drop ECN support for this release

2026-08-16 00:47:02 +02:00 · 2026-07-31 13:40:34 -05:00
parent 93946faf7a
commit 549de9fd29
1 changed files with 0 additions and 17 deletions
@@ -415,23 +415,6 @@ logging:
  # This setting is reloadable
  #inactivity_timeout: 10m
  # ecn (default true) propagates ECN (Explicit Congestion Notification) across the tunnel per RFC 6040: the inner
  # packet's ECN codepoint is copied onto the outer carrier header on encapsulation, and an outer CE ("congestion
  # experienced") mark is folded back into the inner header on decapsulation. On linux it additionally stamps
  # RTAX_FEATURE_ECN on the routes nebula installs, so the kernel actively negotiates ECN for connections to mesh
  # prefixes. Disable this only when an underlay middlebox mangles or clears ECN bits unpredictably.
  # This setting is reloadable, BUT flipping it at runtime only updates the datapath (the inner<->outer copy/combine).
  # The RTAX_FEATURE_ECN flag on already-installed routes is NOT revisited on reload, so nebula must be restarted for
  # the route half of this setting to take effect.
  #ecn: true
  # EXPERIMENTAL, Linux only. ecn_mark_threshold turns nebula into the AQM for its own receive queue — the one
  # congested hop on a tunnel path that no kernel AQM can see. When the UDP receive queue's depth exceeds this
  # fraction of the receive buffer (see listen.read_buffer), decapsulated ECT packets are CE-marked so ECN-capable
  # senders back off before the queue overflows and regulates by tail-drop (loss) instead. 0 disables (default).
  # Requires `ecn: true` end to end. Sampled once per receive batch. Reloadable.
  #ecn_mark_threshold: 0.05
 # Nebula security group configuration
 firewall:
  # Action to take when a packet is not allowed by the firewall rules.