270 qdrant

qdrant-monitoring-debugging Skill

使用指标和可观测性工具诊断 Qdrant 生产环境问题。当有人报告“optimizer 卡住”“索引太慢”“内存过高”“OOM 崩溃”“查询很慢”“延迟飙升”或“搜索以前快现在慢”时使用。在配置没有明显变更但性能下降时也可使用。

安装方式:把技能目录放入 ~/.claude/skills/(Claude Code)或在 claude.ai 设置中启用;也可复制右侧安装命令一键添加。

查看源码

技能指令原文(SKILL.md)

How to Debug Qdrant with Metrics

First check optimizer status. Most production issues trace back to active optimizations competing for resources. If optimizer is clean, check memory, then request metrics.

Optimizer Stuck or Too Slow

Use when: optimizer running for hours, not finishing, or showing errors.

  • Use the /collections/{collection_name}/optimizations endpoint (v1.17+) to see queued and running optimizations and their progress; add ?with=queued,completed,idle_segments for the full lists Optimization monitoring
  • Web UI has an Optimizations tab with timeline view and per-task duration metrics Web UI
  • If optimizer_status shows an error in collection info, check logs for disk full or corrupted segments
  • Large merges and HNSW rebuilds legitimately take hours on big datasets. Check progress before assuming it's stuck.

Memory Seems Too High

Use when: memory exceeds expectations, node crashes with OOM, or memory keeps growing.

  • Watch the process memory metrics on /metrics: memory_resident_bytes, memory_allocated_bytes, and process_major_page_faults_total Node metrics
  • Qdrant uses two types of RAM: resident memory (data structures, quantized vectors) and OS page cache (cached disk reads). Page cache filling available RAM is normal. Memory article
  • If resident memory (RSSAnon) exceeds 80% of total RAM, investigate
  • Check /telemetry for per-collection point counts and vector configurations. For a per-component breakdown of RAM, page cache, and disk use, call GET /collections/{collection_name}/memory or open the Web UI's Memory tab (v1.18+) Memory usage
  • Estimate expected memory per component (vectors, HNSW index, ID tracker, payload indexes) for the structures kept in RAM, plus about 20% headroom, and compare it with actual usage Capacity planning
  • Common causes of unexpected growth: quantized vectors pinned in RAM (memory: pinned on Qdrant 1.19 or newer, always_ram: true on 1.18 or older), too many payload indexes, large max_segment_size during optimization

Queries Are Slow

Use when: queries slower than expected and you need to identify the cause.

  • Slow request log (v1.16+, always on): surfaces individual slow queries with timing breakdown. It's per node and resets on restart, so query every node in a cluster by using GET /profiler/slow_requests Slow request log
  • Use the rest_responses_duration_seconds histogram (v1.8+; grpc_responses_duration_seconds for gRPC) for latency percentiles, rather than the avg and max gauges. Add ?per_collection=true to /metrics to break them down by collection Per-collection API metrics
  • Check optimizer status first. Active optimizations compete for CPU and I/O, degrading search latency.
  • Check segment count via collection info. Too many unmerged segments after bulk upload causes slower search.
  • Compare filtered vs unfiltered query times. Large gap means missing payload index. Payload index

What NOT to Do

  • Ignore optimizer status when debugging slow queries (most common root cause)
  • Assume memory leak when page cache fills RAM (normal OS behavior)
  • Make config changes while optimizer is running (causes cascading re-optimizations)
  • Blame Qdrant before checking if bulk upload just finished (unmerged segments)