Files
felhom.eu/hub/internal/intent/hub.go
T
admin 60244727ad feat(hub): Direction-2 immediate-sync wait channel (v0.58.0)
GET /api/v1/wait long-poll: the box holds an authed hanging GET; the hub
completes it the instant any operator intent bumps that customer's in-memory
generation, then the box fires its ordinary report and the ACK delivers
everything through the unchanged machinery. 240s hold with a 25s heartbeat
newline defeats the nginx 60s proxy_read_timeout with no ingress annotation;
WriteTimeout lifted per-connection via ResponseController.

- internal/intent: per-customer generation counter + waiter registry
  (Bump/Wait/Close), coalescing to latest, race-closer, in-memory by design.
  Red-proofs: counter-vs-queue + race-closer (run-fail-reverted).
- api/wait.go: the endpoint (per-customer only; global key 400; A cannot see B).
- web bumps after every intent write (fire-after-commit): config CRUD, claim
  resend, offsite re-issue/freeze, password regen, block/unblock, floors
  (global bumps all config-managed), controller log-tail + log-bundle.
- main.go: one intent hub shared by web+api; Close() before server.Shutdown.

Pairs with controller v0.140.0 (the long-poll client). Grounding:
documentation/audits/SPIKE-immediate-sync-transport-2026-07-16.md.
2026-07-16 20:44:22 +02:00

144 lines
5.2 KiB
Go

// Package intent is the hub's in-memory operator-intent notifier — the server side of the
// Direction-2 immediate-sync arc (SPIKE-immediate-sync-transport-2026-07-16.md, option b).
//
// Every operator action that changes a customer's desired state (config save/delete, claim
// resend, offsite re-issue/freeze, floor change, block/unblock, log-pull request) Bumps that
// customer's generation. A box holds a hanging GET /api/v1/wait against the hub carrying its
// last-seen generation; Wait returns the instant the generation moves (or on timeout/shutdown).
// The box then fires its ordinary out-of-cycle report — the ACK delivers everything through the
// UNCHANGED report machinery. The wait signal is CONTENTLESS: it carries a generation number
// and nothing else (no config, no escrow, no claim).
//
// The counter is IN-MEMORY BY DESIGN — a restart resets all generations to zero, which the box
// treats as "a different generation" (it compares with != , not >), firing exactly one harmless
// full-state report before settling. No persistence, no store schema (Part 2.1 / §5).
//
// Semantics mirror the wgsync reconciler / controller report.Trigger: a per-customer counter,
// coalescing (N bumps during one hold → one completion carrying the LATEST generation), never a
// per-bump queue. Every Wait exit path deregisters its channel, so there is no waiter leak.
package intent
import (
"context"
"sync"
"time"
)
// Hub is the in-memory per-customer generation counter + waiter registry. The zero value is not
// usable; construct with New. All methods are safe for concurrent use.
type Hub struct {
mu sync.Mutex
gen map[string]uint64 // customerID -> current generation (0 = never bumped)
waiters map[string][]chan struct{} // customerID -> registered wake channels (buffered, cap 1)
closed chan struct{} // closed by Close() to wake every waiter
isClosed bool
}
// New builds an empty intent hub.
func New() *Hub {
return &Hub{
gen: make(map[string]uint64),
waiters: make(map[string][]chan struct{}),
closed: make(chan struct{}),
}
}
// Bump advances the customer's generation and wakes every currently-registered waiter for that
// customer, then clears them (each waiter re-reads the LATEST generation under the lock on wake,
// so a burst of bumps collapses into one completion carrying the final value — a counter, not a
// queue). Safe from any goroutine; never blocks. An empty customerID is a no-op, as is any bump
// after Close.
func (h *Hub) Bump(customerID string) {
if customerID == "" {
return
}
h.mu.Lock()
defer h.mu.Unlock()
if h.isClosed {
return
}
h.gen[customerID]++
for _, ch := range h.waiters[customerID] {
// Non-blocking send into a cap-1 channel: the waiter reads the latest gen under the lock,
// so a single wake is all it needs; a second bump before the waiter runs is harmless.
select {
case ch <- struct{}{}:
default:
}
}
delete(h.waiters, customerID)
}
// Wait blocks until the customer's generation differs from lastSeen, ctx is cancelled, maxHold
// elapses, or Close is called — whichever first — and returns the customer's CURRENT generation
// to hand back to the caller. If the generation already differs from lastSeen at entry (a bump
// landed before this wait registered), it returns immediately: the race gap between a bump and a
// (re)connect is closed by this pre-register check, so a bump is never lost.
func (h *Hub) Wait(ctx context.Context, customerID string, lastSeen uint64, maxHold time.Duration) uint64 {
h.mu.Lock()
if h.isClosed {
cur := h.gen[customerID]
h.mu.Unlock()
return cur
}
if cur := h.gen[customerID]; cur != lastSeen {
h.mu.Unlock()
return cur // race-closer: intent already moved; no need to hold
}
ch := make(chan struct{}, 1)
h.waiters[customerID] = append(h.waiters[customerID], ch)
h.mu.Unlock()
timer := time.NewTimer(maxHold)
defer timer.Stop()
select {
case <-ch:
case <-ctx.Done():
case <-timer.C:
case <-h.closed:
}
return h.finish(customerID, ch)
}
// finish deregisters ch (if a Bump/Close hasn't already cleared it) and returns the customer's
// current — i.e. latest — generation. Called on every Wait exit path, so waiters never leak.
func (h *Hub) finish(customerID string, ch chan struct{}) uint64 {
h.mu.Lock()
defer h.mu.Unlock()
ws := h.waiters[customerID]
for i, c := range ws {
if c == ch {
h.waiters[customerID] = append(ws[:i], ws[i+1:]...)
break
}
}
if len(h.waiters[customerID]) == 0 {
delete(h.waiters, customerID)
}
return h.gen[customerID]
}
// Close wakes every registered waiter (each returns its customer's current generation) and makes
// all subsequent Wait calls return immediately. Idempotent. Wired into the hub's graceful
// shutdown BEFORE server.Shutdown so held waits complete instantly instead of eating the grace
// window.
func (h *Hub) Close() {
h.mu.Lock()
defer h.mu.Unlock()
if h.isClosed {
return
}
h.isClosed = true
close(h.closed)
h.waiters = make(map[string][]chan struct{})
}
// Generation returns the customer's current generation (0 if never bumped). Test/observability
// helper — the wait path uses the value returned by Wait directly.
func (h *Hub) Generation(customerID string) uint64 {
h.mu.Lock()
defer h.mu.Unlock()
return h.gen[customerID]
}