From a71369d71c10f4c16d384aa4e7627da4b8ca72bd Mon Sep 17 00:00:00 2001
From: tegwick <bernd.worsch@gmail.com>
Date: Wed, 8 Jul 2026 00:11:08 +0200
Subject: [PATCH] Fix forgejo-backup large dump copy with chunked kubectl cp

Split in-pod dumps over 50MiB into 16MiB parts with retry to avoid
websocket EOF and connection resets on ~700MiB archives.
---
 tools/cmd/forgejo-backup | 39 ++++++++++++++++++++++++++++++++++++---
 1 file changed, 36 insertions(+), 3 deletions(-)

diff --git a/tools/cmd/forgejo-backup b/tools/cmd/forgejo-backup
index 17b2a65..e981873 100755
--- a/tools/cmd/forgejo-backup
+++ b/tools/cmd/forgejo-backup
@@ -74,10 +74,43 @@ ok "forgejo-db" "${FORGEJO_DB_NAMESPACE}/${FORGEJO_DB_POD}"
 # 1. Application dump (repos, packages, attachments, LFS, avatars)
 ok "forgejo dump" "running in pod…"
 DUMP_REMOTE="/tmp/forgejo-backup-${TS}.zip"
-# Single exec: dump then stream stdout (avoids kubectl cp / multi-exec websocket drops)
 kubectl exec -n "${FORGEJO_NAMESPACE}" "${PROD_POD}" -c gitea -- \
-  sh -c "forgejo dump -f '${DUMP_REMOTE}' && cat '${DUMP_REMOTE}' && rm -f '${DUMP_REMOTE}'" \
-  > "${DUMP_PLAIN}"
+  forgejo dump -f "${DUMP_REMOTE}"
+# Large dumps (~700M) fail via stdout stream or single kubectl cp — use chunks.
+dump_size="$(kubectl exec -n "${FORGEJO_NAMESPACE}" "${PROD_POD}" -c gitea -- \
+  stat -c%s "${DUMP_REMOTE}" 2>/dev/null || echo 0)"
+copy_from_pod() {
+  local remote_path="$1" local_path="$2" attempt
+  for attempt in 1 2 3; do
+    if kubectl cp -n "${FORGEJO_NAMESPACE}" -c gitea \
+      "${FORGEJO_NAMESPACE}/${PROD_POD}:${remote_path}" "${local_path}" 2>/dev/null; then
+      return 0
+    fi
+    sleep $((attempt * 3))
+  done
+  return 1
+}
+
+if [[ "${dump_size}" -lt 52428800 ]]; then
+  copy_from_pod "${DUMP_REMOTE}" "${DUMP_PLAIN}" || exit 1
+else
+  ok "forgejo dump" "chunked copy (${dump_size} bytes)…"
+  chunk_dir="$(mktemp -d)"
+  parts="$(kubectl exec -n "${FORGEJO_NAMESPACE}" "${PROD_POD}" -c gitea -- \
+    sh -c "split -b 16m '${DUMP_REMOTE}' /tmp/forgejo-backup-part- && ls /tmp/forgejo-backup-part-*")"
+  for part in ${parts}; do
+    base="$(basename "${part}")"
+    copy_from_pod "${part}" "${chunk_dir}/${base}" || { bad "forgejo dump" "chunk copy failed: ${base}"; exit 1; }
+  done
+  cat "${chunk_dir}"/forgejo-backup-part-* > "${DUMP_PLAIN}"
+  rm -rf "${chunk_dir}"
+  kubectl exec -n "${FORGEJO_NAMESPACE}" "${PROD_POD}" -c gitea -- \
+    rm -f /tmp/forgejo-backup-part-* "${DUMP_REMOTE}" || true
+fi
+if [[ ! -f "${DUMP_PLAIN}" || ! -s "${DUMP_PLAIN}" ]]; then
+  bad "forgejo dump" "failed to copy dump from pod"
+  exit 1
+fi
 ok "forgejo dump" "$(du -h "${DUMP_PLAIN}" | awk '{print $1}') $(basename "${DUMP_PLAIN}")"
 
 # 2. PostgreSQL logical dump (plain SQL to stdout; CNPG root FS is read-only)