Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →SeaCloud Labs says it moved away from one Elasticsearch index per tenant because the shard and cluster-state work attached to many indexes had become an operational concern. Its replacement, SeaSearch, keeps indexes logically separate but puts authoritative index data in shared S3-compatible storage and routes partitions to compute nodes. That shifts the trade-off: fewer data-movement concerns during node changes, but greater dependence on object storage and a local cache that can make cold queries slower.
Why move away from one index per tenant?
A separate index for each tenant has practical benefits: queries can target the relevant tenant’s data, and separation can help limit the effect of one customer’s traffic spike on another. SeaCloud Labs did not argue that this model is inherently wrong. Its concern was the operational overhead of maintaining a growing number of indexes.
As an Amazon Associate I earn from qualifying purchases.
In the company’s account, each index has at least one shard. Shards maintain Lucene data and background work, while mappings and routing information contribute to cluster state replicated across nodes. SeaCloud Labs says mapping updates became sluggish at thousands of indexes and that tens of thousands could make the master node a paging concern. Those counts describe the company’s experience; they are not universal limits or independently measured thresholds.
The important question is therefore not “How many indexes are too many?” but whether shard work, cluster-state size, mapping churn, and the available operational tooling are becoming problems at your own tenant count.
#1 Best Overall
What SeaSearch changes
SeaCloud Labs says it wanted an alternative without starting a search engine from scratch. It used ZincSearch as a base, citing its Go runtime footprint, Bluge indexing, and Elasticsearch-compatible API, then added shared-storage indexing and routing.
SeaSearch still has indexes, but their data is not authoritative on any one compute node. The cluster’s main pieces have distinct jobs:
- Compute nodes serve reads and writes and access a shared S3-compatible bucket for index data.
- etcd stores index metadata and the partition ownership map.
- A cluster manager monitors node health and assigns partition ownership.
- A proxy or gateway consults the ownership map and forwards client requests to the current owner.
Indexes are hashed into a fixed number of partitions. When nodes change, the ownership map is recalculated; a newly assigned owner fetches data from object storage as needed. SeaCloud Labs describes this as a way to change ownership without migrating or replicating the authoritative index data between compute nodes. The object store becomes a critical part of the architecture, not merely a backup destination.
The project README also describes a single-node mode that uses bbolt for index metadata and the local filesystem for index data. That is a different deployment arrangement from the shared-storage cluster design; a single local disk should not be assumed to provide clustered durability.
How the cache affects query latency
Shared object storage can simplify where authoritative data lives, but fetching segments remotely may be slower than reading them from local NVMe. SeaCloud Labs characterizes object-store round trips as roughly an order of magnitude slower than local NVMe, but does not provide the provider, region, object size, workload, or measurement method behind that comparison. Treat it as the company’s description of the trade-off, not a general latency guarantee.
SeaSearch’s index data is organized into immutable segments: a segment can be read or deleted, but not modified. Compute nodes maintain a rotating local-disk cache and evict older segments when it fills. Because cached segments are immutable, the company says local copies can be used safely, while a node’s index can be larger than its local disk capacity.
Rank #3
SeaCloud Labs says it uses parallel segment warm-up and splits queries across nodes to improve cold-path recovery and spread cache pressure. These techniques do not eliminate cold reads: a query can be slower after a node starts or after relevant segments have been evicted. The company’s own use case is file metadata search, where it says the active working set is a small share of the full dataset. A workload that continually reads data uniformly is less likely to benefit from a cache that keeps only a rotating portion of it.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsHow SeaSearch approaches searching across tenants
Combining results from separate indexes raises a ranking problem. BM25 uses index-local statistics, including term frequency, document count, and average field length. A score calculated within a small index and one calculated within a much larger index should not simply be merged and sorted as though both scores had the same scale.
SeaSearch offers a /api/unified_search endpoint for a consistent query across multiple indexes. The request accepts an array of index/query pairs, and filters can differ by index. SeaCloud Labs presents this as an approach to a broad “search everything I can see” experience; it is not evidence that one ranking method will suit every cross-index relevance problem. Validate the results against the product’s intended behavior and representative queries.
Rank #4
Where Elasticsearch compatibility stops
An Elasticsearch-compatible API does not mean feature parity. SeaCloud Labs says SeaSearch does not offer shard or replica settings, because shared storage changes those concepts. Supported field types are text, keyword, numeric, bool, date, and vector. Mappings can add fields, but existing fields cannot be changed.
The company also lists these unsupported search parameters: indices_boost, knn, min_score, retriever, pit, runtime_mappings, seq_no_primary_term, stats, terminate_after, and version. A migration assessment should include the actual mappings and request patterns in use, not just whether basic queries work.
Recommended Free Tools
SeaCloud Labs explicitly says SeaSearch is not intended to replace observability installations that depend on deep aggregation pipelines and ILM policies. If lifecycle management, aggregation behavior, vector queries, or integrations are essential, confirm each requirement before considering a switch.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What to evaluate before adopting this architecture
SeaSearch’s design is most plausible when tenant indexes are numerous, the active working set is smaller than the full corpus, and shared object storage is operationally dependable. That is an inference from the architecture and the trade-offs SeaCloud Labs describes, not a universal recommendation.
- Tenant isolation and noisy neighbors: Decide whether logical index separation meets your isolation requirements. Consider both request routing and the way shared nodes’ cache demand could affect tenants.
- Index-count operations: Measure shard overhead, cluster-state scale, mapping-update behavior, and the capacity of your operational tooling at your actual tenant count.
- Warm and cold latency: Evaluate startup, cache eviction, cache-hit rates, and realistic access distributions. SeaCloud Labs’ reported experience concerns file metadata search, not every search workload.
- Failure and rebalancing: Compare ownership-map updates and object-store retrieval with the replica recovery and data movement in your current system.
- API and feature fit: Inventory mappings, query operators, aggregations, vector requirements, lifecycle policies, and operational integrations against SeaSearch’s stated limits.
- Object-store operations: Validate the chosen S3-compatible service, region, credentials, failure modes, backup and restore process, and actual cost. SeaCloud Labs identifies S3 and a well-run MinIO cluster as examples, but provides no provider comparison or pricing.
- Ranking semantics: Test cross-index relevance and confirm that one shared query with per-index filters matches what users expect.
SeaCloud Labs says durability depends on the object store. That makes the reliability and recovery characteristics of the chosen shared-storage service part of the search system’s durability model; a single disk should not be treated as adequate shared-storage durability.
What this case study does—and does not—establish
SeaSearch is a first-party architecture account from SeaCloud Labs, and the article is labeled AI-assisted. Its statements about implementation, operational experience, and latency trade-offs should be read as the company’s claims, not as an independent benchmark or proof that the design is faster or cheaper for other workloads. It supplies no universal index-count threshold, provider comparison, or cost analysis.
The case is useful as a design pattern: retain separate indexes while shifting authoritative data to shared storage and assigning partitions to compute nodes. Whether that pattern is preferable depends on the operational pain it removes, the cold-read behavior your users can tolerate, the durability of your object-storage setup, and the features your application actually needs.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
