fix: start/stop for ring-only forwards + group sync via ring topology

Three issues fixed:

1. Start/stop returned 404 for forwards owned by other nodes:
   findLinkByTriple only checked local SQLite store. Added
   findLinkFromTopology fallback — startForward/stopForward now look up
   ring topology entries when the local store doesn't have the link.

2. Group labels didn't sync through the ring:
   handleForwardsAssign only updated local SQLite, never the ring
   topology. Added State.UpdateTopologyGroup + Engine.UpdateTopologyGroup
   — handleForwardsAssign now updates both. Added topologySync callback
   called after every e.state = tk.State (OnToken) and e.state = s
   (AdoptState) to re-apply local store group overrides onto the freshly
   adopted topology, so they survive state adoption and propagate via
   the next token forward. handleForwardsGroupDelete also clears ring
   topology entries.

3. Status-merge branch omitted Group field:
   ring-only forwards always showed '未分组'. Added Group: t.Link.Group
   to the topology-merge forwardStatus.

Verified: 4-node cluster, 5 forwards, group assigned on node-a
propagates to all nodes within one token cycle; stop from non-owning
node succeeds (HTTP 200) and worker stops on the owning node.
This commit is contained in:
2026-08-19 21:52:28 +08:00
parent 465a86c755
commit 2552c14faf
4 changed files with 110 additions and 39 deletions

View File

@ -404,6 +404,20 @@ func (s *State) ForwardsOwnedBy(owner string) []*TopoEntry {
return out
}
// UpdateTopologyGroup sets the group label on the topology entry matching the
// (local, remote, port) triple. Returns true if found. The updated entry
// propagates to all nodes via the next token cycle — group changes sync
// through the ring without a dedicated command.
func (s *State) UpdateTopologyGroup(local, remote string, port int, group string) bool {
for _, e := range s.Topology {
if e.Local.Name == local && e.Remote.Name == remote && e.Link.RemotePort == port {
e.Link.Group = group
return true
}
}
return false
}
// OfflineReassign moves all active forwards owned by an offline node back into
// PendingTasks (they become new tasks for the next lowest-load member). Returns
// the reassigned task ids.

View File

@ -44,6 +44,12 @@ type Engine struct {
// every token cycle (OnToken) and on AdoptState. Cleared (pass "") on
// detachAsStandalone — an explicit leave must NOT auto-rejoin.
peerPersist func(peersJSON string) error
// topologySync is called right after e.state = tk.State (OnToken) or
// e.state = s (AdoptState) so the host can re-apply local store overrides
// (e.g. group labels) onto the freshly adopted topology. Without this,
// group changes made via HTTP handlers between token cycles are overwritten
// by the next state adoption and never propagate to other nodes.
topologySync func()
state State
// myAddr maps our Node ID to the address peers dial.
@ -219,6 +225,12 @@ func (e *Engine) OnToken(ctx context.Context, tk *Token) (*Token, error) {
if e.state.PendingTasks == nil {
e.state.PendingTasks = map[string]*Task{}
}
// Re-apply local store overrides (group labels, disabled flags) onto
// the freshly adopted topology so they survive state adoption and
// propagate to all nodes via the next token forward.
if e.topologySync != nil {
e.topologySync()
}
for id, t := range localPending {
e.state.PendingTasks[id] = t
}
@ -695,6 +707,9 @@ func (e *Engine) AdoptState(s State) {
for id, t := range kept {
e.state.PendingTasks[id] = t
}
if e.topologySync != nil {
e.topologySync()
}
e.state.UpsertNode(Node{ID: e.ID, Addr: e.myAddr, Alive: true,
Load: e.loadSnapshot(), Version: e.Version, Cache: e.Cache, NodeKey: e.nodeKey})
if e.Log != nil {
@ -780,6 +795,13 @@ func (e *Engine) HasTask(local, remote string, port int) bool {
return false
}
// UpdateTopologyGroup sets the group label on a topology entry. The change
// propagates to all nodes via the next token cycle. Used by the forwards
// assign handler so group labels sync through the ring.
func (e *Engine) UpdateTopologyGroup(local, remote string, port int, group string) bool {
return e.state.UpdateTopologyGroup(local, remote, port, group)
}
// IsLeader reports whether this node is the current ring leader.
func (e *Engine) IsLeader() bool { return e.state.LeaderID == e.ID }
@ -796,6 +818,12 @@ func (e *Engine) SetKeyPersist(fn func(key string) error) { e.keyPersist = fn }
// to durable storage (store.SetClusterPeers). Called once from main.go.
func (e *Engine) SetPeerPersist(fn func(peersJSON string) error) { e.peerPersist = fn }
// SetTopologySync installs the callback used to re-apply local store overrides
// (group labels, disabled flags) onto the ring topology after each state
// adoption. Called once from main.go. Without this, group changes made via
// HTTP handlers are overwritten by the next e.state = tk.State.
func (e *Engine) SetTopologySync(fn func()) { e.topologySync = fn }
// persistPeers extracts all alive peers (addr + nodeKey, excluding self)
// from the current ring state and persists them via the peerPersist callback.
// Called on every token cycle (OnToken) and on AdoptState so a crashed node

View File

@ -36,34 +36,44 @@ func findLinkByTriple(s *store.Store, local, remote string, port int) (store.Lin
return store.Link{}, false
}
// findLinkFromTopology looks up a forward's local/remote/link data from the
// ring topology when it's not in the local SQLite store (forward owned by
// another node). Returns the store.Local, store.Remote, store.Link, and ok.
func (h *Handler) findLinkFromTopology(local, remote string, port int) (store.Local, store.Remote, store.Link, bool) {
if h.Ring == nil {
return store.Local{}, store.Remote{}, store.Link{}, false
}
snap := h.Ring.Snapshot()
for _, t := range snap.Topology {
if t.Local.Name == local && t.Remote.Name == remote && t.Link.RemotePort == port {
return t.Local, t.Remote, t.Link, true
}
}
return store.Local{}, store.Remote{}, store.Link{}, false
}
// startForward flips a forward to enabled and brings it up. A local-only
// forward restarts/starts its local frpc worker so the re-enabled proxy is
// rendered back in; a cluster (non-localOnly) forward is submitted to the ring
// so the lowest-load member claims and spawns it (plan §令牌环协议). SubmitTask
// is idempotent via HasTask. Returns store.ErrNotFound if the forward is gone.
// is idempotent via HasTask. Falls back to ring topology when the forward
// exists only in the cluster (owned by another node, not in local store).
func (h *Handler) startForward(local, remote string, port int) error {
ln, ok := findLinkByTriple(h.Store, local, remote, port)
if !ok {
return store.ErrNotFound
}
loc, ok := h.Store.GetLocal(local)
if !ok {
return store.ErrNotFound
}
rem, ok := h.Store.GetRemote(remote)
if !ok {
return store.ErrNotFound
loc, lok := h.Store.GetLocal(local)
rem, rok := h.Store.GetRemote(remote)
if !ok || !lok || !rok {
// Fall back to ring topology for forwards owned by other nodes.
tLoc, tRem, tLn, tok := h.findLinkFromTopology(local, remote, port)
if !tok {
return store.ErrNotFound
}
ln, loc, rem = tLn, tLoc, tRem
}
_ = h.Store.SetLinkDisabled(local, remote, port, false)
// Reflect the post-start (enabled) state in the snapshot handed to the
// ring: ln was fetched before SetLinkDisabled, so for a re-start of a
// previously stopped forward it still carries disabled=true. Without this
// the topology entry would embed a stale disabled flag (cosmetically
// wrong, and confusing if anything reads topology's link snapshot).
ln.Disabled = false
if loc.LocalOnly {
if h.Process != nil {
// Re-render (proxy re-added) on a running worker, or start it.
if _, has := h.Process.Status(remote); has {
_ = h.Process.Restart(remote)
} else {
@ -82,23 +92,22 @@ func (h *Handler) startForward(local, remote string, port int) error {
// forward restarts its worker so renderRemote omits the proxy (sibling forwards
// on the same remote keep running — true per-forward stop); a cluster forward
// is revoked so the owning node cancels its worker and drops it from the
// topology (plan §任务撤销). RevokeTask is idempotent.
// topology (plan §任务撤销). RevokeTask is idempotent. Falls back to ring
// topology when the forward exists only in the cluster (owned by another node).
func (h *Handler) stopForward(local, remote string, port int) error {
ln, ok := findLinkByTriple(h.Store, local, remote, port)
if !ok {
return store.ErrNotFound
loc, lok := h.Store.GetLocal(local)
rem, rok := h.Store.GetRemote(remote)
if !ok || !lok || !rok {
tLoc, tRem, tLn, tok := h.findLinkFromTopology(local, remote, port)
if !tok {
return store.ErrNotFound
}
ln, loc, rem = tLn, tLoc, tRem
} else {
_ = h.Store.SetLinkDisabled(local, remote, port, true)
}
loc, ok := h.Store.GetLocal(local)
if !ok {
return store.ErrNotFound
}
rem, ok := h.Store.GetRemote(remote)
if !ok {
return store.ErrNotFound
}
_ = h.Store.SetLinkDisabled(local, remote, port, true)
if loc.LocalOnly {
// Re-render without this proxy; only meaningful while a worker runs.
if h.Process != nil {
if _, has := h.Process.Status(remote); has {
_ = h.Process.Restart(remote)
@ -217,9 +226,9 @@ type assignReq struct {
// handleForwardsAssign changes the group label of a single forward. The
// status page group chip is the quick entry; the canvas port editor also
// carries a group field (both persist via the same store column). The
// change is purely a DB update — no worker/ring action is needed (group is
// a management label, not a runtime knob).
// carries a group field (both persist via the same store column). The local
// DB row is updated if present; the ring topology entry is always updated so
// the group label propagates to all nodes via the next token cycle.
func (h *Handler) handleForwardsAssign(w http.ResponseWriter, r *http.Request) {
if r.Method != http.MethodPost {
methodNotAllowed(w)
@ -234,12 +243,21 @@ func (h *Handler) handleForwardsAssign(w http.ResponseWriter, r *http.Request) {
http.Error(w, "local/remote/remotePort required", http.StatusBadRequest)
return
}
if _, ok := findLinkByTriple(h.Store, req.Local, req.Remote, req.RemotePort); !ok {
http.Error(w, store.ErrNotFound.Error(), http.StatusNotFound)
return
found := false
if _, ok := findLinkByTriple(h.Store, req.Local, req.Remote, req.RemotePort); ok {
found = true
if err := h.Store.SetLinkGroup(req.Local, req.Remote, req.RemotePort, req.Group); err != nil {
http.Error(w, err.Error(), http.StatusInternalServerError)
return
}
}
if err := h.Store.SetLinkGroup(req.Local, req.Remote, req.RemotePort, req.Group); err != nil {
http.Error(w, err.Error(), http.StatusInternalServerError)
if h.Ring != nil {
if h.Ring.UpdateTopologyGroup(req.Local, req.Remote, req.RemotePort, req.Group) {
found = true
}
}
if !found {
http.Error(w, store.ErrNotFound.Error(), http.StatusNotFound)
return
}
writeJSON(w, http.StatusOK, map[string]any{"ok": true})
@ -248,7 +266,8 @@ func (h *Handler) handleForwardsAssign(w http.ResponseWriter, r *http.Request) {
// handleForwardsGroupDelete dissolves a group: every forward in the named
// group is moved to 未分组 (grp=""). The group itself is not a stored entity
// — it exists only as a label on links — so clearing all members is the
// complete "delete". No worker/ring action (group is a management label).
// complete "delete". Clears both local store links and ring topology entries
// so the change syncs to all nodes.
func (h *Handler) handleForwardsGroupDelete(w http.ResponseWriter, r *http.Request) {
if r.Method != http.MethodPost {
methodNotAllowed(w)
@ -274,5 +293,14 @@ func (h *Handler) handleForwardsGroupDelete(w http.ResponseWriter, r *http.Reque
}
_ = h.Store.SetLinkGroup(ln.Local, ln.Remote, ln.RemotePort, "")
}
// Also clear group on ring topology entries (forwards owned by other nodes).
if h.Ring != nil {
snap := h.Ring.Snapshot()
for _, t := range snap.Topology {
if t.Link.Group == req.Group {
h.Ring.UpdateTopologyGroup(t.Local.Name, t.Remote.Name, t.Link.RemotePort, "")
}
}
}
writeJSON(w, http.StatusOK, map[string]any{"ok": true})
}

View File

@ -370,6 +370,7 @@ func (h *Handler) handleStatus(w http.ResponseWriter, r *http.Request) {
Local: t.Local.Name, Remote: t.Remote.Name, RemotePort: t.Link.RemotePort,
LocalOnly: false, Kind: "remote", OwnerID: t.OwnerID, Active: t.Active,
LocalIP: t.Local.IP, LocalPort: t.Local.Port, LocalProto: t.Local.Protocol,
Group: t.Link.Group,
})
seenFwd[k] = true
}