Skip to content

Commit

Permalink
docs: improved the glasskube installation instructions
Browse files Browse the repository at this point in the history
Signed-off-by: Philip Miglinci <[email protected]>
  • Loading branch information
pmig authored and cdxker committed Sep 5, 2024
1 parent f1d0d8c commit 6e114ab
Showing 1 changed file with 55 additions and 3 deletions.
58 changes: 55 additions & 3 deletions glasskube/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -4,9 +4,12 @@
## 1. Start a new minikube cluster

[Minikube](https://minikube.sigs.k8s.io/docs/start/) can be installed via Homebrew or binary downloads.
Also make sure to enable the ingress and metrics plugin

```shell
minikube start -p trieve
minikube addons enable metrics-server -p trieve
minikube addons enable ingress -p trieve
```

## 2. Bootstrap glasskube
Expand Down Expand Up @@ -68,15 +71,64 @@ kubectl port-forward svc/trieve-keycloak-service 8080:8080

Open [dashboard.localtrieve.com](http://dashboard.localtrieve.com) in your browser.

## Optional commands
## Optional commands & debugging info

#### Scale down all embedding servers:
#### Working on low CPU machines

You can scale down all embedding servers to reduce the memory and cpu usage of trieve.

```shell
kubectl scale deployment trieve-embedding-bgem3 trieve-embedding-jina trieve-embedding-reranker trieve-embedding-splade-doc trieve-embedding-splade-query --replicas 0
```

If you still want to do end-to-end test you can only start the `trieve-embedding-bgem3`
and configure the dataset in a specific way:

1. Create a new dataset
2. Disable full text search
3. Import test data

After that the ingestion worker will orchestrate the ingestion.

You can see the ingestion progress by connecting to the redis pod and execute the `redis-cli` and
monitor progress by looking at the queues `llen ingestion` and `llen processing`.

After ingestion is done you can search the dataset by only setting the search type to semantic so
only bgem3 is used.

#### Qdrant

Qdrant throws some weird exceptions during the initial database migrations, but keeps working normally after.

You can monitor the status of the collections by looking at the Qdrant ui:

```shell
glasskube open qdrant
```

You can fetch the API key with:

```
kubectl get secret trieve-qdrant-qdrant-apikey -o jsonpath='{.data.api-key}' | base64 -d
```

#### ClickHouse

In order to test if ClickHouse works you can connect to clickhouse via:

```shell
kubectl scale deployment trieve-embedding-bgem3 trieve-embedding-jina trieve-embedding-reranker trieve-embedding-splade-doc trieve-embedding-splade-query --replicas 0
kubectl exec -it chi-trieve-clickhouse-cluster1-0-0-0 -- clickhouse-client
```

The password is currently `password` but will be changed in the future.

In order to see if the custom embedding function works you can execute:

```
select embed_p('test');
```


#### Deleting the cluster

```shell
Expand Down

0 comments on commit 6e114ab

Please sign in to comment.