Release candidate 1 for 0.17.2. Changes since 0.17.1:
Added
contrib/clboss-recent-earnings --grouparranges the rows in three
groups, each sorted by net earnings: balanced, fills (peers the
rebalancer mostly topped up) and sources (peers it mostly drew
from).contrib/clboss-earnings-historyaccepts a short channel id as the
peer argument, as well as a node id or an alias.contrib/clboss-channel-sizingshows which channels want more
capacity on our side and which carry capital that never moves, from
CLBOSS's balance samples and CLN's forwards. Rows are grouped as
Grow, Shrink, Spent, Right-sized, Liquidity-limited, Too young and
Little or no traffic, with a proposed splice amount and what holds
each row back.
Changed
- The six remaining startup-only options are dynamic:
clboss-min-onchain,clboss-min-channel,clboss-max-channel,
clboss-auto-close,clboss-zerobasefeeand
clboss-min-nodes-to-processchange at runtime with
lightning-cli setconfig, so every CLBOSS option except
clboss-skip-cln-version-checkdoes. Asetconfigvalue that
the startup validation would have adjusted (an amount below its
floor, a channel-size pair the planner cannot use) or cannot
parse is refused with the reason, so the valuelightningd
persists is always the one in effect; the startup path keeps
forcing such values, so existing configuration files still load.
An unknownclboss-zerobasefeeword, which used to meanallow
silently, now keeps the current setting and logs a warning. - The contrib scripts show a peer's newest short channel id in place
of its node id when the peer has no alias. contrib/clboss-routing-statskeeps a channel awaiting splice
lock-in in its table; the channel forwards until the splice locks
in.contrib/clboss-recent-earningsprints PPM values with underscore
separators, like the amount columns.contrib/cln-plugin-bounceis the plain bounce again: ordered
stops, reverse starts. It no longer re-reads config files and
passes edited values on theplugin startline; a restarted plugin
gets the valueslightningdholds, those read at its own startup
as changed since withsetconfig. A config-file edit takes effect
whenlightningdrestarts. The warning about a configured option
that a newly installed build no longer registers stays, since
lightningdkeeps a stale configvar for it until it restarts.
Fixed
- Timestamps and other non-integer numbers in command output keep
their full precision. They were printed with six significant
digits, soclboss-statusshowed anowof 1721640000 beside a
now_humanof 08:29:41. Issue #224. - The channel creator no longer logs an error for a candidate without
a node_announcement, or whose only address is of typedns; both
count as having no address for IP binning. Issue #257. - CLBOSS again keeps at most three connections to peers without
channels. Since 0.13.2 the peer list came fromlistpeerchannels,
which has no entry for a peer without a channel, so the limit never
applied and every channel candidate that was tested for uptime
stayed connected. The peer list now takes those peers from
listpeers(#355). - The uptime test of a channel candidate no longer leaves a
connection behind: it disconnects after a successful connect, and
a candidate that is already connected counts as online without a
connect (#355). - The channel creator checks
listpeerchannelsagain right before
multifundchannel. A plan takes minutes to make and to connect
for, and a planned peer could get a channel with us in that time
(an inbound open, a manual one); it would then have been funded a
second time. - The channel creator runs one creation cycle at a time. Onchain
funds are announced on every block, and a cycle can outlast the gap
between two blocks, so a second cycle could plan the same funds for
the same peers. - With two or more channels to one peer, closing one of them no
longer announces the peer's channel as destroyed while another
is still open, which archived its complaint history, reset its
channel age and flushed its fee state. The create/destroy
monitor now tracks open channels per channel id (#354). - A cooperative close no longer reports the peer as destroyed,
re-created and destroyed again: the create/destroy monitor's
ten-minute reconciliation uses the same rule as its notifications,
so a channel shutting down does not make the peer channeled again
(#366).
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). - @johngribbin: reported the wrong human-readable dates in
clboss-status(#224). - @tsjk: reported the candidate whose only address is of type
dns
(#257).