Hardware Requirements
Enabling the Full Text Search feature doesn’t change the hardware requirements for QMS servers themselves, but the Meilisearch host does need its own resources. Meilisearch is a single, lightweight, Rust-based binary with no additional runtime or coordination service to size for. Because it’s lightweight and self-contained, co-hosting Meilisearch on the QMS Data Service machine is a supported and common configuration, though a separate remote host is also supported and may be preferable for larger installations.
Meilisearch’s own resource guidance is qualitative rather than a fixed specification, since actual usage depends heavily on the volume of text being indexed:
-
CPU: additional CPU cores let Meilisearch handle more concurrent indexing and search operations. A modern multi-core processor is recommended; a dedicated remote host benefits from more cores under sustained indexing load (for example, during a full tenant reindex).
-
Memory: search speed is closely tied to the ratio between available RAM and the size of the on-disk index. More RAM generally means faster search, and indexing operations (including a full tenant reindex) are more memory-intensive than steady-state search traffic.
-
Disk: Meilisearch stores its index on disk using a memory-mapped store. Disk usage grows with the number of searchable and filterable fields configured, as well as with the total amount of indexed content. An SSD is recommended over spinning disk for both indexing and search performance.
|
Actual CPU, memory, and disk usage will vary significantly by deployment size - the number of tenants with full text search enabled, and the volume of transcriptions, emails, chats, and UC messages being indexed. There’s no single official minimum specification published by Meilisearch; size the Meilisearch host based on your own installation’s expected volume, and monitor actual usage after enabling the feature. |
Further information on Meilisearch’s resource characteristics and performance tuning options can be found here: