* test(server): cover recovered-node admission boundaries
Document backend readiness and write-availability checks during rejoin.
Extend the existing regression through the production middleware order
without changing runtime admission or cluster-quorum semantics.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
* fix(server): bound rejoin waits and preserve recovery ownership
Enforce startup and unreachable-grace bounds while consuming peer replies.
Release admission when every set has its required peer majority and cancel
surplus read-only probes without carrying votes between rounds.
Bound local readiness observations and polling sleeps. Retain initialized
IAM across preparation failures, install the context once, and retry only
publication after readiness timeouts. Share preparation with inline startup.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
* perf(server): make peer admission folding incremental
Compile peer-to-set memberships once and retain compact per-edge votes.
Consume each reply once, discard its payload, and update only affected
set counters instead of rescanning all sets and retained responses.
Recheck the captured deadline after polling and folding a reply so late
work cannot admit the node or clear an expired unreachable grace period.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
---------
Co-authored-by: heihutu <heihutu@gmail.com>
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Hauser Zhi
32238a179c