mirror of
https://github.com/immich-app/immich.git
synced 2025-12-30 01:11:52 +03:00
feat: use pgvecto.rs (#3605)
This commit is contained in:
Jason Rasmussen
@@ -45,7 +45,6 @@ The Immich backend is divided into several services, which are run as individual
|
||||
1. `immich-machine-learning` - Execute machine learning models
|
||||
1. `postgres` - Persistent data storage
|
||||
1. `redis`- Queue management for `immich-microservices`
|
||||
1. `typesense`- Specialized database for search, specifically with vector comparison features
|
||||
|
||||
### Immich Server
|
||||
|
||||
@@ -75,7 +74,6 @@ The Immich Microservices image uses the same `Dockerfile` as the Immich Server,
|
||||
- Object Tagging
|
||||
- Facial Recognition
|
||||
- Storage Template Migration
|
||||
- Search (Typesense synchronization)
|
||||
- Sidecar (see [XMP Sidecars](/docs/features/xmp-sidecars.md))
|
||||
- Background jobs (file deletion, user deletion)
|
||||
|
||||
@@ -108,9 +106,3 @@ See [Database Migrations](./database-migrations.md) for more information about h
|
||||
### Redis
|
||||
|
||||
Immich uses [Redis](https://redis.com/) via [BullMQ](https://docs.bullmq.io/) to manage job queues. Some jobs trigger subsequent jobs. For example, object detection relies on thumbnail generation and automatically run after one is generated.
|
||||
|
||||
### Typesense
|
||||
|
||||
Immich synchronizes some of the Postgres data into Typesense, so it can execute vector related queries in order to implement certain features including, facial recognition and CLIP search.
|
||||
|
||||
<!-- - [NGINX](https://www.nginx.com/) for internal communication between containers and load balancing when scaling. -->
|
||||
|
||||
@@ -1,18 +1,10 @@
|
||||
# Search
|
||||
|
||||
Immich uses Typesense as the primary search database to enable high performance search mechanism.
|
||||
Immich uses Postgres as its search database for both metadata and smart search.
|
||||
|
||||
Typesense is a powerful search engine that can be integrated with popular natural language processing (NLP) models like CLIP and SBERT to provide highly accurate and relevant search results. Here are some benefits of using Typesense integrated search for CLIP and SBERT:
|
||||
Smart search is powered by the [pgvecto.rs](https://github.com/tensorchord/pgvecto.rs) extension, utilizing machine learning models like CLIP to provide relevant search results. This allows for freeform searches without requiring specific keywords in the image or video metadata.
|
||||
|
||||
Improved Search Accuracy: Typesense uses a combination of indexing, querying, and ranking algorithms to quickly and accurately retrieve relevant search results. When integrated with CLIP and SBERT, Typesense can leverage the semantic understanding and deep learning capabilities of these models to further improve the accuracy of search results.
|
||||
|
||||
Faster Search Response Times: Typesense is optimized for lightning-fast search response times, making it ideal for applications that require near-instantaneous search results. By integrating with CLIP and SBERT, Typesense can reduce the time required to process complex search queries, making it even faster and more efficient.
|
||||
|
||||
Enhanced Semantic Search Capabilities: CLIP and SBERT are powerful NLP models that can extract the semantic meaning from text, enabling more nuanced search queries. By integrating with Typesense, these models can help to improve the accuracy of semantic search, enabling users to find the most relevant results based on the true meaning of their query.
|
||||
|
||||
Greater Search Flexibility: Typesense provides flexible search capabilities, including fuzzy search, partial search, enabling users to find the information they need quickly and easily. When integrated with CLIP and SBERT, Typesense can offer even greater flexibility, allowing users to refine their search queries using natural language and providing more accurate and relevant results.
|
||||
|
||||
(Generated by Chat-GPT4)
|
||||
Metadata search (prefixed with `m:`) can search specifically by text without the use of a model.
|
||||
|
||||
Some search examples:
|
||||
<img src={require('./img/search-ex-2.webp').default} title='Search Example 1' />
|
||||
|
||||
@@ -88,15 +88,6 @@ UPLOAD_LOCATION=absolute_location_on_your_machine_where_you_want_to_store_the_ba
|
||||
|
||||
LOG_LEVEL=simple
|
||||
|
||||
###################################################################################
|
||||
# Typesense
|
||||
###################################################################################
|
||||
# TYPESENSE_ENABLED=false
|
||||
TYPESENSE_API_KEY=some-random-text
|
||||
# TYPESENSE_HOST: typesense
|
||||
# TYPESENSE_PORT: 8108
|
||||
# TYPESENSE_PROTOCOL: http
|
||||
|
||||
###################################################################################
|
||||
# Reverse Geocoding
|
||||
#
|
||||
@@ -137,7 +128,6 @@ PUBLIC_LOGIN_PAGE_MESSAGE="My Family Photos and Videos Backup Server"
|
||||
- Populate custom database information if necessary.
|
||||
- Populate `UPLOAD_LOCATION` with your preferred location for storing backup assets.
|
||||
- Consider changing `DB_PASSWORD` to something randomly generated
|
||||
- Consider changing `TYPESENSE_API_KEY` to something randomly generated
|
||||
|
||||
### Step 3 - Start the containers
|
||||
|
||||
|
||||
@@ -17,10 +17,10 @@ If this should not work, try running `docker compose up -d --force-recreate`.
|
||||
|
||||
## Docker Compose
|
||||
|
||||
| Variable | Description | Default | Services |
|
||||
| :---------------- | :-------------------- | :-------: | :------------------------------------------------------------- |
|
||||
| `IMMICH_VERSION` | Image tags | `release` | server, microservices, machine learning, web, proxy, typesense |
|
||||
| `UPLOAD_LOCATION` | Host Path for uploads | | server, microservices |
|
||||
| Variable | Description | Default | Services |
|
||||
| :---------------- | :-------------------- | :-------: | :-------------------------------------------------- |
|
||||
| `IMMICH_VERSION` | Image tags | `release` | server, microservices, machine learning, web, proxy |
|
||||
| `UPLOAD_LOCATION` | Host Path for uploads | | server, microservices |
|
||||
|
||||
:::tip
|
||||
|
||||
@@ -124,51 +124,6 @@ Redis (Sentinel) URL example JSON before encoding:
|
||||
}
|
||||
```
|
||||
|
||||
## Typesense
|
||||
|
||||
| Variable | Description | Default | Services |
|
||||
| :------------------- | :----------------------- | :---------: | :------------------------------- |
|
||||
| `TYPESENSE_ENABLED` | Enable Typesense | | server, microservices |
|
||||
| `TYPESENSE_URL` | Typesense URL | | server, microservices |
|
||||
| `TYPESENSE_HOST` | Typesense Host | `typesense` | server, microservices |
|
||||
| `TYPESENSE_PORT` | Typesense Port | `8108` | server, microservices |
|
||||
| `TYPESENSE_PROTOCOL` | Typesense Protocol | `http` | server, microservices |
|
||||
| `TYPESENSE_API_KEY` | Typesense API Key | | server, microservices, typesense |
|
||||
| `TYPESENSE_DATA_DIR` | Typesense Data Directory | `/data` | typesense |
|
||||
|
||||
:::info
|
||||
|
||||
`TYPESENSE_URL` must start with `ha://` and then include a `base64` encoded JSON string for the configuration.
|
||||
|
||||
`TYPESENSE_ENABLED`: Anything other than `false`, behaves as `true`.
|
||||
Even undefined is treated as `true`.
|
||||
|
||||
- When `TYPESENSE_URL` is defined, the other typesense (`TYPESENSE_*`) variables are ignored.
|
||||
|
||||
:::
|
||||
|
||||
Typesense URL example JSON before encoding:
|
||||
|
||||
```json
|
||||
[
|
||||
{
|
||||
"host": "typesense-1.example.net",
|
||||
"port": "443",
|
||||
"protocol": "https"
|
||||
},
|
||||
{
|
||||
"host": "typesense-2.example.net",
|
||||
"port": "443",
|
||||
"protocol": "https"
|
||||
},
|
||||
{
|
||||
"host": "typesense-3.example.net",
|
||||
"port": "443",
|
||||
"protocol": "https"
|
||||
}
|
||||
]
|
||||
```
|
||||
|
||||
## Machine Learning
|
||||
|
||||
| Variable | Description | Default | Services |
|
||||
|
||||
Reference in New Issue
Block a user