From 2cb06debc55e6fa846c689a246e8799331258b21 Mon Sep 17 00:00:00 2001
From: mr-forust <vzlomdsisma@gmail.com>
Date: Mon, 28 Sep 2026 22:52:54 +0200
Subject: [PATCH] fix(deploy): retry registry lookups with a timeout

A single blink of the registry failed render_pinned for the whole file and redded the apply stage. registry_digest now retries 3 times under a 25s timeout with a warning per attempt; empty still means unresolvable and callers report it by name as before.
---
 .gitea/workflows/deploy-lib.sh | 29 ++++++++++++++++++++++-------
 .gitea/workflows/deploy.yaml   | 12 ++++++------
 2 files changed, 28 insertions(+), 13 deletions(-)

diff --git a/.gitea/workflows/deploy-lib.sh b/.gitea/workflows/deploy-lib.sh
index bbb640a..7aabbe5 100644
--- a/.gitea/workflows/deploy-lib.sh
+++ b/.gitea/workflows/deploy-lib.sh
@@ -233,13 +233,28 @@ registry_digest() {
   # pipefail reports the rightmost non-zero stage, so a ref the registry does not
   # have would abort the caller at the assignment instead of yielding an empty
   # string. The callers check for empty themselves and report it by name.
-  docker manifest inspect "$1" 2>/dev/null \
-    | jq -r --arg arch "$arch" '
-        .manifests[]?
-        | select(.platform.os == "linux" and .platform.architecture == $arch)
-        | .digest
-      ' 2>/dev/null \
-    | head -1 || true
+  #
+  # Retried with a hard timeout because the registry has a known hang mode (and
+  # a known blink mode: a single failed lookup aborts the whole apply file in
+  # render_pinned). A short sleep between attempts lets a restarting registry
+  # come back instead of failing the deploy on one bad second.
+  local attempt=0 digest=""
+  while [ "$attempt" -lt 3 ]; do
+    digest="$(timeout 25s docker manifest inspect "$1" 2>/dev/null \
+      | jq -r --arg arch "$arch" '
+          .manifests[]?
+          | select(.platform.os == "linux" and .platform.architecture == $arch)
+          | .digest
+        ' 2>/dev/null \
+      | head -1 || true)"
+    [ -n "$digest" ] && break
+    attempt=$((attempt + 1))
+    if [ "$attempt" -lt 3 ]; then
+      echo "WARNING: registry lookup of $1 failed (attempt $attempt/3), retrying in 5s" >&2
+      sleep 5
+    fi
+  done
+  printf '%s' "$digest"
 }
 
 # The commit this deploy is for: what CI validated, or - on a manual dispatch,
diff --git a/.gitea/workflows/deploy.yaml b/.gitea/workflows/deploy.yaml
index ba2e37f..5120d32 100644
--- a/.gitea/workflows/deploy.yaml
+++ b/.gitea/workflows/deploy.yaml
@@ -95,12 +95,12 @@ jobs:
     # The 45 minutes this was last raised to 45 were still not enough, and the
     # job logs for those runs no longer exist, so what actually consumed the
     # budget is not known - the two measurable candidates above account for
-    # ~15 of it. The one unbounded thing left in this stage is
-    # `docker manifest inspect` at deploy-lib.sh:236, which has no timeout
-    # against a registry with a known hang mode. Bound it, and make the stage
-    # announce what it is working on, before spending any of that on a larger
-    # ceiling: a stage that is killed with a diagnosable last line is a bug
-    # report, one that vanishes is not.
+    # ~15 of it. The unbounded `docker manifest inspect` against the registry's
+    # known hang mode is now bounded inside registry_digest (25s timeout, 3
+    # attempts): a dead registry fails each owned image after ~85s instead of
+    # hanging the stage, and a blinking one is retried instead of failing the
+    # whole apply file. Still open: make the stage announce which manifest it
+    # is working on, so a killed run leaves a diagnosable last line.
     timeout-minutes: 45
     steps:
       - name: Checkout repository