Running MeiliSearch?

MeiliSearch tasks failing or ranking wrong? We clear the backlog.

MeiliSearch is approachable right up until a task queue gets wedged, a ranking rule change moves the wrong results to the top, or an index starts eating memory on a node that used to be comfortable. We pull apart the schema, the ranking rules and the task log, and give you a cluster that returns the right documents in the right order.

  • Since 2015Fixing technology
  • FreeEstimates
  • US teamReal engineers

meilisearch.log

still failing

What it is

MeiliSearch at a glance

MeiliSearch is an open source full text search engine written in Rust and distributed as a single binary. Clement Renault and Thomas Payet started the project, and the first stable release landed in 2020, with version 1.0 following in 2023 and ongoing work on cloud and typo tolerance features.

The design goal is instant search with sensible defaults, so most teams can bring up a usable index in an afternoon. The depth shows up later: ranking rules, filterable and sortable attributes, task queues and the asynchronous settings model all have real failure modes once an index grows past a few hundred thousand documents or once a team starts tuning for relevance rather than taking the defaults.

Type
Open source full text search engine
First released
2020 initial stable release, 1.0 in 2023
License
MIT, open source core, managed cloud available
Maintained by
Meili SAS and open source contributors
Typically used for
Product search, docs search, in app search, autocomplete with typo tolerance

The honest version

MeiliSearch pros and cons

No technology is all upside. This is where MeiliSearch earns its place and where it tends to cause trouble.

Where MeiliSearch is strong

  • Instant search out of the boxTypo tolerance, prefix search and highlighting work with defaults, which means a working search box ships in hours rather than weeks.
  • A single binary to runNo JVM, no coordinator cluster, no cluster state to manage for small deployments. One process, one data directory, one port.
  • The query API is friendlyThe JSON shape for queries, filters and facets is straightforward enough that frontend engineers can write against it without a backend intermediary.
  • Good relevance for the common caseThe default ranking rules order by typo, words, proximity, attribute, sort, exactness, which produces useful results on product and documentation search without hand tuning.
  • Faceted filtering built infilterable attributes and distinct attributes give you faceted search, de duplication and category drilldowns through the same API.
  • Lightweight compared with ElasticsearchMemory and disk footprint on a comparable corpus are a fraction of an Elasticsearch cluster, which keeps small to mid size deployments cheap.

Where MeiliSearch bites

  • Settings changes trigger full reindexChanging searchableAttributes, filterable attributes or ranking rules enqueues a task that reindexes the whole collection. On a large index this can take hours and locks the write path for its duration.
  • Task queue failures are not loudA failed task sits in the log but the index silently stays on the old settings. Teams often find out when search behavior regresses rather than when the task errors.
  • Scaling horizontally is limitedThere is no native sharding in the open source product, so very large corpora push you toward partitioning at the application level or paying for the cloud offering.
  • Ranking rule ordering is subtleThe order of rules matters, and placing sort before relevance rules produces technically valid but unhelpful orderings. The docs cover it, the error messages less so.
  • Document size and payload caps biteDefault 100 MB payload limit on batch imports, and per document size matters. Large catalog imports need batching and tuning.
  • Cloud and self hosted feature driftMeilisearch Cloud lights up features and limits that the open source binary does not expose immediately. Teams on self hosted find themselves behind on observability and analytics.

Sound familiar?

Common MeiliSearch problems we fix

The MeiliSearch problems that bring people to us, and what you have once each one is fixed.

Task failed: Index `products` not found

Tasks are failing against a missing index

A deploy where the index was recreated under a different name, or a settings update enqueued before the index existed, leaving a line of failed tasks in the dashboard.

Tasks all completing cleanly

ranking_rules: sort must come after relevance

Ranking rules are in the wrong order

A configuration where sort was moved to the top of the ranking rules, so results ordered by price or date push obviously better matches down the list.

Results ordered the way users expect

Maximum database size reached

The on disk store has hit its cap

The default max_indexing_memory and max_db_size settings sized for development, not production, so writes start failing when the catalog grows.

Store sized for real data

invalid_search_filter: not filterable

A filter is being rejected

An attribute that was never added to filterable attributes, so the frontend filter works in dev where the config was loose and fails in staging.

Facets and filters work the same everywhere

Search 1.3s, consider trimming

Search latency has drifted upward

Too many searchable attributes, highlighting on huge text fields, or a ranking rule stack that does more work than the UI actually needs.

Searches comfortably under 100ms

primary_key_inference failure

MeiliSearch cannot work out the primary key

A batch where some documents have id and others have uid, so inference finds more than one candidate and rejects the whole batch.

Primary key explicit and consistent

missing_master_key: 403

The master key is missing or wrong

An environment where MEILI_MASTER_KEY was never set in production, so default protections kick in and existing clients start failing on 403.

Keys configured and rotated safely

payload_too_large: 100MB

Bulk imports are being rejected

A migration script that posts the entire catalog in one HTTP request and hits the payload cap, so no documents make it in.

Imports chunked and resumable

What we do

MeiliSearch help, from index design to upkeep

Pass us a broken task queue, or bring the whole search stack across for a review.

Rescue
A MeiliSearch that is failing tasks, returning nonsense orderings or refusing writes. We get the queue clear before we redesign anything.
Index design
Primary keys, searchable attributes, synonyms, stop words and distinct attributes matched to what the UI actually needs.
Ranking relevance
Ranking rule order, sort attributes and custom rules tuned against real queries, with a test set that stays in version control.
Scale and sharding
Memory and disk sizing, batch ingest pipelines, and a partitioning plan when a single node is no longer enough.
Migrations
Moving onto MeiliSearch from Algolia, Elasticsearch or a database LIKE query, without losing tenant isolation or analytics.
Managed Tech
Backups, task queue monitoring, version upgrades and an on call engineer when a settings change locks the write path at the wrong moment.

How it works

From first message to fixed

  1. Look Free estimate

    Tell us what's going wrong. We look at it, then tell you plainly what's wrong and what it will cost.

  2. Fix

    We fix the cause, not the symptom. If something is down, getting it back up comes first.

  3. Harden

    Backups, monitoring, updates and access, so the same problem can't quietly come back.

  4. Hand back or stay on

    You get a plain-language write-up. Or we stay on for hosting, upkeep and new work.

Often found together

Running MeiliSearch alongside something else?

We work on the whole stack, so the problem can be wherever it actually is.

Questions

MeiliSearch questions

Something else? Call+1 737 727 7327

Why are my MeiliSearch tasks stuck in failed?

The task log is the first place to look. Common causes are a settings update enqueued against an index that no longer exists, a batch where the primary key cannot be inferred, or a write that exceeded the configured max_db_size. The index itself keeps serving queries on the previous settings, so the failure is usually quiet until someone notices the new sort or filter is not actually live.

Why is MeiliSearch returning the wrong order?

Almost always a ranking rule ordering issue. If sort sits above relevance rules, results get ordered by price or date before relevance is considered, which pushes obviously better matches down the list. Custom rules and the exactness rule placement also change the top of the result set in ways that are easy to miss in a spot check and show up in production analytics.

Can MeiliSearch replace Algolia or Elasticsearch?

For most product search, docs search and in app search, yes. The ergonomics are similar to Algolia and the cost is far lower; the self hosted route also removes the per record pricing surprise. Elasticsearch is still a better pick when you need deep analytics, log aggregation or sharding that MeiliSearch does not natively offer.

How do we handle large reindexes without downtime?

You do not change settings on the live index. The pattern is to build a new index under a different name, populate it in batches, verify, then swap the alias or the config to point search traffic at the new one. That way the user facing surface keeps running on the old index until the new one is proven.

Why is our MeiliSearch node running out of memory?

Usually a combination of too many searchable attributes, highlighting on large text fields, and max_indexing_memory set too high for the host. The engine holds a working set in RAM during indexing, and the default sizing assumes a developer laptop. Production nodes need the memory settings tuned deliberately.

How much does MeiliSearch help cost?

The quote is on the house. Share a sample query or describe what the index is doing, and we come back with a price before we start indexing anything.

Free estimate

MeiliSearch letting you down?

Send over the task log or a note on the index, and a search engineer replies with a plan and a price.

  • No charge to find out what it will take
  • If it's urgent, say so

Replies come from a person. By sending, you authorize us to email you back. No marketing.