Sharding and Replication

6/4/2026

This post details CarJager's migration from Algolia to Meilisearch, highlighting the benefits of Meilisearch's managed cloud autoscaling for handling traffic peaks and its native approach to sorting without artificial limits, which contrasts with Algolia's replica-based sorting with a 1,000-entity cap. The migration process and the resulting operational and cost benefits are described.

4/20/2026

Introduced sharding and replication as composable features on Meilisearch Cloud, enabling horizontal scaling for datasets that no longer fit on a single machine, high availability through replication for instance restarts, and geographic replication to reduce response times for global users. Also announced upcoming improvements to dashboard visibility for sharded projects.

4/13/2026

This post introduces sharding and replication as new capabilities for Meilisearch Cloud. Sharding addresses the problem of large datasets exceeding single instance capacity by distributing data across multiple nodes, with queries fanning out and results merged. Replication addresses risk by maintaining identical copies of data across multiple nodes for high read availability and failover during updates. The post details how queries work with sharding, the architecture of replication (static leader-based), and use cases for replication including failover during updates and geo-replication. It emphasizes the composability of sharding and replication, allowing for various configurations. Billing is based on the number of nodes. Future roadmap items include optimal geo-replication with sharding, self-serve replication UI, and improved cluster observability.

4/1/2026

This post details how Touchless uses Meilisearch to power instant, scalable vehicle search for large dealer groups. It highlights the switch from Algolia due to Meilisearch's API flexibility, integration with edge infrastructure (Cloudflare edge workers and web sockets), and cost-efficiency. The implementation involves a normalization pipeline, Sanity as a database, and dedicated Meilisearch indexes per dealer group, with new indexes spinning up in milliseconds. The system handles search, filtering, and infinite scroll for inventories of up to 9,000 vehicles across multiple locations, emphasizing the performance benefits of Meilisearch's architecture in a multi-tenant SaaS product.

3/30/2026

This post details the investigation and resolution of memory leaks in Meilisearch, primarily focusing on the interaction between different memory allocators (jemalloc, mimalloc, bumpalo) and the LMDB dependency. It highlights a leak related to `bumpalo::Vec::into_bump_slice` and a more significant issue stemming from the non-cooperation of LMDB's system allocator with Meilisearch's custom global allocator (mimalloc). The solution involved unifying the allocator by using jemalloc and later mimalloc v3 with its override feature, leading to significant memory usage reduction and performance improvements.

3/26/2026

This post details the ongoing development of distributed search capabilities, specifically sharding and replication, to handle larger datasets and improve availability. It also introduces the concept of serverless indexes, where inactive indexes are moved to object storage to reduce costs, and discusses how this will impact free tiers and geo-replication economics. The post also touches on the evolution of hybrid search, AI gateway for managing AI providers, and the development of proprietary AI models for search.

3/18/2026

This post details the implementation of nested read transactions in LMDB, allowing multiple threads to read from an uncommitted write transaction. This was achieved by patching LMDB to modify the behavior of `mdb_txn_begin` and implementing an atomic reference-counted system. The new API was integrated into Hannoy, Meilisearch's HNSW implementation, by replacing the `ImmutableItems` and `ImmutableLinks` data structures with a wrapper for nested read transactions that lazily fetches items and graph links on demand. This optimization resulted in a 3x speedup for Meilisearch's vector store.

3/12/2026

This post announces the release of replicated sharding directly in the Meilisearch engine (v1.37), a major advancement for distributed and resilient search. It highlights this as a stepping stone to one-click replication in the Cloud UI and notes its availability under the Enterprise edition. The post also mentions that the HNSW-backed vector store (Hannoy) is now stabilized and the legacy `vectorStoreSetting` has been removed, with automatic migration during dumpless upgrades. v1.38 further improves embedding indexing performance by eliminating unnecessary full database scans for updates, fixing intermittent connection errors with remote embedders, and resolving an edge case with task and batch deletion.

2/18/2026

This post details how Scenario scaled their asset discovery with Meilisearch, moving from OpenSearch on AWS. It highlights the challenges of performance at scale, cost/operational overhead, and growing user expectations. Scenario chose Meilisearch for its developer-first experience, enterprise readiness, and hybrid search capabilities, enabling them to combine proprietary embeddings with keyword and multi-modal search. The implementation includes continuous indexing, custom embeddings, and vector search integration for an "Exploration" feature, resulting in dramatic reduction in support load, exceptional scale handling (35M+ searches/month, 80% vector), and productivity gains for their engineering team.

1/22/2026

Introduces the ability to add new shards dynamically to Meilisearch, enabling horizontal scaling without a complete architectural overhaul. Mentions that replicated sharding is planned for future development.

12/23/2025

This post introduces Meilisearch's flexible pricing options, including subscription tiers and a new resource-based pricing model, as an alternative to Typesense's resource-based pricing. It details the components of Typesense's pricing (memory, vCPU, add-ons) and highlights the complexity and potential for variable monthly bills. Meilisearch's offering provides a choice between simple, predictable tiers and more granular resource-based control.