy.Engine) *Engine {
e.rpmMember = e.fetchRPMMember
e.npmMember = e.fetchNPMMember
e.gemsMember = e.fetchGemsMember
e.ansibleDoc = e.ansibleMemberDoc
return e
}
@@ -192,16 +198,17 @@ var (
ErrBadPath = errors.New("invalid path")
)
// MemberRedirect maps an rpm or npm virtual package path (<member>/<href>, as
// written by MergeRPM/MergeNPM) to the owning member's route. ok is false for
// any other path; ErrBadPath means the path is absolute or escapes the member.
// MemberRedirect maps an rpm, npm or ansible virtual package path
// (<member>/<href>, as written by MergeRPM/MergeNPM/AnsibleVersion) 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 {
case models.PackageRPM:
if strings.HasPrefix(path, "repodata/") {
return "", false, nil
}
case models.PackageNPM:
case models.PackageNPM, models.PackageAnsible:
default:
return "", false, nil
}