Upload boot-with-log.sh with huggingface_hub

boot-with-log.sh

@@ -0,0 +1,25 @@
+#!/bin/bash
+# Diagnostic wrapper: runs the normal Krea 2 bootstrap and uploads its log,
+# plus disk and process snapshots, to a private Hugging Face dataset every 60 s.
+L=/tmp/boot.log
+: > "$L"
+(/bootstrap-models.sh >>"$L" 2>&1; echo "BOOTSTRAP EXITED $?" >>"$L") &
+upload() {
+  {
+    echo "=== $(date -u +%FT%TZ) uptime $(cut -d' ' -f1 /proc/uptime)s"
+    df -h /comfyui /tmp 2>&1 | tail -3
+    ls -la /comfyui/models/unet /comfyui/models/clip /comfyui/models/vae 2>&1 | grep -v '^total'
+    ps -eo pid,etime,pcpu,args --sort=-pcpu 2>/dev/null | head -8
+    nvidia-smi --query-gpu=name,memory.used,memory.total,utilization.gpu --format=csv,noheader 2>&1
+    echo "=== last 120 log lines"
+    tail -120 "$L"
+  } > /tmp/snap.txt 2>&1
+  B64=$(base64 -w0 /tmp/snap.txt)
+  printf '{"key":"header","value":{"summary":"boot log %s"}}\n{"key":"file","value":{"content":"%s","path":"%s/boot.txt","encoding":"base64"}}\n' \
+    "$(date -u +%T)" "$B64" "${RUNPOD_POD_ID:-worker}" > /tmp/commit.ndjson
+  curl -s -m 60 -X POST "https://huggingface.co/api/datasets/${LOG_REPO}/commit/main" \
+    -H "Authorization: Bearer ${LOG_TOKEN}" -H "Content-Type: application/x-ndjson" \
+    --data-binary @/tmp/commit.ndjson > /tmp/commit.out 2>&1
+}
+sleep 20; upload
+while true; do sleep 60; upload; done