From da186c5404718c8861dc8163c2ec2f3c2e9a551c Mon Sep 17 00:00:00 2001
From: Joseph Yaksich <gitcommit90@users.noreply.github.com>
Date: Wed, 29 Jul 2026 20:54:14 +0000
Subject: [PATCH] Retry transient fresh LXC starts

---
 CHANGELOG.md              |  5 +++++
 scripts/1helm-lxc-runtime | 27 +++++++++++++++++++++------
 test/site.mjs             |  1 +
 3 files changed, 27 insertions(+), 6 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 76c8dd3..c5669af 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -32,6 +32,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   bootstrap. A channel no longer repeatedly creates and cleans up its computer
   after starting `apt` during the guest-network race.
 
+- A fresh channel now verifies and retries the same container when LXC's first
+  daemonized start transiently loses its state-socket reply. The one-off
+  `wait_on_daemonized_start` error no longer makes the channel appear broken
+  while the automatic fleet pass successfully creates it moments later.
+
 - 1Helm's narrow owned LXC forwarding rules are inserted ahead of host firewall
   policies such as Docker's `FORWARD=DROP`, allowing resident guests to reach
   package mirrors without replacing Docker or Tailscale chains.
diff --git a/scripts/1helm-lxc-runtime b/scripts/1helm-lxc-runtime
index 6a90ba0..cbd652c 100755
--- a/scripts/1helm-lxc-runtime
+++ b/scripts/1helm-lxc-runtime
@@ -78,13 +78,28 @@ cpu_count() {
   printf '%s' "$total"
 }
 start() {
-  [[ "$(state "$1")" == "RUNNING" ]] || lxc-start -P "$LXC_PATH" -n "$1" -d
-  local attempt
-  for attempt in {1..200}; do
-    [[ "$(state "$1")" == "RUNNING" ]] && return
-    sleep 0.1
+  local name="$1" start_attempt state_attempt detail=""
+  [[ "$(state "$name")" == "RUNNING" ]] && return
+  # A freshly unpacked container can hit LXC's daemonized-start socket race:
+  # lxc-start exits after wait_on_daemonized_start loses the state message even
+  # though the container definition is intact and immediately startable. Check
+  # the authoritative state after every invocation and retry that same exact
+  # validated machine instead of failing channel creation on the transient.
+  for start_attempt in {1..3}; do
+    if ! detail="$(lxc-start -P "$LXC_PATH" -n "$name" -d 2>&1)"; then
+      for state_attempt in {1..50}; do
+        [[ "$(state "$name")" == "RUNNING" ]] && return
+        sleep 0.1
+      done
+      continue
+    fi
+    for state_attempt in {1..200}; do
+      [[ "$(state "$name")" == "RUNNING" ]] && return
+      sleep 0.1
+    done
   done
-  die "container did not start"
+  [[ -z "$detail" ]] || printf '%s\n' "$detail" >&2
+  die "container did not start after 3 verified attempts"
 }
 wait_for_guest_network() {
   local name="$1" attempt
diff --git a/test/site.mjs b/test/site.mjs
index 439d57c..250508a 100644
--- a/test/site.mjs
+++ b/test/site.mjs
@@ -210,6 +210,7 @@ test("installer assets are explicit and syntax-valid", () => {
   assert.match(lxcHelper, /wait_for_guest_network[\s\S]*10\\\.0\\\.3\\\.[\s\S]*via 10\\\.0\\\.3\\\.1 dev eth0/, "fresh LXC provisioning waits for its DHCP address and default route");
   assert.match(lxcHelper, /archive\.ubuntu\.com[\s\S]*security\.ubuntu\.com[\s\S]*curl -4[\s\S]*InRelease/, "fresh LXC provisioning proves DNS and outbound IPv4 HTTP before package bootstrap");
   assert.match(lxcHelper, /start "\$name"[\s\S]*wait_for_guest_network "\$name"[\s\S]*apt-get update -o Acquire::ForceIPv4=true -o APT::Update::Error-Mode=any/, "package bootstrap begins only after bounded guest-network readiness");
+  assert.match(lxcHelper, /for start_attempt in \{1\.\.3\}[\s\S]*lxc-start[\s\S]*state "\$name"[\s\S]*continue[\s\S]*container did not start after 3 verified attempts/, "a transient daemonized LXC state-socket failure is retried against the same validated machine");
   assert.match(lxcHelper, /cpuset\.cpus\.effective/, "LXC CPU limits are selected from the service's actually delegated host CPUs");
   assert.match(lxcHelper, /cpu_count/, "LXC inspection counts noncontiguous delegated CPU lists correctly");
   assert.match(lxcNetwork, /1helm-lxc-net-owned/, "the bridge wrapper stops only a bridge it started");