Elasticsearch Was Never a Database

Why It Matters

Treating Elasticsearch as a primary database introduces data integrity risks and operational complexity, jeopardizing reliability for applications that need a true source of truth.

Key Takeaways

• Elasticsearch lacks ACID transactions beyond single documents.
• Index mappings are immutable, forcing full reindex for schema changes.
• Join operations require denormalization or complex application‑side stitching.
• Operational overhead rises due to shard balancing and JVM tuning.
• ParadeDB offers combined OLTP and search, avoiding sync pipelines.

Pulse Analysis

The appeal of Elasticsearch is obvious: developers can spin up a fast, full‑text engine without the overhead of a traditional database. When an existing relational store struggles with scaling text search, teams often add an Elasticsearch cluster as a secondary index. Over time the sync layer becomes fragile, prompting a tempting shortcut—write directly to the index and discard the original database. This migration appears to simplify architecture, but it masks deeper trade‑offs that only surface once the index is expected to serve as the source of truth.

At the technical level, Elasticsearch diverges sharply from relational databases. It provides no multi‑document transactions, so related writes can become inconsistent if one fails. Mappings are immutable; any schema evolution forces a full reindex, a risky operation when the index holds the only copy of data. Query capabilities, while powerful for search, lack native joins and rely on denormalization or application‑side stitching to emulate relational logic. These gaps force engineers to build custom retry, reconciliation, and migration tooling, inflating code complexity and maintenance burden.

From a business perspective, the operational cost of running Elasticsearch as a primary store is substantial. Shard rebalancing, JVM heap tuning, and rolling upgrades demand specialized expertise, and failures can leave half‑applied transactions that jeopardize data correctness. Organizations that need both reliable OLTP and high‑performance search are better served by solutions purpose‑built for that dual role, such as ParadeDB, which embeds Lucene‑style search within a PostgreSQL‑compatible engine. By keeping Elasticsearch in its intended role—as a secondary index—companies preserve data integrity, reduce engineering overhead, and maintain the scalability that made the technology popular in the first place.