Harden worker federation and operator UI

This commit is contained in:
2026-07-29 13:30:55 +04:00
parent 95a96d87a5
commit 1ca9d64e89
35 changed files with 1195 additions and 581 deletions
+87
View File
@@ -1,6 +1,8 @@
package federation
import (
"os"
"path/filepath"
"testing"
"time"
)
@@ -27,6 +29,68 @@ func TestCursorIsMonotonicAndAuthenticationIsRequired(t *testing.T) {
}
}
func TestPendingApprovalSurvivesRegistryRestart(t *testing.T) {
path := filepath.Join(t.TempDir(), "federation-state.json")
r := &Registry{StatePath: path}
if err := r.Load(); err != nil {
t.Fatal(err)
}
if err := r.Register(Worker{ID: "w", Token: "t"}, ""); err != nil {
t.Fatal(err)
}
capture, err := r.PutCapture("w", Capture{TaskID: "task", PaneID: "pane", Text: "Allow command?"})
if err != nil {
t.Fatal(err)
}
queued, err := r.Queue("w", Command{TaskID: "task", Kind: "grant_approval", PaneID: "pane", CaptureRevision: capture.Revision})
if err != nil {
t.Fatal(err)
}
info, err := os.Stat(path)
if err != nil {
t.Fatal(err)
}
if info.Mode().Perm() != 0600 {
t.Fatalf("federation state permissions = %o, want 0600", info.Mode().Perm())
}
restarted := &Registry{StatePath: path}
if err := restarted.Load(); err != nil {
t.Fatal(err)
}
if err := restarted.Register(Worker{ID: "w", Token: "intruder"}, ""); err != ErrUnauthorized {
t.Fatalf("recovered worker identity was hijackable: %v", err)
}
// A restart does not mark the worker online; it must prove its retained
// identity by registering again before recovered controls become available.
if err := restarted.Register(Worker{ID: "w", Token: "t"}, ""); err != nil {
t.Fatal(err)
}
gotCapture, ok := restarted.Capture("w", "task")
if !ok || gotCapture.Revision != capture.Revision || gotCapture.Text != capture.Text {
t.Fatalf("capture after restart = %#v, present=%v", gotCapture, ok)
}
commands, err := restarted.Commands("w")
if err != nil || len(commands) != 1 || commands[0].ID != queued.ID {
t.Fatalf("commands after restart = %#v, err=%v", commands, err)
}
if err := restarted.CompleteCommand("w", queued.ID, "acknowledged", ""); err != nil {
t.Fatal(err)
}
again := &Registry{StatePath: path}
if err := again.Load(); err != nil {
t.Fatal(err)
}
if err := again.Register(Worker{ID: "w", Token: "t"}, ""); err != nil {
t.Fatal(err)
}
commands, err = again.Commands("w")
if err != nil || len(commands) != 0 {
t.Fatalf("resolved command recovered as pending: %#v, err=%v", commands, err)
}
}
func TestRegisterRequiresAdmitTokenAndOwnToken(t *testing.T) {
r := &Registry{AdmitToken: "admit-secret"}
if err := r.Register(Worker{ID: "workpc", Token: "secret"}, "wrong"); err != ErrUnauthorized {
@@ -76,6 +140,29 @@ func TestOfflineHookRunsOnceOnTransition(t *testing.T) {
}
}
func TestHeartbeatProjectsWorkerOwnedHealth(t *testing.T) {
r := &Registry{}
if err := r.Register(Worker{ID: "workpc-opencode", Token: "secret"}, ""); err != nil {
t.Fatal(err)
}
checked := time.Now().UTC().Round(0)
errAt := checked.Add(-time.Minute)
if err := r.Heartbeat("workpc-opencode", WorkerHealth{
HerdrStatus: "unreachable", CheckedAt: checked, ActiveTask: "task-1", ActivePane: "pane-1",
LastError: "local herdr: connection refused", ErrorAt: errAt,
}); err != nil {
t.Fatal(err)
}
workers := r.Snapshot()
if len(workers) != 1 {
t.Fatalf("workers=%#v", workers)
}
h := workers[0].Health
if h.HerdrStatus != "unreachable" || h.ActiveTask != "task-1" || h.ActivePane != "pane-1" || h.LastError == "" || !h.CheckedAt.Equal(checked) || !h.ErrorAt.Equal(errAt) {
t.Fatalf("health=%#v", h)
}
}
func TestCaptureRevisionAndCommandQueue(t *testing.T) {
r := &Registry{}
if err := r.Register(Worker{ID: "w", Token: "t"}, ""); err != nil {