Elasticsearch

elasticsearch

Use this skill for Elasticsearch（ES）相关设计、性能与运维（DBA/中间件）任务。

Defaults / assumptions to confirm

• ES version and deployment (self-hosted / managed)
• Cluster topology (nodes, roles, storage type)
• Data volume and retention requirements
• Query patterns (search vs analytics) and latency SLO

Workflow

1. Understand use-cases and query patterns

• Primary user journeys: keyword search, filtering, aggregations, sorting.
• Write patterns: append-only logs vs frequent updates.
• Required consistency and freshness (near real-time delay tolerance).

2. Index & mapping design

• Define index naming convention and templates.
• Choose correct field types (keyword vs text, date, long, scaled_float).
• Analyze/analyzer strategy for language (e.g., Chinese tokenizer) if needed.
• Avoid mapping explosion; control dynamic mappings.
• Plan _source and stored fields usage; consider doc values.

3. Shards and replicas

• Pick shard count with future growth and reindex cost in mind.
• Avoid too many small shards; target shard size range (e.g., 10–50GB) depending on workload.
• Set replicas for availability and read scaling.

4. Query tuning

• Use profile and slow logs to find bottlenecks.
• Reduce heavy aggregations; precompute when possible.
• Use filters with keyword fields; cache-friendly queries.
• Pagination: prefer search_after for deep pages; avoid large from+size.

5. Lifecycle management

• Use ILM (hot-warm-cold-delete) for time-series data.
• Rollover policies by size/time; manage retention.

6. Cluster operations & stability

• Monitor heap, GC, CPU, disk watermarks, thread pool rejections.
• Detect hot keys/indices; rebalance shards carefully.
• Snapshot/restore; restore drills; retention policy.

7. Safe changes

• Mapping changes often require reindex; plan alias-based migrations.
• Use index aliases for zero-downtime cutover.
• Upgrade runbook: compatibility, rolling upgrade, backout plan.

Outputs

• Mapping/index template proposal + rationale.
• Shard/replica sizing plan + expected capacity.
• Performance diagnosis report (evidence → root cause → fixes).
• Migration plan (reindex + alias cutover + verification).