mirror of
https://github.com/tailscale/tailscale.git
synced 2026-05-05 04:06:35 +02:00
dfdf61c383
sortedPeers() copies the entire peers map into a slice and sorts it on every call. With large peer sets this is expensive — at 50k peers, each call costs ~9ms and 402KB for the MapValues copy and sort. Cache the sorted result in sortedPeersCache, protected by peersMu. The cache is invalidated (set to nil) at the start of updatePeersStateFromResponse, which is the only writer of the peers map. sortedPeers uses a read-lock fast path with double-checked locking on upgrade to write lock. At 50,000 peers: - Cache hit (repeated calls): 17.7ns, 0 allocs (was 9.2ms, 402KB) - After peer change: 9.2ms (unchanged — must re-sort) Fixes tailscale/corp#40144 Signed-off-by: Mike O'Driscoll <mikeo@tailscale.com>