zakura-core/zakura v1.8.4

on GitHub

Added

• Added an optional loopback-only raw trace endpoint. GET /traces lists the
  JSONL trace tables, and GET /traces/<table> streams one table with standard
  HTTP byte ranges and an ETag, so collectors can follow a growing table with
  If-Range. It is disabled by default and requires trace_dir
  (#2072).

Changed

• Verifying a block computes each v5 and v6 transaction's ID and authorizing-data
  digest once, and reuses them for transaction verification and the block's
  authorizing-data commitment check, instead of recomputing the digests
  (#2022).
• Reduced block and transaction-expiry cleanup work by removing known mempool
  transaction IDs without rescanning the pool
  (#2048).
• Raised the minimum fee recommendation from getstandardfee to 1,000
  zatoshis per logical action
  (#2060).
• Classify the compatible performance improvements and fee recommendation
  increase as a patch node release; library crates retain their independent
  version requirements (#2074).
• Raised the default advert_window_min and advert_window_max from 32 and 64
  to 1000, so every block is advertised to up to 1000 peers at once. Relayed
  blocks previously reached the last peers 9 to 12 s late in 3 s waves. Config
  files that still hold the previous defaults, such as files made by
  zakurad generate, now load as the new defaults. Other configured values are
  kept (#2078).
• Reduced how long the legacy block advertiser waits for a peer to request an
  advertised block from 3 s to 2 s, so peers that already have the block hold
  up new-block fan-out for less time
  (#2079).

Fixed

• Fixed header sync re-requesting a VCT repair target from a peer that refused
  it before that peer advertised a new status, and capped such refusals at
  three per peer per repair episode. A peer's repair failures now survive its
  session being replaced, and hold for one readmission cooldown after it
  disconnects (#2000).
• Fixed block sync giving a peer a fresh Status allowance each time it reopened
  the block-sync stream. The connection now keeps its Status pacing across
  stream replacements
  (#2012).
• Fixed block sync advertising free download slots for a peer whose outbound
  queue was full
  (#2012).
• Fixed native discovery authoring an older self-record after a backward clock
  step across a restart, which left peers dialing the node's previous addresses
  until the old record expired. The node now stores its last authored sequence
  beside its identity key, in
  <identity_dir>/<network>.<node_id>.zakura-discovery-sequence
  (#2013).
• Fixed a native P2P peer being able to stall its connection for up to 10
  seconds by withholding stream credit while the node reopened one of its
  service sessions. The reopen now runs beside the connection's other work and
  reserves the service only once its streams are open. The connection's first
  session opens also end at once when the node closes the connection
  (#2014).
• Reduced getblocktemplate transaction-selection work for large mempools by
  avoiding repeated rebuilds of fee-sampling weights
  (#2054).
• Fixed maintained Zakura redials that credited connection setup time toward
  a healthy connection, and an alternate connection that lifted the ten-minute
  ALPN-mismatch backoff
  (#2062).
• Reduced redundant block and transaction decoding when fetching inventory over
  Zakura compatibility streams
  (#2057).