Clément Renault
Clément Renault
When many labels and values are in a graph, and the graph is too small to display all of them, the shown labels are not interesting enough. The labels shown...
[While debugging a very big client](https://github.com/meilisearch/meilisearch-support/issues/110), we first thought the engine would respect the maximum memory usage defined by the `MEILI_MAX_INDEXING_MEMORY` env variable. However, after some time, the engine started...
When modifying a small number of documents, they appear in the wrong order. Related to #4374, #4339, and #4390. **To Reproduce** It happens when a few documents are modified in...
This PR is related to #4391 and must be merged after it. It introduces a new tool to filter the profile logs that are associated with high memory usage only....
This PR disables the facet search indexation, which can take much indexing time. This is only a temporary fix for a customer and will be fixed by #4354. I am...
When [investigating a bug with a big customer](https://github.com/meilisearch/meilisearch-support/issues/95), we found out that it can take up to forty minutes to load 150 updates of about 121k documents. Most of the...
When [debugging a very big client](https://github.com/meilisearch/meilisearch-support/issues/95), we noticed that executing the differential indexing on 148M documents with nine filterable and three sortable attributes. When disabling the incremental indexing and only...
Vectors can weigh a lot when they are stored as text. A typical OpenAI vector has 1536 dimensions, which is 6KiB. When stored as JSON, it is text, and the...
Opening the unnamed database is fallible as it uses the same API as the named ones. As the unnamed database always exists, there is no reason to make creating the...
The issue with the current heed library is that it is not ensuring either of those two points. Here is [the documentation of `mdb_dbi_open` LMDB function](https://github.com/LMDB/lmdb/blob/mdb.master3/libraries/liblmdb/lmdb.h#L1181-L1252): 1. The database handle...