state/nodestore: in memory representation of nodes

Initial work on a nodestore which stores all of the nodes
and their relations in memory with relationship for peers
precalculated.

It is a copy-on-write structure, replacing the "snapshot"
when a change to the structure occurs. It is optimised for reads,
and while batches are not fast, they are grouped together
to do less of the expensive peer calculation if there are many
changes rapidly.

Writes will block until commited, while reads are never
blocked.

Signed-off-by: Kristoffer Dalby <kristoffer@tailscale.com>
This commit is contained in:
Kristoffer Dalby
2025-07-05 23:30:47 +02:00
committed by Kristoffer Dalby
parent 38be30b6d4
commit 9d236571f4
35 changed files with 3960 additions and 1317 deletions

View File

@@ -137,9 +137,10 @@ func NewHeadscale(cfg *types.Config) (*Headscale, error) {
// Initialize ephemeral garbage collector
ephemeralGC := db.NewEphemeralGarbageCollector(func(ni types.NodeID) {
node, err := app.state.GetNodeByID(ni)
if err != nil {
log.Err(err).Uint64("node.id", ni.Uint64()).Msgf("failed to get ephemeral node for deletion")
node, ok := app.state.GetNodeByID(ni)
if !ok {
log.Error().Uint64("node.id", ni.Uint64()).Msg("Ephemeral node deletion failed")
log.Debug().Caller().Uint64("node.id", ni.Uint64()).Msg("Ephemeral node deletion failed because node not found in NodeStore")
return
}
@@ -379,15 +380,8 @@ func (h *Headscale) httpAuthenticationMiddleware(next http.Handler) http.Handler
writer http.ResponseWriter,
req *http.Request,
) {
log.Trace().
Caller().
Str("client_address", req.RemoteAddr).
Msg("HTTP authentication invoked")
authHeader := req.Header.Get("authorization")
if !strings.HasPrefix(authHeader, AuthPrefix) {
log.Error().
if err := func() error {
log.Trace().
Caller().
Str("client_address", req.RemoteAddr).
Msg(`missing "Bearer " prefix in "Authorization" header`)
@@ -501,11 +495,12 @@ func (h *Headscale) createRouter(grpcMux *grpcRuntime.ServeMux) *mux.Router {
// Serve launches the HTTP and gRPC server service Headscale and the API.
func (h *Headscale) Serve() error {
var err error
capver.CanOldCodeBeCleanedUp()
if profilingEnabled {
if profilingPath != "" {
err := os.MkdirAll(profilingPath, os.ModePerm)
err = os.MkdirAll(profilingPath, os.ModePerm)
if err != nil {
log.Fatal().Err(err).Msg("failed to create profiling directory")
}
@@ -559,12 +554,9 @@ func (h *Headscale) Serve() error {
// around between restarts, they will reconnect and the GC will
// be cancelled.
go h.ephemeralGC.Start()
ephmNodes, err := h.state.ListEphemeralNodes()
if err != nil {
return fmt.Errorf("failed to list ephemeral nodes: %w", err)
}
for _, node := range ephmNodes {
h.ephemeralGC.Schedule(node.ID, h.cfg.EphemeralNodeInactivityTimeout)
ephmNodes := h.state.ListEphemeralNodes()
for _, node := range ephmNodes.All() {
h.ephemeralGC.Schedule(node.ID(), h.cfg.EphemeralNodeInactivityTimeout)
}
if h.cfg.DNSConfig.ExtraRecordsPath != "" {
@@ -794,23 +786,14 @@ func (h *Headscale) Serve() error {
continue
}
changed, err := h.state.ReloadPolicy()
changes, err := h.state.ReloadPolicy()
if err != nil {
log.Error().Err(err).Msgf("reloading policy")
continue
}
if changed {
log.Info().
Msg("ACL policy successfully reloaded, notifying nodes of change")
h.Change(changes...)
err = h.state.AutoApproveNodes()
if err != nil {
log.Error().Err(err).Msg("failed to approve routes after new policy")
}
h.Change(change.PolicySet)
}
default:
info := func(msg string) { log.Info().Msg(msg) }
log.Info().
@@ -1020,6 +1003,6 @@ func readOrCreatePrivateKey(path string) (*key.MachinePrivate, error) {
// Change is used to send changes to nodes.
// All change should be enqueued here and empty will be automatically
// ignored.
func (h *Headscale) Change(c change.ChangeSet) {
h.mapBatcher.AddWork(c)
func (h *Headscale) Change(cs ...change.ChangeSet) {
h.mapBatcher.AddWork(cs...)
}