minimax: ship boot log on ERR too

worker/minimax_start.sh

@@ -13,6 +13,22 @@ set -eo pipefail
 LOG=/tmp/minimax-boot.log
 exec > >(tee -a "$LOG") 2>&1
 
+ship_log() {
+  local py="${PY:-$(command -v python3)}"
+  "$py" -m pip show huggingface_hub >/dev/null 2>&1 || "$py" -m pip install -q huggingface_hub || true
+  "$py" - <<'PYEOF' || true
+import os
+from huggingface_hub import HfApi
+api = HfApi(token=os.environ.get("HF_TOKEN"))
+pod = os.environ.get("RUNPOD_POD_ID", "unknown")
+api.upload_file(path_or_fileobj="/tmp/minimax-boot.log",
+                path_in_repo=f"logs/boot-{pod}.log",
+                repo_id="nikdevs/qwen", commit_message=f"boot log {pod}")
+print("[minimax] boot log shipped")
+PYEOF
+}
+trap 'echo "[minimax] BOOT FAILED (line $LINENO)"; ship_log' ERR
+
 VARIANT="${MODEL_VARIANT:-fl2va}"
 REPO="nikdevs/minimax-h3-${VARIANT}"
 TARGET="${COMFY_VERSION:-v0.31.0}"
@@ -86,22 +102,9 @@ open("/comfyui/custom_nodes/lora_from_url.py", "wb").write(urllib.request.urlope
 print("[minimax] lora_from_url node installed")
 PYEOF
 
-# Background log shipper: uploads the boot log to HF a few times so the tail
-# of a crashed ComfyUI startup is visible from outside.
+# Background log shipper: uploads the boot log a few times so the tail of a
+# crashed ComfyUI startup is visible from outside.
 "$PY" -m pip show huggingface_hub >/dev/null 2>&1 || "$PY" -m pip install -q huggingface_hub
-(
-  for i in 1 2 3 4; do
-    sleep 300
-    "$PY" - <<'PYEOF' || true
-import os
-from huggingface_hub import HfApi
-api = HfApi(token=os.environ.get("HF_TOKEN"))
-pod = os.environ.get("RUNPOD_POD_ID", "unknown")
-api.upload_file(path_or_fileobj="/tmp/minimax-boot.log",
-                path_in_repo=f"logs/boot-{pod}.log",
-                repo_id="nikdevs/qwen", commit_message=f"boot log {pod}")
-PYEOF
-  done
-) &
+( for i in 1 2 3 4; do sleep 300; ship_log; done ) &
 
 exec /start.sh