krzysztof/landing_page
1
0
Fork 0
mirror of https://github.com/qdrant/landing_page.git synced 2026-09-25 14:08:30 +02:00
Code Issues Packages Projects Releases Wiki Activity

suggestions (#1523)

Browse Source
This commit is contained in:
Andrey Vasnetsov
2025-03-24 19:51:41 +01:00
committed by GitHub
parent 20bd5fb206
commit 84edb02f6c
1 changed files with 6 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
qdrant-landing/content/articles/indexing-optimization.md
+6 -4
View File
@@ -33,15 +33,16 @@ Sparse vectors use an **inverted index**. This index is updated at the **time of
When performing high-volume vector ingestion, you have **two primary options** for handling indexing overhead. You should choose one depending on your specific workload and memory constraints:
- **The `"m"` parameter**
- **Disable HNSW indexing**
To reduce memory and CPU pressure during bulk ingestion, you can **disable HNSW indexing entirely** by setting `"m": 0`.
For dense vectors, the `m` parameter defines how many edges each node in the HNSW graph can have.
This way, no dense vector index will be built, preventing unnecessary CPU usage during ingestion.
**Figure 1:** A description of three key HNSW parameters.
<img src="/articles_data/indexing-optimization/hnsw-parameters.png" width="600">
To reduce memory and CPU pressure during bulk ingestion, you can **disable HNSW indexing entirely** by setting `"m": 0`. This way, no dense vector index will be built, no matter the `indexing_threshold`, preventing unnecessary memory spikes.
```json
PATCH /collections/your_collection
@@ -52,9 +53,10 @@ PATCH /collections/your_collection
}
```
**After ingestion is complete**, you can **re-enable HNSW** by setting `m` back to a production value (commonly 16 or 32). Remember that the vectors won't be searchable via HNSW until `m` is re-enabled.
**After ingestion is complete**, you can **re-enable HNSW** by setting `m` back to a production value (commonly 16 or 32).
Remember that search won't use HNSW until the index is built, so search performance may be slower during this period.
- **The `indexing_threshold` parameter**
- **Disabling optimizations completely**
The `indexing_threshold` tells Qdrant how many unindexed dense vectors can accumulate in a segment before building the HNSW graph. Setting `"indexing_threshold"=0` defers indexing entirely, keeping **ingestion speed at maximum**. However, this means uploaded vectors are not moved to disk while uploading, which can lead to **high RAM usage**.