feat(ops): expand sentinel checks for lan services and embeddings
This commit is contained in:
@@ -10,6 +10,12 @@ BACKUP_MAX_AGE_HOURS="${BACKUP_MAX_AGE_HOURS:-8}"
|
||||
SEARX_URL="${SEARX_URL:-http://192.168.153.113:18803}"
|
||||
WHISPER_URL="${WHISPER_URL:-http://192.168.153.113:18801}"
|
||||
MCP_URL="${MCP_URL:-http://192.168.153.113:18802/mcp}"
|
||||
OLLAMA_URL="${OLLAMA_URL:-http://127.0.0.1:11434}"
|
||||
OLLAMA_EMBED_MODEL="${OLLAMA_EMBED_MODEL:-nomic-embed-text:latest}"
|
||||
OLLAMA_REQUIRED="${OLLAMA_REQUIRED:-false}"
|
||||
MINIO_URL="${MINIO_URL:-http://192.168.153.253:9000}"
|
||||
DOCKER_EXPECTED_CONTAINERS="${DOCKER_EXPECTED_CONTAINERS:-searxng whisper-server brave-search}"
|
||||
DOCKER_REQUIRE_EXPECTED="${DOCKER_REQUIRE_EXPECTED:-false}"
|
||||
WARN_DISK_PCT="${WARN_DISK_PCT:-85}"
|
||||
WARN_MEM_PCT="${WARN_MEM_PCT:-85}"
|
||||
|
||||
@@ -122,12 +128,63 @@ else
|
||||
add_next "Create backup log or fix backup script path"
|
||||
fi
|
||||
|
||||
# 3) Key LAN service probes
|
||||
# 3) Docker service health (containers + health state)
|
||||
if command -v docker >/dev/null 2>&1; then
|
||||
if docker ps --format '{{.Names}}\t{{.Status}}' >"$ARTIFACT_DIR/docker-ps.txt" 2>"$ARTIFACT_DIR/docker-ps.err"; then
|
||||
for svc in $DOCKER_EXPECTED_CONTAINERS; do
|
||||
if grep -qE "^${svc}[[:space:]]" "$ARTIFACT_DIR/docker-ps.txt"; then
|
||||
line="$(grep -E "^${svc}[[:space:]]" "$ARTIFACT_DIR/docker-ps.txt" | head -n1)"
|
||||
if echo "$line" | grep -qiE 'unhealthy|dead|exited|restarting'; then
|
||||
add_now "P1 Docker container ${svc} unhealthy (${line#*$'\t'})"
|
||||
mark_p1
|
||||
add_next "Check logs: docker logs --tail=200 ${svc}"
|
||||
else
|
||||
add_watch "P4 Docker container ${svc} running"
|
||||
fi
|
||||
else
|
||||
if [[ "${DOCKER_REQUIRE_EXPECTED}" == "true" ]]; then
|
||||
add_now "P1 Docker container missing: ${svc}"
|
||||
mark_p1
|
||||
add_next "Start or restore container: ${svc}"
|
||||
else
|
||||
add_watch "P3 Docker container not found locally: ${svc} (may run on another host)"
|
||||
fi
|
||||
fi
|
||||
done
|
||||
else
|
||||
add_watch "P3 docker ps failed"
|
||||
add_next "Check Docker daemon permissions/state"
|
||||
fi
|
||||
else
|
||||
add_watch "P3 docker CLI not available"
|
||||
fi
|
||||
|
||||
# 4) Key LAN + local service probes
|
||||
http_probe "searxng" "$SEARX_URL" '^200$'
|
||||
http_probe "whisper" "$WHISPER_URL" '^200$'
|
||||
http_probe "brave-mcp" "$MCP_URL" '^(200|406)$'
|
||||
http_probe "minio-live" "${MINIO_URL%/}/minio/health/live" '^200$'
|
||||
|
||||
# 4) Host pressure: disk + memory
|
||||
# 5) Ollama embeddings availability + target model
|
||||
if curl -sS -m 6 "${OLLAMA_URL%/}/api/tags" >"$ARTIFACT_DIR/ollama-tags.json" 2>"$ARTIFACT_DIR/ollama-tags.err"; then
|
||||
if jq -e --arg model "$OLLAMA_EMBED_MODEL" '.models[]? | select(.name == $model)' "$ARTIFACT_DIR/ollama-tags.json" >/dev/null 2>&1; then
|
||||
add_watch "P4 Ollama up; embedding model present (${OLLAMA_EMBED_MODEL})"
|
||||
else
|
||||
add_soon "P2 Ollama up but embedding model missing (${OLLAMA_EMBED_MODEL})"
|
||||
mark_p2
|
||||
add_next "Pull model: ollama pull ${OLLAMA_EMBED_MODEL}"
|
||||
fi
|
||||
else
|
||||
if [[ "${OLLAMA_REQUIRED}" == "true" ]]; then
|
||||
add_now "P1 Ollama unreachable (${OLLAMA_URL})"
|
||||
mark_p1
|
||||
add_next "Check Ollama service and port (default 11434)"
|
||||
else
|
||||
add_watch "P3 Ollama unreachable at configured URL (${OLLAMA_URL}); set OLLAMA_URL if remote"
|
||||
fi
|
||||
fi
|
||||
|
||||
# 6) Host pressure: disk + memory
|
||||
root_disk_pct="$(df -P / | awk 'NR==2 {gsub(/%/,"",$5); print $5}' 2>/dev/null || echo 0)"
|
||||
if [[ "$root_disk_pct" =~ ^[0-9]+$ ]]; then
|
||||
if (( root_disk_pct >= 95 )); then
|
||||
|
||||
Reference in New Issue
Block a user