hub v0.159.0: a refused whole-guest backup has its own alarm (R-1150); a reinstall re-issues the claim code and the off-site login when the old host is stale or off (R-1154)
R-1150: the agent's localapi:backup-cadence-cap is a separate per-host state in HostCapabilityChecker: alone it raises whole_guest_backup_refused (warning, plain text, agent reason in details) and whole_guest_backup_refusal_cleared (info); other critical caps keep agent_capability_degraded/_recovered unchanged. New types operator-only and hub-origin only (not in allowedEventTypes: the suspected controller must not post them). agent_capability_degraded/_recovered join operatorOnlyEvents (the checker's "never the customer" claim was configuration only). R-1154: handleHostEnroll's reuse path re-issues (claim code + off-site login) when no controller wait is open and either deliberate_off_at is held or both the host report and the controller report are older than alerting.stale_threshold (wired from main); once per silence (host_reenroll_reissued record, also written at mint); a live box gets nothing re-issued and the skip is recorded (host_reenroll_not_reissued). 05 section 8 [DESIGN]. Tests red-proved; go build/vet/test green.
This commit is contained in:
@@ -222,6 +222,25 @@ Today: `GET /config/{id}` with an `X-Retrieval-Password` (Hungarian passphrase)
|
||||
The existing `configgen` deep-merge + Hungarian-passphrase machinery is the base; it grows the
|
||||
agent-first + key-pinning + restore-mode steps.
|
||||
|
||||
**A reinstall over an existing host record re-issues when the old box is gone [DESIGN, 2026-10-19 — R-1154, brief
|
||||
2026-10-19; hub v0.159.0].** `POST /host-enroll` is the installer's step 4, run on every pass (a `--resume` too), with
|
||||
the customer's passphrase; an existing host record is REUSED (same id and key — the reinstalled box takes it over).
|
||||
The clean-slate re-issues (the claim code, the off-site login — both live in the CONTROLLER) used to run only when a
|
||||
record was minted, so a reinstall that kept the old record skipped both in silence. The reuse path now runs them
|
||||
when, and only when, the old box is gone: **no controller wait connection open now**, and either the household
|
||||
switched the box off on purpose (`deliberate_off_at` inside its hold) or **both** the agent's host report and the
|
||||
controller's report are older than `alerting.stale_threshold` (or absent) — the threshold and comparison
|
||||
`hostStatus` uses, no second definition. **Once per silence:** not again when it already ran since the host's last
|
||||
report (an installer `--resume` before the new agent's first report). **Why this rule and not "re-issue on every
|
||||
reuse":** a live box re-enrolls legitimately (a `--resume` after its agent started, a repair re-run), and re-issuing
|
||||
there rotates the household's claim code and resets its off-site login under a running box; not re-issuing costs a
|
||||
reinstall two operator presses, which the skip record now names. The controller report counts because the codes
|
||||
live there: a box whose agent is broken but whose controller reports is live. **Known residual:** a reinstall done
|
||||
less than the threshold after the old box's last report is still treated as live — it is recorded
|
||||
(`host_reenroll_not_reissued`, operator-only, info) instead of silent, and the operator presses the two buttons or
|
||||
deletes the old host first. The re-issue writes `host_reenroll_reissued` (also the once-per-silence record); both
|
||||
types are hub-origin only. `hub/internal/api/reenroll.go`, pinned by `hub/internal/api/r1154_reenroll_test.go`.
|
||||
|
||||
## 9. DR model
|
||||
|
||||
The headline: the **old heavy infra-backup push retires** — not because the hub authors everything
|
||||
|
||||
@@ -1,3 +1,41 @@
|
||||
## v0.159.0 — a refused whole-guest backup has its own alarm text (R-1150); a reinstall re-issues the claim code and the off-site login when the old host is stale or off (R-1154) (2026-10-19)
|
||||
|
||||
- **`whole_guest_backup_refused`** (R-1150). The agent's R-973 cadence cap reports a CRITICAL capability
|
||||
`localapi:backup-cadence-cap` as degraded for 24 h after it refused a whole-guest backup a guest asked for beyond its
|
||||
schedule; the hub mailed that as `agent_capability_degraded` „agent privileged capability degraded" — read as a
|
||||
broken sudo grant. `HostCapabilityChecker` now keeps it as a SECOND, independent state per host: the cadence cap
|
||||
alone raises `whole_guest_backup_refused` (warning) „Host <id>: a whole-guest backup was refused — the guest asked
|
||||
for more whole-guest backups than its schedule allows. Its controller may be misbehaving or compromised: check the
|
||||
guest.", the agent's reason (guest, target, time) in the details; its end is `whole_guest_backup_refusal_cleared`
|
||||
(info: recorded, never mailed). Every other critical capability keeps `agent_capability_degraded` /
|
||||
`_recovered` with unchanged text and never names the cadence cap; both on one report → one event per state, never
|
||||
two for one; a grant restored while the refusal stands → `agent_capability_recovered` alone. The agent is unchanged
|
||||
(its status name stays). Both new types are operator-only and hub-origin only — **deliberately NOT in
|
||||
`allowedEventTypes`** (the brief suggested it): the alarm suspects the guest's controller, which holds the
|
||||
`/event` credential, and a type it can post it can fake or post first to spend the operator cooldown (the R-895
|
||||
reasoning, `TestR1150_RefusedBackupEventsAreOperatorOnlyAndHubOrigin`). **Also:** `agent_capability_degraded` /
|
||||
`_recovered` enter `operatorOnlyEvents` — the checker has said „Customer is NEVER notified" since v0.44.0 and only
|
||||
`enabled_events` configuration made it true (the R-97c shape). Tests `TestR1150_CadenceCapAloneRaisesWholeGuestBackupRefused`,
|
||||
`TestR1150_OtherCapStillRaisesCapabilityDegraded`, `TestR1150_BornRefusedAlertsOnce` (red-proved: the pre-fix
|
||||
shape emits `agent_capability_degraded` instead), `TestR1150_RefusedBackupAndCapabilityEventsNeverReachTheCustomer`
|
||||
(red-proved: without the register lines the household is mailed).
|
||||
- **A reinstall over the old host record re-issues** (R-1154; `05` §8 [DESIGN]). `handleHostEnroll` reused an
|
||||
existing host record and returned at once, so `reissueOnReenroll` (claim code + off-site login) ran only on a mint —
|
||||
a reinstall that kept the old record skipped both in silence (demo-hp 2026-08-21; Tester4's standing reinstalls).
|
||||
The reuse path (same host id and key, unchanged) now re-issues when the old box is gone: no controller wait
|
||||
connection open, and either `deliberate_off_at` inside its hold or BOTH the host report and the controller report
|
||||
older than `alerting.stale_threshold` (or absent) — the `hostStatus` threshold, wired from main
|
||||
(`SetStaleThreshold`; unwired = never re-issues). Once per silence (an installer `--resume` before the new agent
|
||||
reports does not re-issue again): the record is the info event `host_reenroll_reissued`, written by the mint path
|
||||
too. A live box re-enrolling gets nothing re-issued and the skip is recorded (`host_reenroll_not_reissued`), so a
|
||||
reinstall within the threshold is visible instead of silent. Both types operator-only, hub-origin only.
|
||||
`internal/api/reenroll.go`. Tests `TestR1154_StaleOldHostIsReissued`, `TestR1154_DeliberatelyOffOldHostIsReissued`,
|
||||
`TestR1154_LiveHostIsNotReissued` (host report / controller report / wait connection / connected-while-off),
|
||||
`TestR1154_ResumeDoesNotReissueTwice`, `TestR1154_NeverReportedAndUnwiredEdges`, `TestR1154_EventsAreOperatorOnlyAndHubOrigin`,
|
||||
`TestR1154_MainWiresTheStaleThresholdIntoTheAPI` — each red-proved (the pre-fix early return: „claim generation 1→1,
|
||||
claim mails 0, off-site re-issues 0"; an unconditional re-issue: „claim generation 1→2 … off-site re-issues 1" on a
|
||||
live box; no once-per-silence check: „claim generation 1→3, off-site re-issues 2").
|
||||
|
||||
## v0.158.0 — agent-only vouch, the hub's own snapshot alarms, the hub's own access log (2026-10-17) — built that night; DEPLOYED 2026-10-18 on the operator's yes (12:16)
|
||||
|
||||
- **An agent can be vouched mid-week** (`09` §3 decision 250 items 3–4). The Day-0 save's R-120 golden guard refused
|
||||
|
||||
+8
-2
@@ -148,7 +148,13 @@ Events are the primary monitoring mechanism. Each event has: customer_id, event_
|
||||
- `node_stale` / `node_down` / `node_recovered` — dead man's switch from staleness checker (every 60s)
|
||||
- `host_stale` / `host_down` / `host_recovered` — host-domain dead man's switch (agent host reports)
|
||||
- `agent_capability_degraded` / `agent_capability_recovered` — `HostCapabilityChecker`: an agent's
|
||||
required `sudo -n` grant went missing (the non-root cutover class)
|
||||
required `sudo -n` grant went missing (the non-root cutover class). Operator-only since v0.159.0.
|
||||
- `whole_guest_backup_refused` (warning) / `whole_guest_backup_refusal_cleared` (info) — the same checker, a
|
||||
SEPARATE state (R-1150, v0.159.0): the agent's `localapi:backup-cadence-cap` reports that it refused a whole-guest
|
||||
backup a guest asked for beyond its schedule (R-973). Operator-only; hub-origin only (not POST-able by a box).
|
||||
- `host_reenroll_reissued` / `host_reenroll_not_reissued` (info) — `POST /host-enroll` reused an existing host
|
||||
record and re-issued the claim code and the off-site login (the old box is gone), or did not (the box is live)
|
||||
(R-1154, v0.159.0). Operator-only; hub-origin only.
|
||||
- `host_leaf_changed` — `HostLeafChecker` (`monitor/host_leaf.go`, every 60s): an agent's reported
|
||||
local-API **leaf fingerprint** changed (an agent re-key) — proactive, fleet-wide, independent of any
|
||||
controller's channel-health check. Trust-on-first-report baseline; empty fp = unknown (never alerts).
|
||||
@@ -181,7 +187,7 @@ Config retrieval uses a separate per-customer retrieval password (not the API ke
|
||||
|--------|------|-------------|
|
||||
| `POST` | `/api/v1/host-enroll` | Mint-or-reuse a host credential, body `{customer_id}` (auth: `X-Retrieval-Password` header) → `{host_id, api_key}` |
|
||||
|
||||
The Day-0 first-boot handshake (`documentation/audits/SPIKE-day0-firstboot-handshake-2026-06-26.md`) enrolls a Proxmox host's agent **using the customer's retrieval passphrase only** — the operator-tier global key never enters the field deploy path. `host-enroll` mints the per-host credential on the first call and **reuses it byte-for-byte** on every later call (idempotent — re-running the host-bootstrap never orphans a running agent's key), backed by `Store.GetHostByCustomer`. Auth is checked **before** any mint, so a wrong passphrase never creates a host row. Behaviour: first → `201`, reuse → `200`, wrong/missing passphrase → `401`, unknown customer → `404`, missing `customer_id` → `400`.
|
||||
The Day-0 first-boot handshake (`documentation/audits/SPIKE-day0-firstboot-handshake-2026-06-26.md`) enrolls a Proxmox host's agent **using the customer's retrieval passphrase only** — the operator-tier global key never enters the field deploy path. `host-enroll` mints the per-host credential on the first call and **reuses it byte-for-byte** on every later call (idempotent — re-running the host-bootstrap never orphans a running agent's key), backed by `Store.GetHostByCustomer`. Auth is checked **before** any mint, so a wrong passphrase never creates a host row. Behaviour: first → `201`, reuse → `200`, wrong/missing passphrase → `401`, unknown customer → `404`, missing `customer_id` → `400`. **Reinstall (R-1154, v0.159.0):** on a reuse whose old box is gone — its host report AND its controller report both older than `alerting.stale_threshold` (or absent), or the household switched it off on purpose, and no controller wait connection open — the enroll also runs the clean-slate re-issues (claim code, off-site login), once per silence; a live box re-enrolling gets nothing re-issued (`internal/api/reenroll.go`, `05` §8).
|
||||
|
||||
The global-key `POST /api/v1/admin/hosts` (operator/HQ pre-mint) remains as the escape hatch pending the enrollment-cutover lock-down; it and `GET /config/{id}` are unchanged by this endpoint.
|
||||
|
||||
|
||||
@@ -337,6 +337,9 @@ func main() {
|
||||
}
|
||||
apiHandler := api.New(dataStore, cfg.API.ReportAPIKey, cfg.Notifications.ResendAPIKey, cfg.Notifications.FromEmail, templateProvider, logger)
|
||||
apiHandler.SetAssetManager(assetsMgr)
|
||||
// R-1154: the host-enroll reuse path judges "the old box is gone" by the same threshold as hostStatus and the
|
||||
// staleness checkers (unwired = it never re-issues). Pinned by r1154_wiring_test.go.
|
||||
apiHandler.SetStaleThreshold(staleThreshold)
|
||||
|
||||
// Initialize notification dispatcher
|
||||
dispatcher := notify.NewDispatcher(
|
||||
|
||||
@@ -0,0 +1,41 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"go/ast"
|
||||
"go/parser"
|
||||
"go/token"
|
||||
"testing"
|
||||
)
|
||||
|
||||
// R-1154 seam wiring: main() must hand the api handler the SAME stale threshold the staleness checkers get
|
||||
// (apiHandler.SetStaleThreshold(staleThreshold)). Unwired, the host-enroll reuse path never re-issues the claim
|
||||
// code and the off-site login on a reinstall — the R-1154 defect, back in silence.
|
||||
//
|
||||
// COMPANION RED-PROOF (observed 2026-10-19): with the call deleted from main.go this fails with
|
||||
// "cmd/hub/main.go never calls apiHandler.SetStaleThreshold(staleThreshold)". Restored.
|
||||
func TestR1154_MainWiresTheStaleThresholdIntoTheAPI(t *testing.T) {
|
||||
f, err := parser.ParseFile(token.NewFileSet(), "main.go", nil, 0)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
wired := false
|
||||
ast.Inspect(f, func(n ast.Node) bool {
|
||||
c, ok := n.(*ast.CallExpr)
|
||||
if !ok {
|
||||
return true
|
||||
}
|
||||
sel, ok := c.Fun.(*ast.SelectorExpr)
|
||||
if !ok || sel.Sel.Name != "SetStaleThreshold" || len(c.Args) != 1 {
|
||||
return true
|
||||
}
|
||||
recv, _ := sel.X.(*ast.Ident)
|
||||
arg, _ := c.Args[0].(*ast.Ident)
|
||||
if recv != nil && recv.Name == "apiHandler" && arg != nil && arg.Name == "staleThreshold" {
|
||||
wired = true
|
||||
}
|
||||
return true
|
||||
})
|
||||
if !wired {
|
||||
t.Error("cmd/hub/main.go never calls apiHandler.SetStaleThreshold(staleThreshold)")
|
||||
}
|
||||
}
|
||||
@@ -111,6 +111,17 @@ type Handler struct {
|
||||
// box ticks in seconds instead of ≤15 min. Shared with the web server (same *poke.Notifier). nil
|
||||
// = poke disabled (a no-op; the report cycle still reconciles).
|
||||
poker Poker
|
||||
|
||||
// staleThreshold (R-1154, v0.159.0) is the hub's report-liveness threshold (alerting.stale_threshold — the
|
||||
// same value hostStatus and both staleness checkers read). The host-enroll REUSE path re-issues the
|
||||
// clean-slate credentials only when the old box has been silent past it (reenroll.go). 0 = not wired: the
|
||||
// reuse path then never re-issues (the pre-v0.159.0 behaviour — the fail-safe direction). main wires it.
|
||||
staleThreshold time.Duration
|
||||
}
|
||||
|
||||
// SetStaleThreshold wires the report-liveness threshold the host-enroll reuse path judges by (R-1154).
|
||||
func (h *Handler) SetStaleThreshold(d time.Duration) {
|
||||
h.staleThreshold = d
|
||||
}
|
||||
|
||||
// SetClaimEngine wires the customer-claim code engine (nil-safe everywhere it is used).
|
||||
@@ -1205,7 +1216,10 @@ func (h *Handler) handleHostEnroll(w http.ResponseWriter, r *http.Request) {
|
||||
return
|
||||
}
|
||||
|
||||
// Mint-once-reuse: an existing host for this customer is returned as-is (idempotent).
|
||||
// Mint-once-reuse: an existing host for this customer is returned as-is (idempotent) — the SAME host id and
|
||||
// key, so re-running the installer never orphans a live agent's key and a reinstalled box takes over the
|
||||
// record. R-1154 (v0.159.0): when the old box has gone silent (or was switched off on purpose), this
|
||||
// enroll is a reinstall and the clean-slate re-issues run here too — reenroll.go decides.
|
||||
existing, err := h.store.GetHostByCustomer(req.CustomerID)
|
||||
if err != nil {
|
||||
h.logger.Printf("[ERROR] host-enroll: host lookup failed for %s: %v", req.CustomerID, err)
|
||||
@@ -1219,6 +1233,7 @@ func (h *Handler) handleHostEnroll(w http.ResponseWriter, r *http.Request) {
|
||||
http.Error(w, "Internal error", http.StatusInternalServerError)
|
||||
return
|
||||
}
|
||||
h.reissueOnReuseIfReinstall(cc, existing, time.Now())
|
||||
w.Header().Set("Content-Type", "application/json")
|
||||
w.WriteHeader(http.StatusOK)
|
||||
json.NewEncoder(w).Encode(map[string]string{"host_id": existing.HostID, "api_key": existing.APIKey})
|
||||
@@ -1245,13 +1260,17 @@ func (h *Handler) handleHostEnroll(w http.ResponseWriter, r *http.Request) {
|
||||
h.logger.Printf("[INFO] host enrolled: %s (customer %s)", hostID, req.CustomerID)
|
||||
|
||||
// F2 (v0.57.0) — clean-slate reinstall of an existing customer. This mint path fires exactly
|
||||
// once per fresh host record (the clean-slate flow deletes the stale host, so re-enroll mints),
|
||||
// so it is the natural single-shot re-enroll hook. For a CLAIMED customer the fresh box has no
|
||||
// once per fresh host record (a host deleted before the reinstall), so it is one of the two
|
||||
// single-shot re-enroll hooks — the other is the reuse path above for a silent old host (R-1154).
|
||||
// For a CLAIMED customer the fresh box has no
|
||||
// password; auto-issue a reset code so the customer isn't stranded at the claim page hunting for
|
||||
// the manual "request a new code" button (delivery rides the report ACK). No-op for an unclaimed
|
||||
// customer (first provision). Also re-issue offsite credentials to the fresh box (F3) — the
|
||||
// one-time offsite password only ever reached the OLD controller, so the fresh one has no target.
|
||||
h.reissueOnReenroll(cc)
|
||||
// R-1154: the record that the re-issues ran for this host record — an installer --resume that
|
||||
// enrolls again before the new agent's first report must not re-issue a second time.
|
||||
h.recordReenrollReissued(cc.CustomerID, hostID, "a new host record")
|
||||
|
||||
w.Header().Set("Content-Type", "application/json")
|
||||
w.WriteHeader(http.StatusCreated)
|
||||
|
||||
@@ -0,0 +1,23 @@
|
||||
package api
|
||||
|
||||
import (
|
||||
"testing"
|
||||
|
||||
"gitea.dooplex.hu/admin/felhom-hub/internal/monitor"
|
||||
"gitea.dooplex.hu/admin/felhom-hub/internal/notify"
|
||||
)
|
||||
|
||||
// R-1150 (v0.159.0): the refused whole-guest backup pair is operator-only AND hub-origin only — not in
|
||||
// allowedEventTypes, so no box can POST one to /api/v1/event. The alarm says the guest's controller may be
|
||||
// compromised; that controller holds the /event credential, and a type it could post it could fake, or post
|
||||
// first to spend the operator cooldown and drown the real one (the R-895 reasoning). Both registers in one test.
|
||||
func TestR1150_RefusedBackupEventsAreOperatorOnlyAndHubOrigin(t *testing.T) {
|
||||
for _, et := range []string{monitor.EventWholeGuestBackupRefused, monitor.EventWholeGuestBackupRefusalCleared} {
|
||||
if !notify.IsOperatorOnly(et) {
|
||||
t.Errorf("%s is not operator-only", et)
|
||||
}
|
||||
if allowedEventTypes[et] {
|
||||
t.Errorf("%s is in allowedEventTypes — a box could POST it", et)
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,331 @@
|
||||
package api
|
||||
|
||||
import (
|
||||
"context"
|
||||
"database/sql"
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"net/http"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
"gitea.dooplex.hu/admin/felhom-hub/internal/intent"
|
||||
"gitea.dooplex.hu/admin/felhom-hub/internal/notify"
|
||||
"gitea.dooplex.hu/admin/felhom-hub/internal/store"
|
||||
)
|
||||
|
||||
// R-1154 (v0.159.0) — a reinstall while the customer's OLD host record still exists re-issues the claim code and
|
||||
// the off-site login when the old box is gone (silent past the stale threshold on BOTH report streams, or switched
|
||||
// off on purpose), and NEVER while the box is live. reenroll.go carries the rule; `05` §8 the design.
|
||||
|
||||
const r1154Host = "c1-abc123"
|
||||
|
||||
type r1154Fixture struct {
|
||||
h *Handler
|
||||
st *store.Store
|
||||
db *sql.DB
|
||||
mail *nullMailer
|
||||
offsite int // offsite re-issuer calls
|
||||
gen0 int // the claim generation before the enroll
|
||||
}
|
||||
|
||||
// newR1154 builds a CLAIMED customer c1 with an existing host record that has reported, the hub threshold at 45 m
|
||||
// (the production value), a claim engine and a counting off-site re-issuer.
|
||||
func newR1154(t *testing.T) *r1154Fixture {
|
||||
t.Helper()
|
||||
h, st, path := newTestHandler(t)
|
||||
f := &r1154Fixture{h: h, st: st}
|
||||
cc := &store.CustomerConfig{CustomerID: "c1", APIKey: "ckey", RetrievalPassword: "pass-phrase",
|
||||
Email: "owner@example.hu", Domain: "example.hu", ConfigJSON: "{}"}
|
||||
if err := st.SaveCustomerConfig(cc); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
f.mail = withClaimEngine(t, h, st)
|
||||
if _, err := h.claimEngine.EnsureIssued(cc); err != nil {
|
||||
t.Fatalf("EnsureIssued: %v", err)
|
||||
}
|
||||
if _, err := st.MarkClaimed("c1"); err != nil {
|
||||
t.Fatalf("MarkClaimed: %v", err)
|
||||
}
|
||||
h.SetOffsiteReissuer(func(_ context.Context, id string) error {
|
||||
if id == "c1" {
|
||||
f.offsite++
|
||||
}
|
||||
return nil
|
||||
})
|
||||
h.SetStaleThreshold(45 * time.Minute)
|
||||
if err := st.UpsertHost(&store.Host{HostID: r1154Host, CustomerID: "c1", APIKey: "old-host-key-0123456789"}); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if err := st.SaveHostReport(r1154Host, "c1", []byte(`{"host_id":"`+r1154Host+`"}`), store.HostReportDenorm{}); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
db, err := sql.Open("sqlite", path)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
t.Cleanup(func() { db.Close() })
|
||||
f.db = db
|
||||
f.exec(t, `UPDATE hosts SET created_at = datetime('now', '-30 days') WHERE host_id = ?`, r1154Host)
|
||||
f.mail.sends = 0
|
||||
f.gen0 = f.generation(t)
|
||||
return f
|
||||
}
|
||||
|
||||
func (f *r1154Fixture) exec(t *testing.T, q string, args ...any) {
|
||||
t.Helper()
|
||||
if _, err := f.db.Exec(q, args...); err != nil {
|
||||
t.Fatalf("%s: %v", q, err)
|
||||
}
|
||||
}
|
||||
|
||||
func (f *r1154Fixture) hostReportedAgo(t *testing.T, d time.Duration) {
|
||||
f.exec(t, `UPDATE hosts SET last_report_at = datetime('now', ?) WHERE host_id = ?`, fmt.Sprintf("-%d seconds", int(d.Seconds())), r1154Host)
|
||||
}
|
||||
|
||||
func (f *r1154Fixture) controllerReportedAgo(t *testing.T, d time.Duration) {
|
||||
t.Helper()
|
||||
if err := f.st.SaveReport("c1", []byte(`{"customer_id":"c1"}`)); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
f.exec(t, `UPDATE reports SET received_at = datetime('now', ?) WHERE customer_id = 'c1'`, fmt.Sprintf("-%d seconds", int(d.Seconds())))
|
||||
}
|
||||
|
||||
func (f *r1154Fixture) generation(t *testing.T) int {
|
||||
t.Helper()
|
||||
cs, err := f.st.GetClaim("c1")
|
||||
if err != nil || cs == nil {
|
||||
t.Fatalf("GetClaim: %+v %v", cs, err)
|
||||
}
|
||||
return cs.Generation
|
||||
}
|
||||
|
||||
// enroll runs the installer's call and checks the reuse contract that must hold on EVERY branch: 200, the same
|
||||
// host id and key.
|
||||
func (f *r1154Fixture) enroll(t *testing.T) {
|
||||
t.Helper()
|
||||
rr := doEnroll(f.h, "c1", "pass-phrase")
|
||||
if rr.Code != http.StatusOK {
|
||||
t.Fatalf("enroll = %d, want 200 (reuse), body=%s", rr.Code, rr.Body.String())
|
||||
}
|
||||
var got enrollResp
|
||||
if err := json.Unmarshal(rr.Body.Bytes(), &got); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if got.HostID != r1154Host || got.APIKey != "old-host-key-0123456789" {
|
||||
t.Fatalf("reuse returned another credential: %s", got.HostID)
|
||||
}
|
||||
if n := countHostsForCustomer(t, f.st, "c1"); n != 1 {
|
||||
t.Fatalf("host rows = %d, want 1 (the record is reused, never duplicated)", n)
|
||||
}
|
||||
}
|
||||
|
||||
// reissued reports what the enroll did: claim generation moved, mails sent, off-site calls.
|
||||
func (f *r1154Fixture) assertReissued(t *testing.T, want bool) {
|
||||
t.Helper()
|
||||
gen := f.generation(t)
|
||||
if want {
|
||||
if gen != f.gen0+1 || f.mail.sends != 1 || f.offsite != 1 {
|
||||
t.Fatalf("want re-issued once: claim generation %d→%d, claim mails %d, off-site re-issues %d", f.gen0, gen, f.mail.sends, f.offsite)
|
||||
}
|
||||
return
|
||||
}
|
||||
if gen != f.gen0 || f.mail.sends != 0 || f.offsite != 0 {
|
||||
t.Fatalf("want NOT re-issued: claim generation %d→%d, claim mails %d, off-site re-issues %d", f.gen0, gen, f.mail.sends, f.offsite)
|
||||
}
|
||||
}
|
||||
|
||||
func (f *r1154Fixture) eventAt(t *testing.T, et string) time.Time {
|
||||
t.Helper()
|
||||
at, err := f.st.LastEventAt("c1", et)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
return at
|
||||
}
|
||||
|
||||
// The old box silent on both streams past the threshold → the enroll re-issues both, keeps the record, records it.
|
||||
//
|
||||
// COMPANION RED-PROOF (observed 2026-10-19): with the pre-fix reuse path restored (the existing host returned at
|
||||
// once, reissueOnReuseIfReinstall not called) this fails with
|
||||
//
|
||||
// r1154_reenroll_test.go: want re-issued once: claim generation 1→1, claim mails 0, off-site re-issues 0
|
||||
//
|
||||
// Restored.
|
||||
func TestR1154_StaleOldHostIsReissued(t *testing.T) {
|
||||
f := newR1154(t)
|
||||
f.hostReportedAgo(t, 2*time.Hour)
|
||||
f.controllerReportedAgo(t, 2*time.Hour)
|
||||
f.enroll(t)
|
||||
f.assertReissued(t, true)
|
||||
if f.eventAt(t, eventHostReenrollReissued).IsZero() {
|
||||
t.Error("host_reenroll_reissued not recorded")
|
||||
}
|
||||
if !f.eventAt(t, "claim_reissued_reenroll").After(time.Time{}) {
|
||||
t.Error("claim_reissued_reenroll (the claim leg's own event) not recorded")
|
||||
}
|
||||
}
|
||||
|
||||
// Switched off on purpose minutes ago (both report clocks still fresh) → re-issued: the box said it was going down.
|
||||
//
|
||||
// COMPANION RED-PROOF (observed 2026-10-19): with the deliberate-off branch removed from decideReenroll this fails
|
||||
// with
|
||||
//
|
||||
// r1154_reenroll_test.go: want re-issued once: claim generation 1→1, claim mails 0, off-site re-issues 0
|
||||
//
|
||||
// Restored.
|
||||
func TestR1154_DeliberatelyOffOldHostIsReissued(t *testing.T) {
|
||||
f := newR1154(t)
|
||||
f.hostReportedAgo(t, 3*time.Minute)
|
||||
f.controllerReportedAgo(t, 3*time.Minute)
|
||||
off := time.Now().UTC().Add(-5 * time.Minute).Truncate(time.Second)
|
||||
if _, err := f.st.SetHostDeliberateOff(r1154Host, &off); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
f.enroll(t)
|
||||
f.assertReissued(t, true)
|
||||
}
|
||||
|
||||
// A LIVE box enrolling again (an installer --resume after its agent started, a repair re-run) keeps every
|
||||
// credential: nothing re-issued, and the skip is recorded so a too-early reinstall is visible. Three ways to be
|
||||
// live, each alone: the host report, the controller report (the codes live in the controller), the wait connection.
|
||||
//
|
||||
// COMPANION RED-PROOF (observed 2026-10-19): with decideReenroll replaced by an unconditional re-issue on the reuse
|
||||
// path this fails with
|
||||
//
|
||||
// r1154_reenroll_test.go: host report fresh: want NOT re-issued: claim generation 1→2, claim mails 1, off-site re-issues 1
|
||||
//
|
||||
// (and the same for the other two cases). Restored.
|
||||
func TestR1154_LiveHostIsNotReissued(t *testing.T) {
|
||||
cases := []struct {
|
||||
name string
|
||||
setup func(t *testing.T, f *r1154Fixture)
|
||||
}{
|
||||
{"host report fresh", func(t *testing.T, f *r1154Fixture) {
|
||||
f.hostReportedAgo(t, 2*time.Minute)
|
||||
f.controllerReportedAgo(t, 2*time.Hour)
|
||||
}},
|
||||
{"controller report fresh, agent silent", func(t *testing.T, f *r1154Fixture) {
|
||||
f.hostReportedAgo(t, 2*time.Hour)
|
||||
f.controllerReportedAgo(t, 5*time.Minute)
|
||||
}},
|
||||
{"controller connected, both reports old", func(t *testing.T, f *r1154Fixture) {
|
||||
f.hostReportedAgo(t, 2*time.Hour)
|
||||
f.controllerReportedAgo(t, 2*time.Hour)
|
||||
hub := intent.New()
|
||||
f.h.SetIntentHub(hub)
|
||||
hub.MarkWaitStart("c1")
|
||||
}},
|
||||
{"controller connected, deliberately off", func(t *testing.T, f *r1154Fixture) {
|
||||
f.hostReportedAgo(t, 2*time.Minute)
|
||||
f.controllerReportedAgo(t, 2*time.Minute)
|
||||
off := time.Now().UTC().Add(-time.Minute).Truncate(time.Second)
|
||||
f.st.SetHostDeliberateOff(r1154Host, &off)
|
||||
hub := intent.New()
|
||||
f.h.SetIntentHub(hub)
|
||||
hub.MarkWaitStart("c1")
|
||||
}},
|
||||
}
|
||||
for _, c := range cases {
|
||||
t.Run(c.name, func(t *testing.T) {
|
||||
f := newR1154(t)
|
||||
c.setup(t, f)
|
||||
f.enroll(t)
|
||||
if gen := f.generation(t); gen != f.gen0 || f.mail.sends != 0 || f.offsite != 0 {
|
||||
t.Fatalf("%s: want NOT re-issued: claim generation %d→%d, claim mails %d, off-site re-issues %d",
|
||||
c.name, f.gen0, gen, f.mail.sends, f.offsite)
|
||||
}
|
||||
if f.eventAt(t, eventHostReenrollNotReissued).IsZero() {
|
||||
t.Errorf("%s: the live skip was not recorded (host_reenroll_not_reissued)", c.name)
|
||||
}
|
||||
if !f.eventAt(t, eventHostReenrollReissued).IsZero() {
|
||||
t.Errorf("%s: a re-issue record was written for a live box", c.name)
|
||||
}
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
// Once per silence: an installer --resume of the same reinstall enrolls again before the new agent's first report
|
||||
// — the old box is still "silent", and the re-issues must not run twice (the claim engine's single-shot rule).
|
||||
// Then the new agent reports, goes silent again past the threshold (the next reinstall): re-issued again.
|
||||
//
|
||||
// COMPANION RED-PROOF (observed 2026-10-19): with the once-per-silence check removed from decideReenroll this fails
|
||||
// with
|
||||
//
|
||||
// r1154_reenroll_test.go: the --resume re-issued again: claim generation 1→3, off-site re-issues 2
|
||||
//
|
||||
// Restored.
|
||||
func TestR1154_ResumeDoesNotReissueTwice(t *testing.T) {
|
||||
f := newR1154(t)
|
||||
f.hostReportedAgo(t, 2*time.Hour)
|
||||
f.controllerReportedAgo(t, 2*time.Hour)
|
||||
f.enroll(t)
|
||||
f.enroll(t) // the --resume
|
||||
if gen := f.generation(t); gen != f.gen0+1 || f.offsite != 1 {
|
||||
t.Fatalf("the --resume re-issued again: claim generation %d→%d, off-site re-issues %d", f.gen0, gen, f.offsite)
|
||||
}
|
||||
// The record is backdated so the next report is strictly after it (second-granularity clocks).
|
||||
f.exec(t, `UPDATE events SET created_at = datetime('now', '-3 hours') WHERE event_type = ?`, eventHostReenrollReissued)
|
||||
f.exec(t, `UPDATE hosts SET last_report_at = datetime('now', '-2 hours', '-30 minutes') WHERE host_id = ?`, r1154Host)
|
||||
f.enroll(t)
|
||||
if gen := f.generation(t); gen != f.gen0+2 || f.offsite != 2 {
|
||||
t.Fatalf("a later reinstall (the box reported after the last re-issue, then went silent) must re-issue: claim generation %d→%d, off-site re-issues %d",
|
||||
f.gen0, gen, f.offsite)
|
||||
}
|
||||
}
|
||||
|
||||
// Edges the code shows. A host record that never reported: a provisional record (operator /admin/hosts, no
|
||||
// re-issue ran for it) → re-issued, like the mint it stands in for; a record the enroll itself minted (the
|
||||
// re-issues ran at mint) → not again on the --resume. Not wired (no threshold) → nothing re-issued, the
|
||||
// pre-v0.159.0 behaviour.
|
||||
//
|
||||
// COMPANION RED-PROOF (observed 2026-10-19): with the pre-fix reuse path restored, „provisional never reported"
|
||||
// fails with „want re-issued once: claim generation 1→1, claim mails 0, off-site re-issues 0"; with the
|
||||
// once-per-silence check removed, „minted by the enroll" fails with „off-site re-issues = 2, want 1"; with an
|
||||
// unconditional re-issue, „threshold not wired" fails with „want NOT re-issued: claim generation 1→2, claim mails
|
||||
// 1, off-site re-issues 1". Restored.
|
||||
func TestR1154_NeverReportedAndUnwiredEdges(t *testing.T) {
|
||||
t.Run("provisional never reported", func(t *testing.T) {
|
||||
f := newR1154(t)
|
||||
f.exec(t, `UPDATE hosts SET last_report_at = NULL WHERE host_id = ?`, r1154Host)
|
||||
f.enroll(t)
|
||||
f.assertReissued(t, true)
|
||||
})
|
||||
t.Run("minted by the enroll, never reported", func(t *testing.T) {
|
||||
h, st, _ := newTestHandler(t)
|
||||
st.SaveCustomerConfig(&store.CustomerConfig{CustomerID: "c1", APIKey: "ckey", RetrievalPassword: "pass-phrase"})
|
||||
h.SetStaleThreshold(45 * time.Minute)
|
||||
calls := 0
|
||||
h.SetOffsiteReissuer(func(context.Context, string) error { calls++; return nil })
|
||||
if rr := doEnroll(h, "c1", "pass-phrase"); rr.Code != http.StatusCreated {
|
||||
t.Fatalf("first enroll = %d", rr.Code)
|
||||
}
|
||||
if rr := doEnroll(h, "c1", "pass-phrase"); rr.Code != http.StatusOK {
|
||||
t.Fatalf("resume enroll = %d", rr.Code)
|
||||
}
|
||||
if calls != 1 {
|
||||
t.Fatalf("off-site re-issues = %d, want 1 (at the mint only, not again on the --resume)", calls)
|
||||
}
|
||||
})
|
||||
t.Run("threshold not wired", func(t *testing.T) {
|
||||
f := newR1154(t)
|
||||
f.h.SetStaleThreshold(0)
|
||||
f.hostReportedAgo(t, 2*time.Hour)
|
||||
f.controllerReportedAgo(t, 2*time.Hour)
|
||||
f.enroll(t)
|
||||
f.assertReissued(t, false)
|
||||
})
|
||||
}
|
||||
|
||||
// Both R-1154 event types: operator-only, and hub-origin only — a box able to POST host_reenroll_reissued could
|
||||
// plant the once-per-silence record and block the re-issue of its own reinstall.
|
||||
func TestR1154_EventsAreOperatorOnlyAndHubOrigin(t *testing.T) {
|
||||
for _, et := range []string{eventHostReenrollReissued, eventHostReenrollNotReissued} {
|
||||
if !notify.IsOperatorOnly(et) {
|
||||
t.Errorf("%s is not operator-only", et)
|
||||
}
|
||||
if allowedEventTypes[et] {
|
||||
t.Errorf("%s is in allowedEventTypes — a box could POST it", et)
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,148 @@
|
||||
package api
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"time"
|
||||
|
||||
"gitea.dooplex.hu/admin/felhom-hub/internal/intent"
|
||||
"gitea.dooplex.hu/admin/felhom-hub/internal/store"
|
||||
)
|
||||
|
||||
// R-1154 (hub v0.159.0, brief 2026-10-19) — a reinstall while the customer's OLD host record still exists.
|
||||
//
|
||||
// POST /host-enroll is called by the installer (felhom-host-install.sh step 4) on EVERY pass, a --resume
|
||||
// included, with the customer's retrieval passphrase. An existing host record is reused (same host id and
|
||||
// key — a reinstalled box takes over the record). Until v0.159.0 the reuse path returned at once, so the
|
||||
// clean-slate re-issues (reissueOnReenroll: the claim code, the off-site login) ran ONLY when the old host
|
||||
// had been deleted first, and a plain reinstall skipped both in silence: the household stuck at the claim
|
||||
// page, the off-site tier stuck (proven on demo-hp 2026-08-21).
|
||||
//
|
||||
// THE RULE (`05` §8). The reuse path re-issues when the OLD BOX IS GONE, judged by the hub's own clocks:
|
||||
//
|
||||
// 1. NEVER while the box's controller holds a wait connection right now (intent presence "connected").
|
||||
// 2. The household switched the box off on purpose (deliberate_off_at inside its hold) — the report clocks
|
||||
// are not consulted: the box said it was going down.
|
||||
// 3. Otherwise BOTH the agent's host report AND the controller's report are older than staleThreshold (or
|
||||
// absent). The claim code and the off-site login live in the CONTROLLER, so a box whose agent is
|
||||
// silent but whose controller still reports is live for this purpose (an installer re-run to repair a
|
||||
// broken agent must not cost the household its codes). Same threshold, same comparison as hostStatus:
|
||||
// no second definition of "live".
|
||||
// 4. Once per silence: not when the re-issues already ran for this host since its last report (or, for a
|
||||
// never-reported host, since the record was created) — a --resume of the same reinstall enrolls again
|
||||
// before the new agent's first report. The record is the hub-origin info event host_reenroll_reissued.
|
||||
//
|
||||
// A live box that enrolls again (a --resume after its agent started, a repair re-run) gets its credentials
|
||||
// and NOTHING re-issued; that skip is recorded as the info event host_reenroll_not_reissued, so a reinstall
|
||||
// done within the threshold of the old box's last report is visible on the event list instead of silent.
|
||||
// Not wired (staleThreshold 0) or a failed lookup: nothing is re-issued — the pre-v0.159.0 behaviour.
|
||||
//
|
||||
// Both event types are hub-origin only (not in allowedEventTypes — a box must not be able to plant the
|
||||
// once-per-silence record) and operator-only. Pinned by r1154_reenroll_test.go.
|
||||
|
||||
const (
|
||||
eventHostReenrollReissued = "host_reenroll_reissued"
|
||||
eventHostReenrollNotReissued = "host_reenroll_not_reissued"
|
||||
)
|
||||
|
||||
// reenrollDecision is the verdict on a reuse-path enroll.
|
||||
type reenrollDecision struct {
|
||||
reissue bool // run the clean-slate re-issues
|
||||
live bool // skipped because the box is live (recorded as an event)
|
||||
why string // operator English, for the log line and the event
|
||||
}
|
||||
|
||||
func (h *Handler) decideReenroll(host *store.Host, now time.Time) reenrollDecision {
|
||||
if h.staleThreshold <= 0 {
|
||||
return reenrollDecision{why: "no stale threshold wired"}
|
||||
}
|
||||
if h.intentHub != nil {
|
||||
if p := h.intentHub.Presence(host.CustomerID, now); p.State == intent.PresenceConnected {
|
||||
return reenrollDecision{live: true, why: "the box's controller is connected to the hub right now"}
|
||||
}
|
||||
}
|
||||
var why string
|
||||
if store.DeliberateOffHeld(host.DeliberateOffAt, now) {
|
||||
why = "the household switched the box off at " + host.DeliberateOffAt.UTC().Format("2006-01-02 15:04 UTC")
|
||||
} else {
|
||||
hostAge := "never reported"
|
||||
if host.LastReportAt != nil {
|
||||
age := now.Sub(*host.LastReportAt)
|
||||
if age <= h.staleThreshold {
|
||||
return reenrollDecision{live: true, why: "the host reported " + roundAge(age) + " ago"}
|
||||
}
|
||||
hostAge = "last reported " + roundAge(age) + " ago"
|
||||
}
|
||||
found, ctrlAt, _, err := h.store.LatestReportOffsitePresence(host.CustomerID)
|
||||
if err != nil {
|
||||
h.logger.Printf("[WARN] host-enroll: controller report lookup failed for %s: %v — nothing re-issued", host.CustomerID, err)
|
||||
return reenrollDecision{why: "the controller report lookup failed"}
|
||||
}
|
||||
ctrlAge := "the controller never reported"
|
||||
if found {
|
||||
age := now.Sub(ctrlAt)
|
||||
if age <= h.staleThreshold {
|
||||
return reenrollDecision{live: true, why: "the controller reported " + roundAge(age) + " ago"}
|
||||
}
|
||||
ctrlAge = "the controller last reported " + roundAge(age) + " ago"
|
||||
}
|
||||
why = "the host " + hostAge + ", " + ctrlAge + " (stale threshold " + h.staleThreshold.String() + ")"
|
||||
}
|
||||
// Once per silence.
|
||||
since := host.CreatedAt
|
||||
if host.LastReportAt != nil && host.LastReportAt.After(since) {
|
||||
since = *host.LastReportAt
|
||||
}
|
||||
last, err := h.store.LastEventAt(host.CustomerID, eventHostReenrollReissued)
|
||||
if err != nil {
|
||||
h.logger.Printf("[WARN] host-enroll: re-issue record lookup failed for %s: %v — nothing re-issued", host.CustomerID, err)
|
||||
return reenrollDecision{why: "the re-issue record lookup failed"}
|
||||
}
|
||||
if !last.IsZero() && !last.Before(since) {
|
||||
return reenrollDecision{why: "already re-issued at " + last.UTC().Format("2006-01-02 15:04:05 UTC") + " for this install"}
|
||||
}
|
||||
return reenrollDecision{reissue: true, why: why}
|
||||
}
|
||||
|
||||
// reissueOnReuseIfReinstall runs the clean-slate re-issues on the reuse path when decideReenroll says the
|
||||
// old box is gone, and records the outcome. Best-effort, like reissueOnReenroll: it never fails the enroll.
|
||||
func (h *Handler) reissueOnReuseIfReinstall(cc *store.CustomerConfig, host *store.Host, now time.Time) {
|
||||
d := h.decideReenroll(host, now)
|
||||
switch {
|
||||
case d.reissue:
|
||||
h.logger.Printf("[INFO] host re-enrolled: %s (customer %s) — re-issuing the claim code and the off-site login: %s",
|
||||
host.HostID, cc.CustomerID, d.why)
|
||||
h.reissueOnReenroll(cc)
|
||||
h.recordReenrollReissued(cc.CustomerID, host.HostID, d.why)
|
||||
case d.live:
|
||||
h.logger.Printf("[INFO] host re-enrolled: %s (customer %s) — nothing re-issued, the box is live: %s",
|
||||
host.HostID, cc.CustomerID, d.why)
|
||||
msg := "Host " + host.HostID + " enrolled again while the box is live (" + d.why + ") — the claim code and " +
|
||||
"the off-site login were NOT re-issued. If this box was reinstalled, send a new claim code and re-issue " +
|
||||
"the off-site login from the customer page."
|
||||
h.saveReenrollEvent(cc.CustomerID, eventHostReenrollNotReissued, msg, host.HostID, d.why)
|
||||
default:
|
||||
h.logger.Printf("[INFO] host re-enrolled: %s (customer %s) — nothing re-issued: %s", host.HostID, cc.CustomerID, d.why)
|
||||
}
|
||||
}
|
||||
|
||||
// recordReenrollReissued stores the once-per-silence record (and the operator's line) after the re-issues ran.
|
||||
func (h *Handler) recordReenrollReissued(customerID, hostID, why string) {
|
||||
msg := "Host " + hostID + " enrolled — " + why + " — the clean-slate re-issues ran (the claim code if the " +
|
||||
"customer is claimed, the off-site login if it is provisioned)"
|
||||
h.saveReenrollEvent(customerID, eventHostReenrollReissued, msg, hostID, why)
|
||||
}
|
||||
|
||||
func (h *Handler) saveReenrollEvent(customerID, eventType, msg, hostID, why string) {
|
||||
details, _ := json.Marshal(map[string]string{"host_id": hostID, "reason": why})
|
||||
if _, err := h.store.SaveEvent(customerID, eventType, "info", msg, string(details), "hub"); err != nil {
|
||||
h.logger.Printf("[WARN] save %s for %s: %v", eventType, customerID, err)
|
||||
}
|
||||
}
|
||||
|
||||
func roundAge(d time.Duration) string {
|
||||
if d < time.Minute {
|
||||
return fmt.Sprintf("%ds", int(d.Seconds()))
|
||||
}
|
||||
return d.Round(time.Minute).String()
|
||||
}
|
||||
@@ -191,8 +191,9 @@ func (e *Engine) RequestReset(cc *store.CustomerConfig) error {
|
||||
//
|
||||
// No-op for an UNCLAIMED customer — that is the first-provision path where EnsureIssued already
|
||||
// owns the first code; re-enrolling before the first claim must NOT rotate. The CALLER guarantees
|
||||
// single-shot by invoking this only on a genuinely fresh host record (the host-enroll mint path,
|
||||
// which fires exactly once per reinstall). Returns (generation, reissued, error).
|
||||
// single-shot: the host-enroll mint path (a fresh host record), and since hub v0.159.0 (R-1154) the
|
||||
// reuse path once per silence of the old box (api/reenroll.go). Pinned by api/r1154_reenroll_test.go
|
||||
// TestR1154_ResumeDoesNotReissueTwice. Returns (generation, reissued, error).
|
||||
func (e *Engine) ReissueForReenroll(cc *store.CustomerConfig) (gen int, reissued bool, err error) {
|
||||
cs, err := e.Store.GetClaim(cc.CustomerID)
|
||||
if err != nil {
|
||||
|
||||
@@ -20,17 +20,49 @@ import (
|
||||
// (a missing user-facing grant — the multi-drive-flapping class). Non-critical degradations ride
|
||||
// the report + the agent's own logs but do NOT alert the operator (avoid noise). The event is
|
||||
// attributed to the host's customer, so the existing operator notification UX picks it up
|
||||
// unchanged. Customer is NEVER notified — internal capability health is operator-only.
|
||||
// unchanged. Customer is NEVER notified — internal capability health is operator-only: both event
|
||||
// types sit in notify.operatorOnlyEvents (R-1150, hub v0.159.0; before that only configuration kept
|
||||
// them from a household). Pinned by notify/r1150_operator_only_test.go.
|
||||
//
|
||||
// R-1150 (hub v0.159.0) — ONE capability is not a grant. The agent's `localapi:backup-cadence-cap`
|
||||
// (felhom-agent cmd/felhom-agent/main.go backupCapStatusName, R-973) is CRITICAL and "degraded" for
|
||||
// 24 h after the agent REFUSED a whole-guest backup a guest asked for beyond its schedule. Reported
|
||||
// as „agent privileged capability degraded" it read as a broken sudo grant. It is therefore a SECOND,
|
||||
// independent state per host with its own event pair:
|
||||
//
|
||||
// - the grants (every other critical capability): ok↔degraded → agent_capability_degraded /
|
||||
// agent_capability_recovered, exactly as before, the cadence cap never named in them;
|
||||
// - the backup refusal (the cadence cap alone): ok→degraded → whole_guest_backup_refused (warning),
|
||||
// degraded→ok → whole_guest_backup_refusal_cleared (info: recorded, never mailed).
|
||||
//
|
||||
// Two states rather than one verdict with a special case, because the two can change on the SAME
|
||||
// report independently (a refusal while a grant is missing, a grant restored while the refusal
|
||||
// stands): each fires on its own edge, so one report flip is at most one event per state — never
|
||||
// two events for one state, and never a "grants restored" line when only the refusal window ended.
|
||||
// Pinned by host_capability_test.go TestR1150_*.
|
||||
type HostCapabilityChecker struct {
|
||||
store *store.Store
|
||||
logger *log.Logger
|
||||
onEvent EventNotifyFunc
|
||||
|
||||
mu sync.Mutex
|
||||
states map[string]string // hostID → "ok" | "degraded"
|
||||
states map[string]string // hostID → "ok" | "degraded" (the grants)
|
||||
refusals map[string]string // hostID → "ok" | "degraded" (the backup cadence cap, R-1150)
|
||||
customerOf map[string]string // hostID → customerID
|
||||
}
|
||||
|
||||
// backupCadenceCapName is the agent's R-973 cadence-cap status name. It must equal felhom-agent's
|
||||
// backupCapStatusName (cmd/felhom-agent/main.go) — the agent's status name stays (R-1150).
|
||||
const backupCadenceCapName = "localapi:backup-cadence-cap"
|
||||
|
||||
// R-1150 event types — hub-origin ONLY (deliberately not in api.allowedEventTypes: the guest's
|
||||
// controller is the party this alarm suspects, and a box able to POST the type could fake it or,
|
||||
// by spending the operator cooldown first, drown the real one) and operator-only.
|
||||
const (
|
||||
EventWholeGuestBackupRefused = "whole_guest_backup_refused"
|
||||
EventWholeGuestBackupRefusalCleared = "whole_guest_backup_refusal_cleared"
|
||||
)
|
||||
|
||||
// NewHostCapabilityChecker creates the checker and seeds state from the current snapshots. No
|
||||
// events are generated during initialization (mirrors the staleness checker).
|
||||
func NewHostCapabilityChecker(s *store.Store, onEvent EventNotifyFunc, logger *log.Logger) *HostCapabilityChecker {
|
||||
@@ -39,6 +71,7 @@ func NewHostCapabilityChecker(s *store.Store, onEvent EventNotifyFunc, logger *l
|
||||
logger: logger,
|
||||
onEvent: onEvent,
|
||||
states: make(map[string]string),
|
||||
refusals: make(map[string]string),
|
||||
customerOf: make(map[string]string),
|
||||
}
|
||||
rows, err := s.GetHostCapabilities()
|
||||
@@ -46,7 +79,7 @@ func NewHostCapabilityChecker(s *store.Store, onEvent EventNotifyFunc, logger *l
|
||||
logger.Printf("[WARN] Host capability checker: failed to seed states: %v", err)
|
||||
return cc
|
||||
}
|
||||
var okCount, degCount int
|
||||
var okCount, degCount, refusedCount int
|
||||
for _, row := range rows {
|
||||
if s.IsCustomerBlocked(row.CustomerID) {
|
||||
continue
|
||||
@@ -56,14 +89,20 @@ func NewHostCapabilityChecker(s *store.Store, onEvent EventNotifyFunc, logger *l
|
||||
// F2: seed only HEALTHY hosts. A host already degraded at (re)start is left UNSEEDED so the
|
||||
// first Check() observes oldState=="" and emits once — otherwise a box already broken when the
|
||||
// hub restarts would stay silent forever. The dispatcher's 1h cooldown dedups the re-emit.
|
||||
// The same rule, separately, for the backup refusal (R-1150).
|
||||
if st == "degraded" {
|
||||
degCount++
|
||||
continue
|
||||
} else {
|
||||
cc.states[row.HostID] = st
|
||||
okCount++
|
||||
}
|
||||
if rst, _ := backupRefusalState(row.Capabilities); rst == "degraded" {
|
||||
refusedCount++
|
||||
} else {
|
||||
cc.refusals[row.HostID] = rst
|
||||
}
|
||||
cc.states[row.HostID] = st
|
||||
okCount++
|
||||
}
|
||||
logger.Printf("[INFO] Host capability checker initialized: %d ok, %d degraded (degraded left unseeded → first Check emits)", okCount, degCount)
|
||||
logger.Printf("[INFO] Host capability checker initialized: %d ok, %d degraded, %d with a refused whole-guest backup (degraded/refused left unseeded → first Check emits)", okCount, degCount, refusedCount)
|
||||
return cc
|
||||
}
|
||||
|
||||
@@ -83,37 +122,44 @@ func (cc *HostCapabilityChecker) Check() {
|
||||
seen[row.HostID] = true
|
||||
if cc.store.IsCustomerBlocked(row.CustomerID) {
|
||||
delete(cc.states, row.HostID)
|
||||
delete(cc.refusals, row.HostID)
|
||||
continue
|
||||
}
|
||||
cc.customerOf[row.HostID] = row.CustomerID
|
||||
|
||||
newState, names, features := capabilityState(row.Capabilities)
|
||||
oldState := cc.states[row.HostID]
|
||||
if oldState == "" {
|
||||
// F2: first observation (incl. a hub restart). A healthy first-obs seeds silently; a host
|
||||
// already DEGRADED at first sight emits once (else a host that was already broken when the
|
||||
// hub (re)started would never alert). The dispatcher's 1h operator cooldown dedups a re-emit
|
||||
// across a hub restart.
|
||||
cc.states[row.HostID] = newState
|
||||
if newState == "degraded" {
|
||||
cc.emitTransition(row.HostID, row.CustomerID, "unknown", newState, names, features)
|
||||
}
|
||||
continue
|
||||
if old, changed := step(cc.states, row.HostID, newState); changed {
|
||||
cc.emitTransition(row.HostID, row.CustomerID, old, newState, names, features)
|
||||
}
|
||||
if oldState == newState {
|
||||
continue
|
||||
refState, reason := backupRefusalState(row.Capabilities)
|
||||
if old, changed := step(cc.refusals, row.HostID, refState); changed {
|
||||
cc.emitRefusal(row.HostID, row.CustomerID, old, refState, reason)
|
||||
}
|
||||
cc.states[row.HostID] = newState
|
||||
cc.emitTransition(row.HostID, row.CustomerID, oldState, newState, names, features)
|
||||
}
|
||||
|
||||
for id := range cc.states {
|
||||
if !seen[id] {
|
||||
delete(cc.states, id)
|
||||
for _, m := range []map[string]string{cc.states, cc.refusals} {
|
||||
for id := range m {
|
||||
if !seen[id] {
|
||||
delete(m, id)
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// step records newState for hostID in m and reports whether an event is due, with the state it left
|
||||
// ("unknown" on a first observation). F2: a first observation (incl. a hub restart) seeds a healthy
|
||||
// state silently, and a state already DEGRADED at first sight is due once (else a host that was already
|
||||
// broken when the hub (re)started would never alert). The dispatcher's 1h operator cooldown dedups a
|
||||
// re-emit across a hub restart. A steady state is never due.
|
||||
func step(m map[string]string, hostID, newState string) (oldState string, due bool) {
|
||||
oldState = m[hostID]
|
||||
m[hostID] = newState
|
||||
if oldState == "" {
|
||||
return "unknown", newState == "degraded"
|
||||
}
|
||||
return oldState, oldState != newState
|
||||
}
|
||||
|
||||
// GetState returns the current capability state for a host ("unknown" if unseen).
|
||||
func (cc *HostCapabilityChecker) GetState(hostID string) string {
|
||||
cc.mu.Lock()
|
||||
@@ -125,11 +171,27 @@ func (cc *HostCapabilityChecker) GetState(hostID string) string {
|
||||
return s
|
||||
}
|
||||
|
||||
// GetBackupRefusalState returns the host's whole-guest backup refusal state (R-1150): "ok", "degraded"
|
||||
// (the agent refused one inside its window), or "unknown" if unseen.
|
||||
func (cc *HostCapabilityChecker) GetBackupRefusalState(hostID string) string {
|
||||
cc.mu.Lock()
|
||||
defer cc.mu.Unlock()
|
||||
s := cc.refusals[hostID]
|
||||
if s == "" {
|
||||
return "unknown"
|
||||
}
|
||||
return s
|
||||
}
|
||||
|
||||
// capabilityState reduces a snapshot to "ok"/"degraded" plus the degraded CRITICAL capability names
|
||||
// + features (for the event detail). An empty/absent snapshot (pre-v0.44.0 agent) is "ok" — an old
|
||||
// agent can't trip a false alert.
|
||||
// agent can't trip a false alert. The backup cadence cap is NOT a grant and is left to
|
||||
// backupRefusalState (R-1150).
|
||||
func capabilityState(caps []store.CapabilityStatus) (state string, names, features []string) {
|
||||
for _, c := range caps {
|
||||
if c.Name == backupCadenceCapName {
|
||||
continue
|
||||
}
|
||||
if c.Critical && c.Status == "degraded" {
|
||||
names = append(names, c.Name)
|
||||
features = append(features, c.Feature)
|
||||
@@ -175,6 +237,50 @@ func (cc *HostCapabilityChecker) emitTransition(hostID, customerID, oldState, ne
|
||||
}
|
||||
}
|
||||
|
||||
// backupRefusalState reads the agent's cadence-cap status (R-1150): "degraded" with the agent's own
|
||||
// reason while it reports a refused whole-guest backup, else "ok" (absent on an agent without the cap).
|
||||
// The agent always reports the cap as critical; Critical is required here as it is for every alarm.
|
||||
func backupRefusalState(caps []store.CapabilityStatus) (state, reason string) {
|
||||
for _, c := range caps {
|
||||
if c.Name == backupCadenceCapName && c.Critical && c.Status == "degraded" {
|
||||
return "degraded", c.Reason
|
||||
}
|
||||
}
|
||||
return "ok", ""
|
||||
}
|
||||
|
||||
// emitRefusal raises the operator's alarm for a refused whole-guest backup (R-1150) — the agent turned
|
||||
// down a backup the guest asked for beyond its schedule — and records its end.
|
||||
func (cc *HostCapabilityChecker) emitRefusal(hostID, customerID, oldState, newState, reason string) {
|
||||
var eventType, severity, message string
|
||||
switch {
|
||||
case newState == "degraded":
|
||||
eventType = EventWholeGuestBackupRefused
|
||||
severity = "warning"
|
||||
message = "Host " + hostID + ": a whole-guest backup was refused — the guest asked for more whole-guest " +
|
||||
"backups than its schedule allows. Its controller may be misbehaving or compromised: check the guest."
|
||||
case newState == "ok" && oldState == "degraded":
|
||||
eventType = EventWholeGuestBackupRefusalCleared
|
||||
severity = "info"
|
||||
message = "Host " + hostID + ": the agent no longer reports a refused whole-guest backup (its refusal window has ended)"
|
||||
default:
|
||||
return
|
||||
}
|
||||
details, _ := json.Marshal(map[string]any{
|
||||
"host_id": hostID,
|
||||
"capability": backupCadenceCapName,
|
||||
"agent_reason": reason,
|
||||
})
|
||||
cc.logger.Printf("[INFO] Host capability: %s backup refusal %s → %s (%s)", hostID, oldState, newState, eventType)
|
||||
if _, err := cc.store.SaveEvent(customerID, eventType, severity, message, string(details), "hub"); err != nil {
|
||||
cc.logger.Printf("[WARN] Failed to save backup refusal event for %s: %v", hostID, err)
|
||||
return
|
||||
}
|
||||
if cc.onEvent != nil {
|
||||
cc.onEvent(customerID, eventType, severity, message, string(details), "hub")
|
||||
}
|
||||
}
|
||||
|
||||
func dedup(in []string) []string {
|
||||
seen := make(map[string]bool, len(in))
|
||||
var out []string
|
||||
|
||||
@@ -4,6 +4,7 @@ import (
|
||||
"io"
|
||||
"log"
|
||||
"path/filepath"
|
||||
"strings"
|
||||
"testing"
|
||||
|
||||
"gitea.dooplex.hu/admin/felhom-hub/internal/store"
|
||||
@@ -130,3 +131,170 @@ func TestHostCapabilityChecker_F2_BornDegradedAlerts(t *testing.T) {
|
||||
t.Fatalf("steady degraded must not re-emit, got %v", events)
|
||||
}
|
||||
}
|
||||
|
||||
// ── R-1150: a refused whole-guest backup has its own alarm ───────────────────────────────────────
|
||||
|
||||
const capCadenceRefused = `[{"name":"guest-init-pid","feature":"drive-gate","critical":true,"status":"ok"},` +
|
||||
`{"name":"localapi:backup-cadence-cap","feature":"whole-guest restore points","critical":true,"status":"degraded",` +
|
||||
`"reason":"refused a whole-guest backup of guest 9202 on target \"local\" at 2026-10-19T03:15:00Z — more than the schedule allows; the guest's controller may be misbehaving or compromised: check the guest"}]`
|
||||
const capCadenceOK = `[{"name":"guest-init-pid","feature":"drive-gate","critical":true,"status":"ok"},` +
|
||||
`{"name":"localapi:backup-cadence-cap","feature":"whole-guest restore points","critical":true,"status":"ok"}]`
|
||||
const capGrantAndCadenceDegraded = `[{"name":"guest-init-pid","feature":"drive-gate guest-sees","critical":true,"status":"degraded","reason":"sudo policy denied"},` +
|
||||
`{"name":"localapi:backup-cadence-cap","feature":"whole-guest restore points","critical":true,"status":"degraded","reason":"refused a whole-guest backup of guest 9202"}]`
|
||||
const capGrantDegradedCadenceOK = `[{"name":"guest-init-pid","feature":"drive-gate guest-sees","critical":true,"status":"degraded","reason":"sudo policy denied"},` +
|
||||
`{"name":"localapi:backup-cadence-cap","feature":"whole-guest restore points","critical":true,"status":"ok"}]`
|
||||
|
||||
type capEvent struct{ eventType, severity, message, details string }
|
||||
|
||||
func recordCapEvents(evs *[]capEvent) EventNotifyFunc {
|
||||
return func(_, eventType, severity, message, details, _ string) {
|
||||
*evs = append(*evs, capEvent{eventType, severity, message, details})
|
||||
}
|
||||
}
|
||||
|
||||
func capTypes(evs []capEvent) []string {
|
||||
out := make([]string, len(evs))
|
||||
for i, e := range evs {
|
||||
out[i] = e.eventType
|
||||
}
|
||||
return out
|
||||
}
|
||||
|
||||
// The cadence cap ALONE degraded → exactly one whole_guest_backup_refused (warning, plain text, the agent's
|
||||
// reason in the details) and NOT agent_capability_degraded; steady → nothing; the window ending → one
|
||||
// whole_guest_backup_refusal_cleared (info) and NOT agent_capability_recovered.
|
||||
//
|
||||
// COMPANION RED-PROOF (observed 2026-10-19): with the pre-fix shape restored (capabilityState counting the
|
||||
// cadence cap like any grant, no separate refusal state) this test fails with
|
||||
//
|
||||
// host_capability_test.go: cadence cap alone → want [whole_guest_backup_refused], got [agent_capability_degraded]
|
||||
//
|
||||
// Restored.
|
||||
func TestR1150_CadenceCapAloneRaisesWholeGuestBackupRefused(t *testing.T) {
|
||||
st := newCapStore(t)
|
||||
st.SaveHostReport("h1", "c1", reportWith(capCadenceOK), store.HostReportDenorm{})
|
||||
var evs []capEvent
|
||||
cc := NewHostCapabilityChecker(st, recordCapEvents(&evs), log.New(io.Discard, "", 0))
|
||||
|
||||
st.SaveHostReport("h1", "c1", reportWith(capCadenceRefused), store.HostReportDenorm{})
|
||||
cc.Check()
|
||||
if got := capTypes(evs); len(got) != 1 || got[0] != "whole_guest_backup_refused" {
|
||||
t.Fatalf("cadence cap alone → want [whole_guest_backup_refused], got %v", got)
|
||||
}
|
||||
e := evs[0]
|
||||
if e.severity != "warning" {
|
||||
t.Errorf("severity = %q, want warning", e.severity)
|
||||
}
|
||||
if !strings.Contains(e.message, "a whole-guest backup was refused") || !strings.Contains(e.message, "check the guest") {
|
||||
t.Errorf("message does not say what happened in plain words: %q", e.message)
|
||||
}
|
||||
if strings.Contains(e.message, "privileged capability") {
|
||||
t.Errorf("message still reads as a broken grant: %q", e.message)
|
||||
}
|
||||
if !strings.Contains(e.details, "guest 9202") || !strings.Contains(e.details, "localapi:backup-cadence-cap") {
|
||||
t.Errorf("details must carry the agent's reason and the capability name: %s", e.details)
|
||||
}
|
||||
if cc.GetState("h1") != "ok" {
|
||||
t.Errorf("the grants state = %s, want ok (the cadence cap is not a grant)", cc.GetState("h1"))
|
||||
}
|
||||
if cc.GetBackupRefusalState("h1") != "degraded" {
|
||||
t.Errorf("refusal state = %s, want degraded", cc.GetBackupRefusalState("h1"))
|
||||
}
|
||||
// The event is STORED too (the operator's event list), under the new type.
|
||||
if at, err := st.LastEventAt("c1", "whole_guest_backup_refused"); err != nil || at.IsZero() {
|
||||
t.Errorf("whole_guest_backup_refused not stored: at=%v err=%v", at, err)
|
||||
}
|
||||
|
||||
st.SaveHostReport("h1", "c1", reportWith(capCadenceRefused), store.HostReportDenorm{})
|
||||
cc.Check()
|
||||
if len(evs) != 1 {
|
||||
t.Fatalf("steady refusal must not re-emit, got %v", capTypes(evs))
|
||||
}
|
||||
|
||||
st.SaveHostReport("h1", "c1", reportWith(capCadenceOK), store.HostReportDenorm{})
|
||||
cc.Check()
|
||||
if got := capTypes(evs); len(got) != 2 || got[1] != "whole_guest_backup_refusal_cleared" {
|
||||
t.Fatalf("refusal window ended → want [.. whole_guest_backup_refusal_cleared], got %v", got)
|
||||
}
|
||||
if evs[1].severity != "info" {
|
||||
t.Errorf("cleared severity = %q, want info (recorded, never mailed)", evs[1].severity)
|
||||
}
|
||||
}
|
||||
|
||||
// Another critical cap degraded → agent_capability_degraded unchanged (the cadence cap never named in it);
|
||||
// both degraded on ONE report → exactly one event per state; the grant restored while the refusal stands →
|
||||
// agent_capability_recovered alone; nothing for a steady state.
|
||||
//
|
||||
// COMPANION RED-PROOF (observed 2026-10-19): with the pre-fix shape restored the both-degraded report emits a
|
||||
// single event naming both caps and the test fails with
|
||||
//
|
||||
// host_capability_test.go: grant + cadence on one report → want [agent_capability_degraded whole_guest_backup_refused], got [agent_capability_degraded]
|
||||
//
|
||||
// Restored.
|
||||
func TestR1150_OtherCapStillRaisesCapabilityDegraded(t *testing.T) {
|
||||
st := newCapStore(t)
|
||||
st.SaveHostReport("h1", "c1", reportWith(capCadenceOK), store.HostReportDenorm{})
|
||||
var evs []capEvent
|
||||
cc := NewHostCapabilityChecker(st, recordCapEvents(&evs), log.New(io.Discard, "", 0))
|
||||
|
||||
// A grant alone: the old event, unchanged text.
|
||||
st.SaveHostReport("h1", "c1", reportWith(capGrantDegradedCadenceOK), store.HostReportDenorm{})
|
||||
cc.Check()
|
||||
if got := capTypes(evs); len(got) != 1 || got[0] != "agent_capability_degraded" {
|
||||
t.Fatalf("grant alone → want [agent_capability_degraded], got %v", got)
|
||||
}
|
||||
if !strings.Contains(evs[0].message, "agent privileged capability degraded — guest-init-pid") {
|
||||
t.Errorf("agent_capability_degraded text changed: %q", evs[0].message)
|
||||
}
|
||||
// Back to clean, then both on one report.
|
||||
st.SaveHostReport("h1", "c1", reportWith(capCadenceOK), store.HostReportDenorm{})
|
||||
cc.Check()
|
||||
evs = nil
|
||||
st.SaveHostReport("h1", "c1", reportWith(capGrantAndCadenceDegraded), store.HostReportDenorm{})
|
||||
cc.Check()
|
||||
got := capTypes(evs)
|
||||
if len(got) != 2 || got[0] != "agent_capability_degraded" || got[1] != "whole_guest_backup_refused" {
|
||||
t.Fatalf("grant + cadence on one report → want [agent_capability_degraded whole_guest_backup_refused], got %v", got)
|
||||
}
|
||||
if strings.Contains(evs[0].message, "backup-cadence-cap") || strings.Contains(evs[0].details, "backup-cadence-cap") {
|
||||
t.Errorf("agent_capability_degraded must not name the cadence cap: %q / %s", evs[0].message, evs[0].details)
|
||||
}
|
||||
// Steady: nothing.
|
||||
st.SaveHostReport("h1", "c1", reportWith(capGrantAndCadenceDegraded), store.HostReportDenorm{})
|
||||
cc.Check()
|
||||
if len(evs) != 2 {
|
||||
t.Fatalf("steady both-degraded must not re-emit, got %v", capTypes(evs))
|
||||
}
|
||||
// The grant restored, the refusal still standing: only the grants recover.
|
||||
st.SaveHostReport("h1", "c1", reportWith(capCadenceRefused), store.HostReportDenorm{})
|
||||
cc.Check()
|
||||
if got := capTypes(evs); len(got) != 3 || got[2] != "agent_capability_recovered" {
|
||||
t.Fatalf("grant restored, refusal standing → want [.. agent_capability_recovered], got %v", got)
|
||||
}
|
||||
if cc.GetBackupRefusalState("h1") != "degraded" {
|
||||
t.Errorf("refusal state = %s, want still degraded", cc.GetBackupRefusalState("h1"))
|
||||
}
|
||||
}
|
||||
|
||||
// F2 for the refusal: a host already reporting a refused backup when the hub (re)starts alerts ONCE on the
|
||||
// first Check (not at construction), and never again while steady.
|
||||
//
|
||||
// COMPANION RED-PROOF (observed 2026-10-19): with the pre-fix shape restored it fails with
|
||||
//
|
||||
// host_capability_test.go: born-refused → want exactly [whole_guest_backup_refused], got [agent_capability_degraded]
|
||||
//
|
||||
// Restored.
|
||||
func TestR1150_BornRefusedAlertsOnce(t *testing.T) {
|
||||
st := newCapStore(t)
|
||||
st.SaveHostReport("h1", "c1", reportWith(capCadenceRefused), store.HostReportDenorm{})
|
||||
var evs []capEvent
|
||||
cc := NewHostCapabilityChecker(st, recordCapEvents(&evs), log.New(io.Discard, "", 0))
|
||||
if len(evs) != 0 {
|
||||
t.Fatalf("construction must not emit, got %v", capTypes(evs))
|
||||
}
|
||||
cc.Check()
|
||||
cc.Check()
|
||||
if got := capTypes(evs); len(got) != 1 || got[0] != "whole_guest_backup_refused" {
|
||||
t.Fatalf("born-refused → want exactly [whole_guest_backup_refused], got %v", got)
|
||||
}
|
||||
}
|
||||
|
||||
@@ -880,6 +880,22 @@ var operatorOnlyEvents = map[string]bool{
|
||||
// definition: someone pushed a catalog change asking for container rights the box does not give. The household can
|
||||
// do nothing; its app page says the app keeps its last good definition. Registered in the same commit that mints it.
|
||||
"app_definition_refused": true,
|
||||
// R-1150 (v0.159.0). The agent refused a whole-guest backup a guest asked for beyond its schedule (the R-973
|
||||
// cadence cap): the guest's controller may be misbehaving or compromised — the operator's question, never the
|
||||
// household's. Hub-origin only (monitor.HostCapabilityChecker; deliberately NOT in api.allowedEventTypes — the
|
||||
// suspected controller must not be able to post or pre-empt it). Listed in the SAME commit that mints them.
|
||||
"whole_guest_backup_refused": true,
|
||||
"whole_guest_backup_refusal_cleared": true,
|
||||
// R-1150 (v0.159.0): the agent's privileged-capability pair. Its checker has said „Customer is NEVER notified"
|
||||
// since v0.44.0, and only configuration (enabled_events) made that true — the R-97c shape. Now structural.
|
||||
"agent_capability_degraded": true,
|
||||
"agent_capability_recovered": true,
|
||||
// R-1154 (v0.159.0): a host-enroll that reused the old host record — re-issued the clean-slate credentials, or
|
||||
// did NOT because the box is live. Info (recorded, never mailed), host ids and report ages: the operator's
|
||||
// record. Hub-origin only (api/reenroll.go; not in api.allowedEventTypes — a box must not plant the
|
||||
// once-per-silence record). Listed in the SAME commit that mints them.
|
||||
"host_reenroll_reissued": true,
|
||||
"host_reenroll_not_reissued": true,
|
||||
}
|
||||
|
||||
// IsOperatorOnly reports whether an event type is barred from customer dispatch. Exported so the
|
||||
|
||||
@@ -0,0 +1,43 @@
|
||||
package notify
|
||||
|
||||
import "testing"
|
||||
|
||||
// R-1150 (v0.159.0) — the refused whole-guest backup pair and the agent's capability pair never reach a
|
||||
// household, under the configuration that would break it: the customer HAS the type in enabled_events and
|
||||
// HAS an e-mail. The operator still gets the warning. (The api package pins the other register: none of
|
||||
// these is in allowedEventTypes, so no box can POST one.)
|
||||
//
|
||||
// COMPANION RED-PROOF (observed 2026-10-19): with the four lines removed from operatorOnlyEvents this fails
|
||||
// with
|
||||
//
|
||||
// r1150_operator_only_test.go: whole_guest_backup_refused: the household was mailed (customer@example.com) — enabled_events opted it in
|
||||
// r1150_operator_only_test.go: agent_capability_degraded: the household was mailed (customer@example.com) — enabled_events opted it in
|
||||
//
|
||||
// Restored.
|
||||
func TestR1150_RefusedBackupAndCapabilityEventsNeverReachTheCustomer(t *testing.T) {
|
||||
for _, et := range []string{"whole_guest_backup_refused", "agent_capability_degraded"} {
|
||||
st := opOnlyStore(t)
|
||||
if err := st.SaveNotificationPrefs("c1", "customer@example.com", []string{et}, 6); err != nil {
|
||||
t.Fatalf("SaveNotificationPrefs: %v", err)
|
||||
}
|
||||
rec := &sentTo{}
|
||||
d := opOnlyDispatcher(t, st, rec)
|
||||
d.ProcessEvent("c1", et, "warning", "Host c1-abc123: test", `{"host_id":"c1-abc123"}`, "hub")
|
||||
var customer, operator bool
|
||||
for _, to := range rec.to {
|
||||
customer = customer || to == "customer@example.com"
|
||||
operator = operator || to == "operator@felhom.eu"
|
||||
}
|
||||
if customer {
|
||||
t.Errorf("%s: the household was mailed (customer@example.com) — enabled_events opted it in", et)
|
||||
}
|
||||
if !operator {
|
||||
t.Errorf("%s: the operator was not mailed — the guard must be customer-only", et)
|
||||
}
|
||||
}
|
||||
for _, et := range []string{"whole_guest_backup_refusal_cleared", "agent_capability_recovered"} {
|
||||
if !operatorOnlyEvents[et] {
|
||||
t.Errorf("%s must be operator-only (the pair's other edge)", et)
|
||||
}
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user