ksedgwic/clboss v0.17.2

v0.17.2: Lost in the Flow

on GitHub

CLBOSS v0.17.2 "Lost in the Flow" is a maintenance release. Every option is now dynamic, channel candidates that used to read as unreachable get funded, a connection leak is closed, and a new experimental tool says which channels want more capacity. The requirements are those of v0.17.0: Core Lightning v26.04 or later, and the xrebalance plugin v0.4.5 or later.

Posts

• What's new in CLBOSS v0.17.2: a short summary for busy operators.

Changed

• Options. Every CLBOSS option except clboss-skip-cln-version-check is dynamic: lightning-cli setconfig changes it at runtime. A value the startup validation would have adjusted, or cannot parse, is refused with the reason, so the value lightningd persists is the one in effect (#364).
• contrib/cln-plugin-bounce is the plain bounce again, now that every option is dynamic (#370).

Fixed

• Channel candidates. The dowser measured a candidate with one probe at the largest channel size, so a candidate whose reachable flow fell between the smallest and the largest size read as no flow and was never funded. It now searches for the flow the candidate can reach (#342).
• Connections. CLBOSS again keeps at most three connections to peers without channels. Since 0.13.2 the limit never applied, and every candidate tested for uptime stayed connected. With CLN 26.06.8, which shares connectd's CPU budget across connected peers, those lingering candidates made its per-peer throttle trip on real channel peers (#355).
• Duplicate funding. The channel creator checks the channel list again right before funding and runs one creation cycle at a time, so a peer that got a channel while the plan was made is not funded again (#363).
• Two channels to one peer. Closing one of them archived the peer's complaints and reset its fee state and channel age, and a cooperative close showed the peer destroyed, re-created and destroyed again. The create/destroy monitor now tracks channels by id (#354, #366).
• Lost candidate batches. Two finders proposing the same node at the same time lost one batch. The preinvestigator now runs one connect per node and answers every waiting case (#341).
• Candidate addresses. A candidate without a node_announcement, or with only a dns address, logged an error on every cycle (#257).
• clboss-status dates. Timestamps keep their full precision; the now field was rounded to six digits (#224).

Added

• contrib/clboss-channel-sizing, experimental and work in progress: which channels want more capacity on our side and which carry capital that never moves, from CLBOSS's balance samples and CLN's forwards, with a proposed splice amount. The groups and the proposed changes are a reading of the data, not a recommendation; it is in this release to show the direction and to collect feedback.
• contrib/clboss-recent-earnings --group arranges the rows as balanced, fills and sources; contrib/clboss-earnings-history accepts a short channel id as the peer argument (#356).

Upgrading from v0.17.1

Install and restart. Nothing needs setting, and no option was removed.

Credits

Thanks to the contributors to this release:

• @Mohil-Ahuja: the create/destroy monitor's per-channel liveness (#365) and its reconciliation by the monitor's own rule (#367).
• @Amperstrand: the preinvestigator's one connect per node for concurrent cases (#341) and the dowser's search for the flow reachable over several paths (#342).
• @johngribbin: reported the wrong human-readable dates in clboss-status (#224).
• @tsjk: reported the candidate whose only address is of type dns (#257), and the gossip-only peers CLBOSS kept connected (#355).

See CHANGELOG.md for details.