JobCannon
全部技能

Elasticsearch

⬢ 層級 2技術
+$25–40k
薪資影響
4 個月
學習時長
困難
難度
—
職業
一覽

Elasticsearch is the industry standard for full-text search, log aggregation, and real-time analytics. Built on Lucene, it powers Wikipedia, GitHub, Stack Overflow. ELK Stack (Elasticsearch, Logstash, Kibana) dominates open-source observability. Learn Query DSL, mapping design, cluster management, and vector search in 3–4 months. Career: SRE/DevOps (centralized logging, $130-200k), Data Engineer (analytics pipelines, $125-190k), Backend (search features, $120-175k). Vector search via ESRE emerging 2026.

什麼是 Elasticsearch

Elasticsearch is the industry-standard distributed search and analytics engine built on Apache Lucene. It powers full-text search (Wikipedia, GitHub, Stack Overflow), log aggregation (ELK Stack = Elasticsearch, Logstash, Kibana), real-time analytics, and emerging vector search (for AI/RAG). Elasticsearch index data with custom analyzers and mappings, query with Query DSL (declarative JSON), and scale horizontally across multiple nodes. In 2026, vector search via ESRE is emerging as new use case, AI applications storing embeddings for semantic search. Elasticsearch is NOT a primary database (no transactions, eventual consistency) but the default search/analytics layer for systems handling >100GB of data. Learning Elasticsearch is valuable for backend engineers building search features, SREs building observability platforms, and data engineers building analytics pipelines. The ELK Stack (Elasticsearch + Logstash + Kibana) is ubiquitous in enterprises; OpenSearch (AWS fork post-license change) is alternative but fundamentally similar.

🔧 工具與生態
ElasticsearchOpenSearchLogstashKibanaBeatsApache SolrAlgoliaTypesenseMeilisearchPineconeWeaviateElastic CloudAWS OpenSearch Service

💰 各地區薪資

地區初級中級高級
USA$110,000$155,000$210,000
UK£65,000£90,000£135,000
EU€70,000€95,000€145,000
CANADACA$120,000CA$165,000CA$225,000

❓ 常見問題

Elasticsearch vs OpenSearch, license change, which should I learn?
Elastic changed to SSPL in 2021; OpenSearch forked as permissive ASSL alternative. Both are functionally similar for core search/logging. Learn Elasticsearch first (larger job market, more tutorials). OpenSearch skills transfer 1:1. Hiring splits: Elastic dominates enterprise (AWS/Google/Meta), OpenSearch growing in AWS-first orgs. For 2026: both relevant; OpenSearch hiring accelerating due to cost savings.
Elasticsearch 8 vector search (ESRE) vs Pinecone/Weaviate, when do I use each?
Elasticsearch ESRE: proprietary vector search, ~$5/GB/month, integrates with existing log/search workflows. Pinecone: serverless, $0.25–1/GB/month, simpler API, no manage. Weaviate: self-hosted or cloud, GraphQL interface, larger community. ESRE for unified stack (search + vectors + logs). Pinecone for pure vector-only RAG. Weaviate for on-prem deployments.
Log aggregation, Elasticsearch vs SaaS (Datadog, New Relic, Grafana)?
Self-hosted Elasticsearch: cheapest at scale (>500GB/day), full control, operational toil (upgrades, tuning). SaaS (Datadog): fixed costs, pre-built dashboards, no ops burden. Crossover: ~100GB/day for Elasticsearch to win on TCO. Startups <50GB/day: SaaS. Enterprise >500GB/day: Elasticsearch or hybrid.
Sharding strategy, how many shards per index?
Rule: each shard = 30–50GB max (disk + memory pressure). 100GB index = 2–3 shards. Too many shards = overhead (per-shard state, merge costs). Too few = slow queries on large indices. Primary + 1 replica = standard HA. Auto-scale with ILM: shard at ingest-time, merge at rollover.
Algolia vs Elasticsearch for product search, which for e-commerce?
Algolia: best UX (instant typo-tolerant faceting), $0–3k/month, managed. Elasticsearch: DIY relevance tuning, cheaper at scale, 15+ fields. Algolia for <10k products, fast time-to-market. Elasticsearch for >100k SKUs, custom ranking, lower budget. Most e-commerce: hybrid (Algolia for storefront, Elasticsearch for analytics/logs).
When should I NOT use Elasticsearch?
Never as primary DB (not ACID, eventual consistency). Not for <1GB datasets (overkill, operational cost). Not for structured transactional data (use PostgreSQL). Not real-time dashboards <100ms (use ClickHouse/TimescaleDB). Ideal: logs, time-series, full-text search, analytics >1TB.
Cluster health, what metrics matter most?
Three pillars: (1) Shard allocation, all shards assigned, no unassigned. (2) GC pause time, avg < 500ms, p99 < 2s (JVM heap tuning). (3) Indexing/query latency, p99 < 1s for reads, p95 < 500ms for writes. Use _cat/health, _nodes/stats, Kibana cluster monitor. Alerts on red status or high GC pause.

不確定這項技能適不適合你？

做一下職業匹配，我們會推薦合適的方向。

找到最適合我的技能 →

找到理想的職業方向

在 2,521 個職業中進行技能匹配。免費。

免費做職業匹配 →