lmcp 0.5.3-1: hub hardening — hard ssh timeout, parallel probes, sticky DOWN cache

This commit is contained in:
2026-04-21 14:00:28 +02:00
parent ef7911dd27
commit c07d1f846c
2 changed files with 19 additions and 2 deletions
+2 -2
View File
@@ -3,7 +3,7 @@
# Source of truth: git.reauktion.de/marfrit/lmcp
pkgname=lmcp
pkgver=0.5.2
pkgver=0.5.3
pkgrel=1
pkgdesc="Lightweight MCP (Model Context Protocol) server in pure Lua"
arch=('any')
@@ -11,7 +11,7 @@ url="https://git.reauktion.de/marfrit/lmcp"
license=('MIT')
depends=('lua' 'lua-socket')
source=("${pkgname}-${pkgver}.tar.gz::https://git.reauktion.de/marfrit/lmcp/archive/v${pkgver}.tar.gz")
sha256sums=('5a5945e6f2511f440de8e961b82ec3deaab5e77cc9208c9999e831fbc33d702d')
sha256sums=('9ee3acf3139db375f79f1ce19f97386ba017a4ed24f3bb6be9b90dc5a2c10bb4')
package() {
cd "${pkgname}"
+17
View File
@@ -1,3 +1,20 @@
lmcp (0.5.3-1) bookworm trixie; urgency=medium
* hub.lua: hard wall-clock cap on ssh fallback via `timeout` —
prevents event-loop wedges on half-dead sshd (auth stall, remote
bash-s hang). Configurable via LMCP_HUB_SSH_HARD_TIMEOUT.
* hub.lua: parallel lmcp probes for remote_list_hosts. Wall clock
for a 12-backend fleet with 8 offline hosts drops from ~28 s
(sequential ssh timeouts) to ~3 s.
* hub.lua: probe is lmcp-only, not ssh. Actual remote_* calls still
fall through to ssh fallback when lmcp is down.
* hub.lua: sticky DOWN cache with exponential backoff (60 → 120 →
240 → 480 → 900 s). UP hosts still 30 s TTL.
* hub.lua: per-request stderr logging (tool/host/via/elapsed) for
future RCA.
-- Markus Fritsche <mfritsche@reauktion.de> Tue, 21 Apr 2026 14:10:00 +0000
lmcp (0.5.2-1) bookworm trixie; urgency=medium
* server.lua: new shell_bg tool launches detached commands via