Files
netbird/client/internal/peer/mailbox.go
Zoltan Papp 5f98524e02 [client] Refactor peer Conn to a single-owner event loop
Replace the mutex-guarded callback model of peer.Conn with a per-peer
event loop that exclusively owns all mutable connection state. External
callers and transport workers post typed events into a non-blocking,
coalescing mailbox instead of contending on conn.mu:

- offers/answers coalesce to the newest message, a new offer flushes
  queued candidates of the superseded session
- candidates are applied in arrival order from a bounded FIFO
- transport state changes are never dropped
- the blocking relay dial runs on a helper goroutine with a single dial
  in flight; signaling I/O (offer/answer sends) runs off the loop

conn.mu now only guards the open/close lifecycle. Close posts a close
event and waits for the loop teardown; the loop also tears down on
engine context cancellation and releases resources of unprocessed
events.

Delete the Handshaker listener machinery (Listen loop, unbuffered
drop-on-busy channels, AsyncOfferListener with its double-processing of
the first offer), the never-wired dispatcher package and the unused
ICEMonitor.ReconnectCh. Fix a goroutine leak in the WG watcher test
that raced with tests mutating the package-level check timing vars.
2026-07-11 13:30:06 +02:00

117 lines
2.7 KiB
Go

package peer
import (
"sync"
)
// maxQueuedCandidates bounds the remote candidate queue; on overflow the
// oldest candidate is dropped. Lost candidates are recovered by the next
// offer exchange triggered by the guard.
const maxQueuedCandidates = 128
// mailbox is the coalescing inbox of the Conn event loop. Posting never
// blocks. Per message kind either the latest value wins (offer, answer,
// guard tick), the values queue in bounded FIFO order (candidates) or in
// unbounded FIFO order (lifecycle and transport state changes, which are
// low-volume and must not be lost). A new offer flushes the queued
// candidates because they belong to the superseded session.
type mailbox struct {
mu sync.Mutex
closed bool
lifecycle []event
transport []event
offer *evRemoteOffer
answer *evRemoteAnswer
candidates []evRemoteCandidate
guardTick bool
wake chan struct{}
}
func newMailbox() *mailbox {
return &mailbox{
wake: make(chan struct{}, 1),
}
}
// post stores the event and wakes the loop. It reports false if the mailbox
// is already closed and the event was not accepted.
func (m *mailbox) post(ev event) bool {
m.mu.Lock()
if m.closed {
m.mu.Unlock()
return false
}
switch e := ev.(type) {
case evClose:
m.lifecycle = append(m.lifecycle, e)
case evRemoteOffer:
m.offer = &e
m.candidates = nil
case evRemoteAnswer:
m.answer = &e
case evRemoteCandidate:
if len(m.candidates) >= maxQueuedCandidates {
m.candidates = m.candidates[1:]
}
m.candidates = append(m.candidates, e)
case evGuardTick:
m.guardTick = true
default:
m.transport = append(m.transport, ev)
}
m.mu.Unlock()
select {
case m.wake <- struct{}{}:
default:
}
return true
}
// drain returns the pending events in processing order: lifecycle first,
// then transport state changes, the coalesced offer and answer, the queued
// candidates and finally the guard tick.
func (m *mailbox) drain() []event {
m.mu.Lock()
defer m.mu.Unlock()
return m.drainLocked()
}
// closeAndDrain marks the mailbox closed so further posts are rejected and
// returns the events that were still pending.
func (m *mailbox) closeAndDrain() []event {
m.mu.Lock()
defer m.mu.Unlock()
m.closed = true
return m.drainLocked()
}
func (m *mailbox) drainLocked() []event {
evs := make([]event, 0, len(m.lifecycle)+len(m.transport)+len(m.candidates)+3)
evs = append(evs, m.lifecycle...)
evs = append(evs, m.transport...)
if m.offer != nil {
evs = append(evs, *m.offer)
}
if m.answer != nil {
evs = append(evs, *m.answer)
}
for _, c := range m.candidates {
evs = append(evs, c)
}
if m.guardTick {
evs = append(evs, evGuardTick{})
}
m.lifecycle = nil
m.transport = nil
m.offer = nil
m.answer = nil
m.candidates = nil
m.guardTick = false
return evs
}