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
Running MeiliSearch?
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.
What it is
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.
The honest version
No technology is all upside. This is where MeiliSearch earns its place and where it tends to cause trouble.
Sound familiar?
The MeiliSearch problems that bring people to us, and what you have once each one is fixed.
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
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
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
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
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
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
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
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
Pass us a broken task queue, or bring the whole search stack across for a review.
How it works
Tell us what's going wrong. We look at it, then tell you plainly what's wrong and what it will cost.
We fix the cause, not the symptom. If something is down, getting it back up comes first.
Backups, monitoring, updates and access, so the same problem can't quietly come back.
You get a plain-language write-up. Or we stay on for hosting, upkeep and new work.
Often found together
We work on the whole stack, so the problem can be wherever it actually is.
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.
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.
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.
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.
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.
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
Send over the task log or a note on the index, and a search engineer replies with a plan and a price.
Your message is with a real person, and we'll get back to you soon.
Rather talk now? Call us: