ipn/ipnlocal, control/controlclient: process node adds/removes in constant time

For large tailnets (~50k+ nodes) with frequent peer churn (ephemeral
GitHub Actions workers etc.), tailscaled used to rebuild the full
netmap and fan it out on the IPN bus on every MapResponse that
added or removed a peer. There were two O(N) costs per delta: the
full netmap rebuild + every Notify.NetMap encode to every bus watcher.

This change tackles both:

  1. Plumb O(1) peer add/remove through the delta path. PeersChanged
     and PeersRemoved no longer prevent the delta happy path; instead,
     they mutate the per-node-backend peer map in place.

  2. Restrict ipn.Notify.NetMap emission to the platforms whose host
     GUIs still depend on it (Windows, macOS, iOS) and migrate
     in-tree consumers off it everywhere else:

     - Migrate reactive consumers (containerboot, kube agents,
       sniproxy, tsconsensus, etc.) off Notify.NetMap to the
       previously-added Notify.SelfChange signal so they no longer
       have to subscribe to the full netmap.
     - Add ipn.NotifyNoNetMap so GUI clients on "legacy-emit" platforms
       that have already migrated can opt out of the per-watcher
       NetMap encode.
     - Gate Notify.NetMap emission on the producer side by a compile-
       time GOOS check, so the supporting code is dead-code-eliminated
       on Linux and other geese where no GUI consumer needs it.

Re-running BenchmarkGiantTailnet from tstest/largetailnet, which was
added along with baseline numbers on unmodified main in ad5436af0d,
the per-delta cost (one peer add+remove pair) is now ~O(1) regardless
of tailnet size N:

    N         no-watcher (ms/op)            bus-watcher (ms/op)
              before    now     factor      before    now     factor
     10000        32   0.11       300x         166   0.13      1300x
     50000       222   0.11      2000x         865   0.13      6700x
    100000       504   0.12      4100x        1765   0.13     13400x
    250000      1551   0.12     12500x        4696   0.15     32400x

Updates #12542

Change-Id: I94e34b37331d1a8ec74c299deffadf4d061fda9e
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>

ipn/ipnlocal/bus.go

@@ -5,13 +5,30 @@ package ipnlocal
 
 import (
 	"context"
+	"runtime"
 	"time"
 
 	"tailscale.com/ipn"
 	"tailscale.com/tailcfg"
 	"tailscale.com/tstime"
+	"tailscale.com/util/mak"
 )
 
+// goosGetsLegacyNetmapNotify reports whether tailscaled, when running on the
+// current GOOS, still emits the legacy [ipn.Notify.NetMap] field on runtime
+// (non-initial) bus messages. It is true on platforms whose host GUIs have
+// not yet finished migrating to the narrower bus signals
+// ([ipn.Notify.SelfChange] / [ipn.Notify.PeerChanges]) and the on-demand
+// [LocalClient.NetMap] fetch.
+//
+// runtime.GOOS is a compile-time constant, so the producer-side code that
+// builds and ships NetMap on the bus is dead-code-eliminated on Linux and
+// other geese where this is false.
+const goosGetsLegacyNetmapNotify = runtime.GOOS == "windows" ||
+	runtime.GOOS == "darwin" ||
+	runtime.GOOS == "ios" ||
+	runtime.GOOS == "android"
+
 type rateLimitingBusSender struct {
 	fn              func(*ipn.Notify) (keepGoing bool)
 	lastFlush       time.Time           // last call to fn, or zero value if none
@@ -126,11 +143,21 @@ func mergeBoringNotifies(dst, src *ipn.Notify) *ipn.Notify {
 	if dst == nil {
 		dst = &ipn.Notify{Version: src.Version}
 	}
-	if src.NetMap != nil {
+	if goosGetsLegacyNetmapNotify && src.NetMap != nil {
+		// Full netmap supersedes any accumulated peer-change deltas.
 		dst.NetMap = src.NetMap
-		dst.PeerChanges = nil // full netmap supersedes any accumulated deltas
-	} else if src.PeerChanges != nil {
-		dst.PeerChanges = mergePeerChanges(dst.PeerChanges, src.PeerChanges)
+		dst.PeerChangedPatch = nil
+	} else if src.PeerChangedPatch != nil {
+		dst.PeerChangedPatch = mergePeerChangedPatch(dst.PeerChangedPatch, src.PeerChangedPatch)
+	}
+	if len(src.PeersChanged) > 0 {
+		dst.PeersChanged = append(dst.PeersChanged, src.PeersChanged...)
+	}
+	if len(src.PeersRemoved) > 0 {
+		dst.PeersRemoved = append(dst.PeersRemoved, src.PeersRemoved...)
+	}
+	for id, up := range src.UserProfiles {
+		mak.Set(&dst.UserProfiles, id, up)
 	}
 	if src.Engine != nil {
 		dst.Engine = src.Engine
@@ -138,10 +165,10 @@ func mergeBoringNotifies(dst, src *ipn.Notify) *ipn.Notify {
 	return dst
 }
 
-// mergePeerChanges merges new peer changes from src into dst, either
-// mutating dst or allocating a new slice if dst is nil, returning the merged result.
-// Values in src override those in dst for the same NodeID.
-func mergePeerChanges(dst, src []*tailcfg.PeerChange) []*tailcfg.PeerChange {
+// mergePeerChangedPatch merges new peer-changed patches from src into dst,
+// either mutating dst or allocating a new slice if dst is nil, returning the
+// merged result. Values in src override those in dst for the same NodeID.
+func mergePeerChangedPatch(dst, src []*tailcfg.PeerChange) []*tailcfg.PeerChange {
 	idxByNode := make(map[tailcfg.NodeID]int, len(dst))
 	for i, d := range dst {
 		idxByNode[d.NodeID] = i
@@ -191,8 +218,7 @@ func mergePeerChangeForIpnBus(old, new *tailcfg.PeerChange) *tailcfg.PeerChange
 // should be sent on the IPN bus immediately (e.g. to GUIs) without
 // rate limiting it for a few seconds.
 //
-// It effectively reports whether n contains any field set that's
-// not NetMap or Engine.
+// PeerChanges and Engine are the only "boring" (rate-limitable) fields.
 func isNotableNotify(n *ipn.Notify) bool {
 	if n == nil {
 		return false
@@ -206,6 +232,7 @@ func isNotableNotify(n *ipn.Notify) bool {
 		n.ErrMessage != nil ||
 		n.LoginFinished != nil ||
 		n.SelfChange != nil ||
+		n.InitialStatus != nil ||
 		!n.DriveShares.IsNil() ||
 		n.Health != nil ||
 		len(n.IncomingFiles) > 0 ||