Vespa Deployment (#330)

Large Change!
2025-04-08 03:48:14 +02:00 · 2023-08-26 15:35:19 -07:00 · 2023-08-26 15:35:19 -07:00 · a2ec1e2cda
commit a2ec1e2cda
parent 642862bede
25 changed files with 584 additions and 289 deletions
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@ -45,9 +45,8 @@ We would love to see you there!

 ## Get Started 🚀
 Danswer being a fully functional app, relies on several external pieces of software, specifically:
- Postgres
- Vector DB ([Qdrant](https://github.com/qdrant/qdrant))
- Search Engine ([Typesense](https://github.com/typesense/typesense))
+- Postgres (Relational DB)
+- [Vespa](https://vespa.ai/) (Vector DB/Search Engine)

 This guide provides instructions to set up the Danswer specific services outside of Docker because it's easier for
 development purposes but also feel free to just use the containers and update with local changes by providing the
@ -101,14 +100,9 @@ Postgres:
 docker compose -f docker-compose.dev.yml -p danswer-stack up -d relational_db
 ```

-Qdrant:
+Vespa:
 ```bash
-docker compose -f docker-compose.dev.yml -p danswer-stack up -d vector_db
-```
-
-Typesense:
-```bash
-docker compose -f docker-compose.dev.yml -p danswer-stack up -d search_engine
+docker compose -f docker-compose.dev.yml -p danswer-stack up -d index
 ```


@ -129,37 +123,52 @@ _for Windows, run:_
 ```


-The first time running Danswer, you will need to run the DB migrations.
+The first time running Danswer, you will need to run the DB migrations for Postgres.
 Navigate to `danswer/backend` and with the venv active, run:
 ```bash
 alembic upgrade head
 ```

-To run the backend API server, navigate to `danswer/backend` and run:
+Additionally, we have to package the Vespa schema deployment:
+Nagivate to `danswer/backend/danswer/datastores/vespa/app_config` and run:
 ```bash
-DISABLE_AUTH=True TYPESENSE_API_KEY=typesense_api_key DYNAMIC_CONFIG_DIR_PATH=./dynamic_config_storage uvicorn danswer.main:app --reload --port 8080
+zip -r ../vespa-app.zip .
+```
+- Note: If you don't have the `zip` utility, you will need to install it prior to running the above
+
+To run the backend API server, navigate back to `danswer/backend` and run:
+```bash
+DISABLE_AUTH=True \
+DYNAMIC_CONFIG_DIR_PATH=./dynamic_config_storage \
+VESPA_DEPLOYMENT_ZIP=./danswer/datastores/vespa/vespa-app.zip \
+uvicorn danswer.main:app --reload --port 8080
 ```
 _For Windows (for compatibility with both PowerShell and Command Prompt):_
 ```bash
-powershell -Command " $env:DISABLE_AUTH='True'; $env:TYPESENSE_API_KEY='typesense_api_key'; $env:DYNAMIC_CONFIG_DIR_PATH='./dynamic_config_storage'; uvicorn danswer.main:app --reload --port 8080 "
+powershell -Command "
+    $env:DISABLE_AUTH='True'
+    $env:DYNAMIC_CONFIG_DIR_PATH='./dynamic_config_storage'
+    $env:VESPA_DEPLOYMENT_ZIP='./danswer/datastores/vespa/vespa-app.zip'
+    uvicorn danswer.main:app --reload --port 8080 
+"
 ```

 To run the background job to check for connector updates and index documents, navigate to `danswer/backend` and run:
 ```bash
-PYTHONPATH=. TYPESENSE_API_KEY=typesense_api_key DYNAMIC_CONFIG_DIR_PATH=./dynamic_config_storage python danswer/background/update.py
+PYTHONPATH=. DYNAMIC_CONFIG_DIR_PATH=./dynamic_config_storage python danswer/background/update.py
 ```
 _For Windows:_
 ```bash
-powershell -Command " $env:PYTHONPATH='.'; $env:TYPESENSE_API_KEY='typesense_api_key'; $env:DYNAMIC_CONFIG_DIR_PATH='./dynamic_config_storage'; python danswer/background/update.py "
+powershell -Command " $env:PYTHONPATH='.'; $env:DYNAMIC_CONFIG_DIR_PATH='./dynamic_config_storage'; python danswer/background/update.py "
 ```

 To run the background job which handles deletion of connectors, navigate to `danswer/backend` and run:
 ```bash
-PYTHONPATH=. TYPESENSE_API_KEY=typesense_api_key DYNAMIC_CONFIG_DIR_PATH=./dynamic_config_storage python danswer/background/connector_deletion.py
+PYTHONPATH=. DYNAMIC_CONFIG_DIR_PATH=./dynamic_config_storage python danswer/background/connector_deletion.py
 ```
 _For Windows:_
 ```bash
-powershell -Command " $env:PYTHONPATH='.'; $env:TYPESENSE_API_KEY='typesense_api_key'; $env:DYNAMIC_CONFIG_DIR_PATH='./dynamic_config_storage'; python danswer/background/connector_deletion.py "
+powershell -Command " $env:PYTHONPATH='.'; $env:DYNAMIC_CONFIG_DIR_PATH='./dynamic_config_storage'; python danswer/background/connector_deletion.py "
 ```

 Note: if you need finer logging, add the additional environment variable `LOG_LEVEL=DEBUG` to the relevant services.
--- a/backend/Dockerfile
+++ b/backend/Dockerfile
@ -3,7 +3,7 @@ FROM python:3.11.4-slim-bullseye
 RUN apt-get update \
    && apt-get install -y git cmake pkg-config libprotobuf-c-dev protobuf-compiler \
       libprotobuf-dev libgoogle-perftools-dev libpq-dev build-essential cron curl \
-       supervisor \
+       supervisor zip \
    && rm -rf /var/lib/apt/lists/*

 COPY ./requirements/default.txt /tmp/requirements.txt
@ -17,6 +17,10 @@ COPY ./alembic /app/alembic
 COPY ./alembic.ini /app/alembic.ini
 COPY supervisord.conf /etc/supervisor/conf.d/supervisord.conf

+WORKDIR /app/danswer/datastores/vespa/app_config
+RUN zip -r /app/danswer/vespa-app.zip .
+WORKDIR /app
+
 ENV PYTHONPATH /app

 # By default this container does nothing, it is used by api server and background which specify their own CMD
--- a/backend/danswer/background/update.py
+++ b/backend/danswer/background/update.py
@ -230,6 +230,14 @@ def _run_indexing(
        db_connector = attempt.connector
        db_credential = attempt.credential

+        update_connector_credential_pair(
+            db_session=db_session,
+            connector_id=db_connector.id,
+            credential_id=db_credential.id,
+            attempt_status=IndexingStatus.IN_PROGRESS,
+            run_dt=run_dt,
+        )
+
        try:
            net_doc_change = 0
            document_count = 0
@ -238,7 +246,6 @@ def _run_indexing(
                logger.debug(
                    f"Indexing batch of documents: {[doc.to_short_descriptor() for doc in doc_batch]}"
                )
-
                index_user_id = (
                    None if db_credential.public_doc else db_credential.user_id
                )
@ -286,6 +293,8 @@ def _run_indexing(
                f"Failed connector elapsed time: {time.time() - run_time} seconds"
            )
            mark_attempt_failed(attempt, db_session, failure_reason=str(e))
+            # The last attempt won't be marked failed until the next cycle's check for still in-progress attempts
+            # The connector_credential_pair is marked failed here though to reflect correctly in UI asap
            update_connector_credential_pair(
                db_session=db_session,
                connector_id=attempt.connector.id,
--- a/backend/danswer/configs/app_configs.py
+++ b/backend/danswer/configs/app_configs.py
@ -67,7 +67,7 @@ MASK_CREDENTIAL_PREFIX = (
 DOCUMENT_INDEX_NAME = "danswer_index"  # Shared by vector/keyword indices
 # Vespa is now the default document index store for both keyword and vector
 DOCUMENT_INDEX_TYPE = os.environ.get(
-    "DOCUMENT_INDEX_TYPE", DocumentIndexType.SPLIT.value
+    "DOCUMENT_INDEX_TYPE", DocumentIndexType.COMBINED.value
 )
 VESPA_HOST = os.environ.get("VESPA_HOST") or "localhost"
 VESPA_PORT = os.environ.get("VESPA_PORT") or "8081"
@ -92,11 +92,11 @@ INDEX_BATCH_SIZE = 16

 # below are intended to match the env variables names used by the official postgres docker image
 # https://hub.docker.com/_/postgres
-POSTGRES_USER = os.environ.get("POSTGRES_USER", "postgres")
-POSTGRES_PASSWORD = os.environ.get("POSTGRES_PASSWORD", "password")
-POSTGRES_HOST = os.environ.get("POSTGRES_HOST", "localhost")
-POSTGRES_PORT = os.environ.get("POSTGRES_PORT", "5432")
-POSTGRES_DB = os.environ.get("POSTGRES_DB", "postgres")
+POSTGRES_USER = os.environ.get("POSTGRES_USER") or "postgres"
+POSTGRES_PASSWORD = os.environ.get("POSTGRES_PASSWORD") or "password"
+POSTGRES_HOST = os.environ.get("POSTGRES_HOST") or "localhost"
+POSTGRES_PORT = os.environ.get("POSTGRES_PORT") or "5432"
+POSTGRES_DB = os.environ.get("POSTGRES_DB") or "postgres"


 #####
--- a/backend/danswer/configs/model_configs.py
+++ b/backend/danswer/configs/model_configs.py
@ -9,20 +9,21 @@ from danswer.configs.constants import ModelHostType
 # Models used must be MIT or Apache license
 # Inference/Indexing speed

-# https://www.sbert.net/docs/pretrained_models.html
-# Use 'multi-qa-MiniLM-L6-cos-v1' if license is added because it is 3x faster (384 dimensional embedding)
-# Context size is 256 for above though
-DOCUMENT_ENCODER_MODEL = "sentence-transformers/all-distilroberta-v1"
-DOC_EMBEDDING_DIM = 768  # Depends on the document encoder model
+# https://huggingface.co/thenlper/gte-small
+DOCUMENT_ENCODER_MODEL = "thenlper/gte-small"
+DOC_EMBEDDING_DIM = 384  # Depends on the document encoder model
+NORMALIZE_EMBEDDINGS = False
+# Certain models like BGE use a prefix for asymmetric retrievals (query generally shorter than docs)
+ASYMMETRIC_PREFIX = ""

 # https://www.sbert.net/docs/pretrained-models/ce-msmarco.html
-# Previously using "cross-encoder/ms-marco-MiniLM-L-6-v2" alone
 CROSS_ENCODER_MODEL_ENSEMBLE = [
    "cross-encoder/ms-marco-MiniLM-L-4-v2",
    "cross-encoder/ms-marco-TinyBERT-L-2-v2",
 ]

 # Better to keep it loose, surfacing more results better than missing results
+# Currently unused by Vespa
 SEARCH_DISTANCE_CUTOFF = 0.1  # Cosine similarity (currently), range of -1 to 1 with -1 being completely opposite

 QUERY_MAX_CONTEXT_SIZE = 256
--- a/backend/danswer/datastores/indexing_pipeline.py
+++ b/backend/danswer/datastores/indexing_pipeline.py
@ -102,7 +102,9 @@ def _indexing_pipeline(
        )
        raise e

-    return len(insertion_records), len(chunks)
+    return len([r for r in insertion_records if r.already_existed is False]), len(
+        chunks
+    )


 def build_indexing_pipeline(
--- a/backend/danswer/datastores/qdrant/store.py
+++ b/backend/danswer/datastores/qdrant/store.py
@ -25,7 +25,7 @@ from danswer.datastores.interfaces import VectorIndex
 from danswer.datastores.qdrant.indexing import index_qdrant_chunks
 from danswer.datastores.qdrant.utils import create_qdrant_collection
 from danswer.datastores.qdrant.utils import list_qdrant_collections
-from danswer.search.search_utils import get_default_embedding_model
+from danswer.search.semantic_search import embed_query
 from danswer.utils.batching import batch_generator
 from danswer.utils.clients import get_qdrant_client
 from danswer.utils.logger import setup_logger
@ -175,11 +175,7 @@ class QdrantIndex(VectorIndex):
        distance_cutoff: float | None = SEARCH_DISTANCE_CUTOFF,
        page_size: int = NUM_RETURNED_HITS,
    ) -> list[InferenceChunk]:
-        query_embedding = get_default_embedding_model().encode(
-            query
-        )  # TODO: make this part of the embedder interface
-        if not isinstance(query_embedding, list):
-            query_embedding = query_embedding.tolist()
+        query_embedding = embed_query(query)

        filter_conditions = _build_qdrant_filters(user_id, filters)

--- a/backend/danswer/datastores/vespa/app_config/schemas/danswer_chunk.sd
+++ b/backend/danswer/datastores/vespa/app_config/schemas/danswer_chunk.sd
@ -14,7 +14,11 @@ schema danswer_chunk {
        # Need to consider that not every doc has a separable title (ie. slack message)
        # Set summary options to enable bolding
        field content type string {
-            indexing: summary | attribute | index
+            indexing: summary | index
+            match {
+                gram
+                gram-size: 3
+            }
            index: enable-bm25
        }
        # https://docs.vespa.ai/en/attributes.html potential enum store for speed, but probably not worth it
@ -38,7 +42,7 @@ schema danswer_chunk {
        field metadata type string {
            indexing: summary | attribute
        }
-        field embeddings type tensor<float>(t{},x[768]) {
+        field embeddings type tensor<float>(t{},x[384]) {
            indexing: attribute
            attribute {
                distance-metric: angular
@ -66,7 +70,7 @@ schema danswer_chunk {

    rank-profile semantic_search inherits default {
        inputs {
-            query(query_embedding) tensor<float>(x[768])
+            query(query_embedding) tensor<float>(x[384])
        }
        first-phase {
            expression: closeness(field, embeddings) * attribute(boost)
@ -76,7 +80,7 @@ schema danswer_chunk {

    rank-profile hybrid_search inherits default {
        inputs {
-            query(query_embedding) tensor<float>(x[768])
+            query(query_embedding) tensor<float>(x[384])
        }
        first-phase {
            expression: bm25(content)
--- a/backend/danswer/datastores/vespa/app_config/services.xml
+++ b/backend/danswer/datastores/vespa/app_config/services.xml
@ -3,6 +3,9 @@
    <container id="default" version="1.0">
        <document-api/>
        <search/>
+        <http>
+            <server id="default" port="8081"/>
+        </http>
        <nodes>
            <node hostalias="node1" />
        </nodes>
@ -15,5 +18,12 @@
        <nodes>
            <node hostalias="node1" distribution-key="0" />
        </nodes>
+        <tuning>
+            <resource-limits>
+                <!-- Default is 75% but this should be increased for personal computer dockerized deployments -->
+                <!-- https://docs.vespa.ai/en/operations/feed-block.html -->
+                <disk>0.98</disk>
+            </resource-limits>
+        </tuning>
    </content>
 </services>
--- a/backend/danswer/datastores/vespa/store.py
+++ b/backend/danswer/datastores/vespa/store.py
@ -1,10 +1,12 @@
 import json
 from collections.abc import Mapping
-from functools import partial
+from typing import Any
 from typing import cast
 from uuid import UUID

 import requests
+from requests import HTTPError
+from requests import Response

 from danswer.chunking.models import IndexChunk
 from danswer.chunking.models import InferenceChunk
@ -39,7 +41,9 @@ from danswer.datastores.interfaces import DocumentIndex
 from danswer.datastores.interfaces import DocumentInsertionRecord
 from danswer.datastores.interfaces import IndexFilter
 from danswer.datastores.interfaces import UpdateRequest
+from danswer.datastores.vespa.utils import remove_invalid_unicode_chars
 from danswer.search.search_utils import get_default_embedding_model
+from danswer.search.semantic_search import embed_query
 from danswer.utils.logger import setup_logger

 logger = setup_logger()
@ -123,7 +127,9 @@ def _index_vespa_chunks(
    chunks: list[IndexChunk],
    index_attempt_metadata: IndexAttemptMetadata,
 ) -> set[DocumentInsertionRecord]:
-    json_header = {"Content-Type": "application/json"}
+    json_header = {
+        "Content-Type": "application/json",
+    }
    insertion_records: set[DocumentInsertionRecord] = set()
    cross_connector_document_metadata_map: dict[
        str, CrossConnectorDocumentMetadata
@ -151,6 +157,7 @@ def _index_vespa_chunks(
                )
            already_existing_documents.add(document.id)

+        # No minichunk documents in vespa, minichunk vectors are stored in the chunk itself
        vespa_chunk_id = str(get_uuid_from_chunk(chunk))

        embeddings = chunk.embeddings
@ -159,32 +166,64 @@ def _index_vespa_chunks(
            for ind, m_c_embed in enumerate(embeddings.mini_chunk_embeddings):
                embeddings_name_vector_map[f"mini_chunk_{ind}"] = m_c_embed

-        vespa_document = {
-            "fields": {
-                DOCUMENT_ID: document.id,
-                CHUNK_ID: chunk.chunk_id,
-                BLURB: chunk.blurb,
-                CONTENT: chunk.content,
-                SOURCE_TYPE: str(document.source.value),
-                SOURCE_LINKS: json.dumps(chunk.source_links),
-                SEMANTIC_IDENTIFIER: document.semantic_identifier,
-                SECTION_CONTINUATION: chunk.section_continuation,
-                METADATA: json.dumps(document.metadata),
-                EMBEDDINGS: embeddings_name_vector_map,
-                BOOST: 1,  # Boost value always starts at 1 for 0 impact on weight
-                ALLOWED_USERS: cross_connector_document_metadata_map[
-                    document.id
-                ].allowed_users,
-                ALLOWED_GROUPS: cross_connector_document_metadata_map[
-                    document.id
-                ].allowed_user_groups,
-            }
+        vespa_document_fields = {
+            DOCUMENT_ID: document.id,
+            CHUNK_ID: chunk.chunk_id,
+            BLURB: chunk.blurb,
+            CONTENT: chunk.content,
+            SOURCE_TYPE: str(document.source.value),
+            SOURCE_LINKS: json.dumps(chunk.source_links),
+            SEMANTIC_IDENTIFIER: document.semantic_identifier,
+            SECTION_CONTINUATION: chunk.section_continuation,
+            METADATA: json.dumps(document.metadata),
+            EMBEDDINGS: embeddings_name_vector_map,
+            BOOST: 1,  # Boost value always starts at 1 for 0 impact on weight
+            ALLOWED_USERS: cross_connector_document_metadata_map[
+                document.id
+            ].allowed_users,
+            ALLOWED_GROUPS: cross_connector_document_metadata_map[
+                document.id
+            ].allowed_user_groups,
        }

-        url = f"{DOCUMENT_ID_ENDPOINT}/{vespa_chunk_id}"
+        def _index_chunk(
+            url: str,
+            headers: dict[str, str],
+            fields: dict[str, Any],
+        ) -> Response:
+            logger.debug(
+                f"Hitting URL '{url}', with headers '{headers}', with fields '{fields}'"
+            )
+            res = requests.post(url, headers=headers, json={"fields": fields})
+            try:
+                res.raise_for_status()
+                return res
+            except Exception as e:
+                logger.error(
+                    f"Failed to index document: '{document.id}'. Got response: '{res.text}'"
+                )
+                raise e

-        res = requests.post(url, headers=json_header, json=vespa_document)
-        res.raise_for_status()
+        vespa_url = f"{DOCUMENT_ID_ENDPOINT}/{vespa_chunk_id}"
+        try:
+            _index_chunk(vespa_url, json_header, vespa_document_fields)
+        except HTTPError as e:
+            if cast(Response, e.response).status_code != 400:
+                raise e
+
+            # if it's a 400 response, try again with invalid unicode chars removed
+            # only doing this on error to avoid having to go through the content
+            # char by char every time
+            vespa_document_fields[BLURB] = remove_invalid_unicode_chars(
+                cast(str, vespa_document_fields[BLURB])
+            )
+            vespa_document_fields[SEMANTIC_IDENTIFIER] = remove_invalid_unicode_chars(
+                cast(str, vespa_document_fields[SEMANTIC_IDENTIFIER])
+            )
+            vespa_document_fields[CONTENT] = remove_invalid_unicode_chars(
+                cast(str, vespa_document_fields[CONTENT])
+            )
+            _index_chunk(vespa_url, json_header, vespa_document_fields)

        insertion_records.add(
            DocumentInsertionRecord(
@ -218,11 +257,13 @@ def _build_vespa_filters(
            }
            for filter_key, filter_val in valid_filters.items():
                if isinstance(filter_val, str):
-                    filter_str += f'{filter_key} = "{filter_val}" and '
+                    filter_str += f'{filter_key} contains "{filter_val}" and '
                elif isinstance(filter_val, list):
-                    quoted_elems = [f'"{elem}"' for elem in filter_val]
-                    filters_or = ",".join(quoted_elems)
-                    filter_str += f"{filter_key} in [{filters_or}] and "
+                    eq_elems = [
+                        f'{filter_key} contains "{elem}"' for elem in filter_val
+                    ]
+                    filters_or = " or ".join(eq_elems)
+                    filter_str += f"({filters_or}) and "
                else:
                    raise ValueError("Invalid filters provided")
    return filter_str
@ -271,6 +312,7 @@ class VespaIndex(DocumentIndex):
        If the changes cannot be applied without conflict with existing data, it will fail with a non 200
        """
        deploy_url = f"{VESPA_APPLICATION_ENDPOINT}/tenant/default/prepareandactivate"
+        logger.debug(f"Sending Vespa zip to {deploy_url}")
        headers = {"Content-Type": "application/zip"}
        with open(self.deployment_zip, "rb") as f:
            response = requests.post(deploy_url, headers=headers, data=f)
@ -365,9 +407,7 @@ class VespaIndex(DocumentIndex):
            + f"({{targetHits: {10 * num_to_retrieve}}}nearestNeighbor(embeddings, query_embedding))"
        )

-        query_embedding = get_default_embedding_model().encode(query)
-        if not isinstance(query_embedding, list):
-            query_embedding = query_embedding.tolist()
+        query_embedding = embed_query(query)

        params = {
            "yql": yql,
@ -388,12 +428,11 @@ class VespaIndex(DocumentIndex):
        yql = (
            VespaIndex.yql_base
            + vespa_where_clauses
-            + f'{{targetHits: {10 * num_to_retrieve}}}nearestNeighbor(embeddings, query_embedding) or {{grammar: "weakAnd"}}userInput(@query)'
+            + f"{{targetHits: {10 * num_to_retrieve}}}nearestNeighbor(embeddings, query_embedding) or "
+            + f'{{grammar: "weakAnd"}}userInput(@query)'
        )

-        query_embedding = get_default_embedding_model().encode(query)
-        if not isinstance(query_embedding, list):
-            query_embedding = query_embedding.tolist()
+        query_embedding = embed_query(query)

        params = {
            "yql": yql,
--- a/backend/danswer/datastores/vespa/utils.py
+++ b/backend/danswer/datastores/vespa/utils.py
@ -0,0 +1,12 @@
+import re
+
+
+_illegal_xml_chars_RE = re.compile(
+    "[\x00-\x08\x0b\x0c\x0e-\x1F\uD800-\uDFFF\uFFFE\uFFFF]"
+)
+
+
+def remove_invalid_unicode_chars(text: str) -> str:
+    """Vespa does not take in unicode chars that aren't valid for XML.
+    This removes them."""
+    return _illegal_xml_chars_RE.sub("", text)
--- a/backend/danswer/search/semantic_search.py
+++ b/backend/danswer/search/semantic_search.py
@ -12,7 +12,9 @@ from danswer.configs.app_configs import ENABLE_MINI_CHUNK
 from danswer.configs.app_configs import MINI_CHUNK_SIZE
 from danswer.configs.app_configs import NUM_RERANKED_RESULTS
 from danswer.configs.app_configs import NUM_RETURNED_HITS
+from danswer.configs.model_configs import ASYMMETRIC_PREFIX
 from danswer.configs.model_configs import BATCH_SIZE_ENCODE_CHUNKS
+from danswer.configs.model_configs import NORMALIZE_EMBEDDINGS
 from danswer.datastores.interfaces import DocumentIndex
 from danswer.datastores.interfaces import IndexFilter
 from danswer.search.models import Embedder
@ -90,7 +92,10 @@ def retrieve_ranked_documents(
        for ranked_chunk in ranked_chunks
        if ranked_chunk.source_links is not None
    ]
-    files_log_msg = f"Top links from semantic search: {', '.join(top_docs)}"
+
+    files_log_msg = (
+        f"Top links from semantic search: {', '.join(list(dict.fromkeys(top_docs)))}"
+    )
    logger.info(files_log_msg)

    return ranked_chunks, top_chunks[num_rerank:]
@ -155,7 +160,12 @@ def encode_chunks(

    embeddings_np: list[numpy.ndarray] = []
    for text_batch in text_batches:
-        embeddings_np.extend(embedding_model.encode(text_batch))
+        # Normalize embeddings is only configured via model_configs.py, be sure to use right value for the set loss
+        embeddings_np.extend(
+            embedding_model.encode(
+                text_batch, normalize_embeddings=NORMALIZE_EMBEDDINGS
+            )
+        )
    embeddings: list[list[float]] = [embedding.tolist() for embedding in embeddings_np]

    embedding_ind_start = 0
@ -177,6 +187,24 @@ def encode_chunks(
    return embedded_chunks


+def embed_query(
+    query: str,
+    embedding_model: SentenceTransformer | None = None,
+    prefix: str = ASYMMETRIC_PREFIX,
+    normalize_embeddings: bool = NORMALIZE_EMBEDDINGS,
+) -> list[float]:
+    model = embedding_model or get_default_embedding_model()
+    prefixed_query = prefix + query
+    query_embedding = model.encode(
+        prefixed_query, normalize_embeddings=normalize_embeddings
+    )
+
+    if not isinstance(query_embedding, list):
+        query_embedding = query_embedding.tolist()
+
+    return query_embedding
+
+
 class DefaultEmbedder(Embedder):
    def embed(self, chunks: list[DocAwareChunk]) -> list[IndexChunk]:
        return encode_chunks(chunks)
--- a/backend/danswer/server/manage.py
+++ b/backend/danswer/server/manage.py
@ -287,7 +287,7 @@ def get_connector_indexing_status(
        latest_index_attempt = cc_pair_to_latest_index_attempt.get(
            (connector.id, credential.id)
        )
-        deletion_attemts = deletion_attempts_by_connector.get(connector.id, [])
+        deletion_attempts = deletion_attempts_by_connector.get(connector.id, [])
        indexing_statuses.append(
            ConnectorIndexingStatus(
                connector=ConnectorSnapshot.from_connector_db_model(connector),
@ -309,7 +309,7 @@ def get_connector_indexing_status(
                    DeletionAttemptSnapshot.from_deletion_attempt_db_model(
                        deletion_attempt
                    )
-                    for deletion_attempt in deletion_attemts
+                    for deletion_attempt in deletion_attempts
                ],
                is_deletable=check_deletion_attempt_is_allowed(
                    connector_credential_pair=cc_pair
--- a/deployment/docker_compose/docker-compose.dev.legacy.yml
+++ b/deployment/docker_compose/docker-compose.dev.legacy.yml
@ -0,0 +1,155 @@
+# This legacy version runs the app with typesense and qdrant together as the document indices
+# Danswer is moving forward with Vespa to offer a consolidated index and a better search experience
+version: '3'
+services:
+  api_server:
+    image: danswer/danswer-backend:latest
+    build:
+      context: ../../backend
+      dockerfile: Dockerfile
+    command: >
+      /bin/sh -c "alembic upgrade head &&
+      echo \"Starting Danswer Api Server\" &&
+      uvicorn danswer.main:app --host 0.0.0.0 --port 8080"
+    depends_on:
+      - relational_db
+      - vector_db
+      - search_engine
+    restart: always
+    ports:
+      - "8080:8080"
+    environment:
+      - DOCUMENT_INDEX_TYPE=split
+      - INTERNAL_MODEL_VERSION=${INTERNAL_MODEL_VERSION:-openai-chat-completion}
+      - GEN_AI_MODEL_VERSION=${GEN_AI_MODEL_VERSION:-gpt-3.5-turbo}
+      - GEN_AI_API_KEY=${GEN_AI_API_KEY:-}
+      - GEN_AI_ENDPOINT=${GEN_AI_ENDPOINT:-}
+      - GEN_AI_HOST_TYPE=${GEN_AI_HOST_TYPE:-}
+      - POSTGRES_HOST=relational_db
+      - QDRANT_HOST=vector_db
+      - TYPESENSE_HOST=search_engine
+      - TYPESENSE_API_KEY=${TYPESENSE_API_KEY:-typesense_api_key}
+      - LOG_LEVEL=${LOG_LEVEL:-info}
+      - DISABLE_AUTH=${DISABLE_AUTH:-True}
+      - QA_TIMEOUT=${QA_TIMEOUT:-}
+      - OAUTH_TYPE=${OAUTH_TYPE:-google}
+      - OPENID_CONFIG_URL=${OPENID_CONFIG_URL:-}
+      - GOOGLE_OAUTH_CLIENT_ID=${GOOGLE_OAUTH_CLIENT_ID:-}
+      - GOOGLE_OAUTH_CLIENT_SECRET=${GOOGLE_OAUTH_CLIENT_SECRET:-}
+      - DISABLE_GENERATIVE_AI=${DISABLE_GENERATIVE_AI:-}
+      - API_BASE_OPENAI=${API_BASE_OPENAI:-}
+      - API_TYPE_OPENAI=${API_TYPE_OPENAI:-}
+      - API_VERSION_OPENAI=${API_VERSION_OPENAI:-}
+      - AZURE_DEPLOYMENT_ID=${AZURE_DEPLOYMENT_ID:-}
+    volumes:
+      - local_dynamic_storage:/home/storage
+      - file_connector_tmp_storage:/home/file_connector_storage
+      - model_cache_torch:/root/.cache/torch/
+      - model_cache_nltk:/root/nltk_data/
+      - model_cache_huggingface:/root/.cache/huggingface/
+  background:
+    image: danswer/danswer-backend:latest
+    build:
+      context: ../../backend
+      dockerfile: Dockerfile
+    command: /usr/bin/supervisord
+    depends_on:
+      - relational_db
+      - vector_db
+    restart: always
+    environment:
+      - DOCUMENT_INDEX_TYPE=split
+      - INTERNAL_MODEL_VERSION=${INTERNAL_MODEL_VERSION:-openai-chat-completion}
+      - GEN_AI_MODEL_VERSION=${GEN_AI_MODEL_VERSION:-gpt-3.5-turbo}
+      - GEN_AI_API_KEY=${GEN_AI_API_KEY:-}
+      - GEN_AI_ENDPOINT=${GEN_AI_ENDPOINT:-}
+      - GEN_AI_HOST_TYPE=${GEN_AI_HOST_TYPE:-}
+      - POSTGRES_HOST=relational_db
+      - QDRANT_HOST=vector_db
+      - TYPESENSE_HOST=search_engine
+      - TYPESENSE_API_KEY=${TYPESENSE_API_KEY:-typesense_api_key}
+      - API_BASE_OPENAI=${API_BASE_OPENAI:-}
+      - API_TYPE_OPENAI=${API_TYPE_OPENAI:-}
+      - API_VERSION_OPENAI=${API_VERSION_OPENAI:-}
+      - AZURE_DEPLOYMENT_ID=${AZURE_DEPLOYMENT_ID:-}
+      - CONTINUE_ON_CONNECTOR_FAILURE=${CONTINUE_ON_CONNECTOR_FAILURE:-}
+      - NUM_INDEXING_WORKERS=${NUM_INDEXING_WORKERS:-}
+      - DANSWER_BOT_SLACK_APP_TOKEN=${DANSWER_BOT_SLACK_APP_TOKEN:-}
+      - DANSWER_BOT_SLACK_BOT_TOKEN=${DANSWER_BOT_SLACK_BOT_TOKEN:-}
+      - LOG_LEVEL=${LOG_LEVEL:-info}
+    volumes:
+      - local_dynamic_storage:/home/storage
+      - file_connector_tmp_storage:/home/file_connector_storage
+      - model_cache_torch:/root/.cache/torch/
+      - model_cache_nltk:/root/nltk_data/
+      - model_cache_huggingface:/root/.cache/huggingface/
+  web_server:
+    image: danswer/danswer-web-server:latest
+    build:
+      context: ../../web
+      dockerfile: Dockerfile
+      args:
+        - NEXT_PUBLIC_DISABLE_STREAMING=${NEXT_PUBLIC_DISABLE_STREAMING:-false}
+    depends_on:
+      - api_server
+    restart: always
+    environment:
+      - INTERNAL_URL=http://api_server:8080
+      - DISABLE_AUTH=${DISABLE_AUTH:-True}
+      - OAUTH_NAME=${OAUTH_NAME:-}
+  relational_db:
+    image: postgres:15.2-alpine
+    restart: always
+    environment:
+      - POSTGRES_USER=${POSTGRES_USER:-postgres}
+      - POSTGRES_PASSWORD=${POSTGRES_PASSWORD:-password}
+    ports:
+      - "5432:5432"
+    volumes:
+      - db_volume:/var/lib/postgresql/data
+  vector_db:
+    image: qdrant/qdrant:v1.3.0
+    restart: always
+    environment:
+      - QDRANT__TELEMETRY_DISABLED=true
+    ports:
+      - "6333:6333"
+    volumes:
+      - qdrant_volume:/qdrant/storage
+  search_engine:
+    image: typesense/typesense:0.24.1
+    restart: always
+    environment:
+      - TYPESENSE_API_KEY=${TYPESENSE_API_KEY:-typesense_api_key}
+      - TYPESENSE_DATA_DIR=/typesense/storage
+    ports:
+      - "8108:8108"
+    volumes:
+      - typesense_volume:/typesense/storage
+  nginx:
+    image: nginx:1.23.4-alpine
+    restart: always
+    # nginx will immediately crash with `nginx: [emerg] host not found in upstream`
+    # if api_server / web_server are not up 
+    depends_on:
+      - api_server
+      - web_server
+    environment:
+      - DOMAIN=localhost
+    ports:
+      - "80:80"
+      - "3000:80"  # allow for localhost:3000 usage, since that is the norm
+    volumes:
+      - ../data/nginx:/etc/nginx/conf.d
+    command: > 
+      /bin/sh -c "envsubst '$$\{DOMAIN\}' < /etc/nginx/conf.d/app.conf.template.dev > /etc/nginx/conf.d/app.conf &&
+      while :; do sleep 6h & wait $${!}; nginx -s reload; done & nginx -g \"daemon off;\""
+volumes:
+  local_dynamic_storage:
+  file_connector_tmp_storage:  # used to store files uploaded by the user temporarily while we are indexing them
+  db_volume:
+  qdrant_volume:
+  typesense_volume:
+  model_cache_torch:
+  model_cache_nltk:
+  model_cache_huggingface:
--- a/deployment/docker_compose/docker-compose.dev.yml
+++ b/deployment/docker_compose/docker-compose.dev.yml
@ -11,8 +11,7 @@ services:
      uvicorn danswer.main:app --host 0.0.0.0 --port 8080"
    depends_on:
      - relational_db
-      - vector_db
-      - search_engine
+      - index
    restart: always
    ports:
      - "8080:8080"
@ -23,9 +22,7 @@ services:
      - GEN_AI_ENDPOINT=${GEN_AI_ENDPOINT:-}
      - GEN_AI_HOST_TYPE=${GEN_AI_HOST_TYPE:-}
      - POSTGRES_HOST=relational_db
-      - QDRANT_HOST=vector_db
-      - TYPESENSE_HOST=search_engine
-      - TYPESENSE_API_KEY=${TYPESENSE_API_KEY:-typesense_api_key}
+      - VESPA_HOST=index
      - LOG_LEVEL=${LOG_LEVEL:-info}
      - DISABLE_AUTH=${DISABLE_AUTH:-True}
      - QA_TIMEOUT=${QA_TIMEOUT:-}
@ -52,7 +49,7 @@ services:
    command: /usr/bin/supervisord
    depends_on:
      - relational_db
-      - vector_db
+      - index
    restart: always
    environment:
      - INTERNAL_MODEL_VERSION=${INTERNAL_MODEL_VERSION:-openai-chat-completion}
@ -61,9 +58,7 @@ services:
      - GEN_AI_ENDPOINT=${GEN_AI_ENDPOINT:-}
      - GEN_AI_HOST_TYPE=${GEN_AI_HOST_TYPE:-}
      - POSTGRES_HOST=relational_db
-      - QDRANT_HOST=vector_db
-      - TYPESENSE_HOST=search_engine
-      - TYPESENSE_API_KEY=${TYPESENSE_API_KEY:-typesense_api_key}
+      - VESPA_HOST=index
      - API_BASE_OPENAI=${API_BASE_OPENAI:-}
      - API_TYPE_OPENAI=${API_TYPE_OPENAI:-}
      - API_VERSION_OPENAI=${API_VERSION_OPENAI:-}
@ -103,25 +98,14 @@ services:
      - "5432:5432"
    volumes:
      - db_volume:/var/lib/postgresql/data
-  vector_db:
-    image: qdrant/qdrant:v1.3.0
+  index:
+    image: vespaengine/vespa:8
    restart: always
-    environment:
-      - QDRANT__TELEMETRY_DISABLED=true
    ports:
-      - "6333:6333"
+      - "19071:19071"
+      - "8081:8081"
    volumes:
-      - qdrant_volume:/qdrant/storage
-  search_engine:
-    image: typesense/typesense:0.24.1
-    restart: always
-    environment:
-      - TYPESENSE_API_KEY=${TYPESENSE_API_KEY:-typesense_api_key}
-      - TYPESENSE_DATA_DIR=/typesense/storage
-    ports:
-      - "8108:8108"
-    volumes:
-      - typesense_volume:/typesense/storage
+      - vespa_volume:/opt/vespa/var
  nginx:
    image: nginx:1.23.4-alpine
    restart: always
@ -144,8 +128,7 @@ volumes:
  local_dynamic_storage:
  file_connector_tmp_storage:  # used to store files uploaded by the user temporarily while we are indexing them
  db_volume:
-  qdrant_volume:
-  typesense_volume:
+  vespa_volume:
  model_cache_torch:
  model_cache_nltk:
  model_cache_huggingface:
--- a/deployment/docker_compose/docker-compose.prod.legacy.yml
+++ b/deployment/docker_compose/docker-compose.prod.legacy.yml
@ -0,0 +1,128 @@
+version: '3'
+services:
+  api_server:
+    image: danswer/danswer-backend:latest
+    build:
+      context: ../../backend
+      dockerfile: Dockerfile
+    command: >
+      /bin/sh -c "alembic upgrade head &&
+      echo \"Starting Danswer Api Server\" &&
+      uvicorn danswer.main:app --host 0.0.0.0 --port 8080"
+    depends_on:
+      - relational_db
+      - vector_db
+      - search_engine
+    restart: always
+    env_file:
+      - .env
+    environment:
+      - DOCUMENT_INDEX_TYPE=split
+      - POSTGRES_HOST=relational_db
+      - QDRANT_HOST=vector_db
+      - TYPESENSE_HOST=search_engine
+    volumes:
+      - local_dynamic_storage:/home/storage
+      - file_connector_tmp_storage:/home/file_connector_storage
+      - model_cache_torch:/root/.cache/torch/
+      - model_cache_nltk:/root/nltk_data/
+      - model_cache_huggingface:/root/.cache/huggingface/
+  background:
+    image: danswer/danswer-backend:latest
+    build:
+      context: ../../backend
+      dockerfile: Dockerfile
+    command: /usr/bin/supervisord
+    depends_on:
+      - relational_db
+      - vector_db
+    restart: always
+    env_file:
+      - .env
+    environment:
+      - DOCUMENT_INDEX_TYPE=split
+      - POSTGRES_HOST=relational_db
+      - QDRANT_HOST=vector_db
+      - TYPESENSE_HOST=search_engine
+    volumes:
+      - local_dynamic_storage:/home/storage
+      - file_connector_tmp_storage:/home/file_connector_storage
+      - model_cache_torch:/root/.cache/torch/
+      - model_cache_nltk:/root/nltk_data/
+      - model_cache_huggingface:/root/.cache/huggingface/
+  web_server:
+    image: danswer/danswer-web-server:latest
+    build:
+      context: ../../web
+      dockerfile: Dockerfile
+    depends_on:
+      - api_server
+    restart: always
+    env_file:
+      - .env
+    environment:
+      - INTERNAL_URL=http://api_server:8080
+  relational_db:
+    image: postgres:15.2-alpine
+    restart: always
+    # POSTGRES_USER and POSTGRES_PASSWORD should be set in .env file
+    env_file:
+      - .env
+    volumes:
+      - db_volume:/var/lib/postgresql/data
+  vector_db:
+    image: qdrant/qdrant:v1.3.0
+    restart: always
+    env_file:
+      - .env
+    environment:
+      - QDRANT__TELEMETRY_DISABLED=true
+    volumes:
+      - qdrant_volume:/qdrant/storage
+  search_engine:
+    image: typesense/typesense:0.24.1
+    restart: always
+    # TYPESENSE_API_KEY must be set in .env file
+    environment:
+      - TYPESENSE_DATA_DIR=/typesense/storage
+    env_file:
+      - .env
+    volumes:
+      - typesense_volume:/typesense/storage
+  nginx:
+    image: nginx:1.23.4-alpine
+    restart: always
+    # nginx will immediately crash with `nginx: [emerg] host not found in upstream`
+    # if api_server / web_server are not up 
+    depends_on:
+      - api_server
+      - web_server
+    ports:
+      - "80:80"
+      - "443:443"
+    volumes:
+      - ../data/nginx:/etc/nginx/conf.d
+      - ../data/certbot/conf:/etc/letsencrypt
+      - ../data/certbot/www:/var/www/certbot
+    command: > 
+      /bin/sh -c "envsubst '$$\{DOMAIN\}' < /etc/nginx/conf.d/app.conf.template > /etc/nginx/conf.d/app.conf 
+      && while :; do sleep 6h & wait $${!}; nginx -s reload; done & nginx -g \"daemon off;\""
+    env_file:
+      - .env.nginx
+  # follows https://pentacent.medium.com/nginx-and-lets-encrypt-with-docker-in-less-than-5-minutes-b4b8a60d3a71
+  certbot:
+    image: certbot/certbot
+    restart: always
+    volumes:
+      - ../data/certbot/conf:/etc/letsencrypt
+      - ../data/certbot/www:/var/www/certbot
+    entrypoint: "/bin/sh -c 'trap exit TERM; while :; do certbot renew; sleep 12h & wait $${!}; done;'"
+volumes:
+  local_dynamic_storage:
+  file_connector_tmp_storage:  # used to store files uploaded by the user temporarily while we are indexing them
+  db_volume:
+  qdrant_volume:
+  typesense_volume:
+  model_cache_torch:
+  model_cache_nltk:
+  model_cache_huggingface:
--- a/deployment/docker_compose/docker-compose.prod.yml
+++ b/deployment/docker_compose/docker-compose.prod.yml
@ -11,15 +11,13 @@ services:
      uvicorn danswer.main:app --host 0.0.0.0 --port 8080"
    depends_on:
      - relational_db
-      - vector_db
-      - search_engine
+      - index
    restart: always
    env_file:
      - .env
    environment:
      - POSTGRES_HOST=relational_db
-      - QDRANT_HOST=vector_db
-      - TYPESENSE_HOST=search_engine
+      - VESPA_HOST=index
    volumes:
      - local_dynamic_storage:/home/storage
      - file_connector_tmp_storage:/home/file_connector_storage
@ -34,14 +32,13 @@ services:
    command: /usr/bin/supervisord
    depends_on:
      - relational_db
-      - vector_db
+      - index
    restart: always
    env_file:
      - .env
    environment:
      - POSTGRES_HOST=relational_db
-      - QDRANT_HOST=vector_db
-      - TYPESENSE_HOST=search_engine
+      - VESPA_HOST=index
    volumes:
      - local_dynamic_storage:/home/storage
      - file_connector_tmp_storage:/home/file_connector_storage
@ -68,25 +65,14 @@ services:
      - .env
    volumes:
      - db_volume:/var/lib/postgresql/data
-  vector_db:
-    image: qdrant/qdrant:v1.3.0
+  index:
+    image: vespaengine/vespa:8
    restart: always
-    env_file:
-      - .env
-    environment:
-      - QDRANT__TELEMETRY_DISABLED=true
+    ports:
+      - "19071:19071"
+      - "8081:8081"
    volumes:
-      - qdrant_volume:/qdrant/storage
-  search_engine:
-    image: typesense/typesense:0.24.1
-    restart: always
-    # TYPESENSE_API_KEY must be set in .env file
-    environment:
-      - TYPESENSE_DATA_DIR=/typesense/storage
-    env_file:
-      - .env
-    volumes:
-      - typesense_volume:/typesense/storage
+      - vespa_volume:/opt/vespa/var
  nginx:
    image: nginx:1.23.4-alpine
    restart: always
@ -119,8 +105,7 @@ volumes:
  local_dynamic_storage:
  file_connector_tmp_storage:  # used to store files uploaded by the user temporarily while we are indexing them
  db_volume:
-  qdrant_volume:
-  typesense_volume:
+  vespa_volume:
  model_cache_torch:
  model_cache_nltk:
  model_cache_huggingface:
--- a/deployment/kubernetes/api_server-service-deployment.yaml
+++ b/deployment/kubernetes/api_server-service-deployment.yaml
@ -42,15 +42,8 @@ spec:
        env:
        - name: POSTGRES_HOST
          value: relational-db-service
-        - name: QDRANT_HOST
-          value: vector-db-service
-        - name: TYPESENSE_HOST
-          value: search-engine-service
-        - name: TYPESENSE_API_KEY
-          valueFrom:
-            secretKeyRef:
-              name: danswer-secrets
-              key: typesense_api_key
+        - name: VESPA_HOST
+          value: document-index-service
        - name: GOOGLE_OAUTH_CLIENT_ID
          valueFrom:
            secretKeyRef:
--- a/deployment/kubernetes/background-deployment.yaml
+++ b/deployment/kubernetes/background-deployment.yaml
@ -20,15 +20,8 @@ spec:
        env:
        - name: POSTGRES_HOST
          value: relational-db-service
-        - name: QDRANT_HOST
-          value: vector-db-service
-        - name: TYPESENSE_HOST
-          value: search-engine-service
-        - name: TYPESENSE_API_KEY
-          valueFrom:
-            secretKeyRef:
-              name: danswer-secrets
-              key: typesense_api_key
+        - name: VESPA_HOST
+          value: document-index-service
        volumeMounts:
        - name: dynamic-storage
          mountPath: /home/storage
--- a/deployment/kubernetes/persistent-volumes.yaml
+++ b/deployment/kubernetes/persistent-volumes.yaml
@ -19,36 +19,3 @@ spec:
  resources:
    requests:
      storage: 1Gi
---
-apiVersion: v1
-kind: PersistentVolumeClaim
-metadata:
-  name: db-volume-claim
-spec:
-  accessModes:
-    - ReadWriteOnce
-  resources:
-    requests:
-      storage: 1Gi
---
-apiVersion: v1
-kind: PersistentVolumeClaim
-metadata:
-  name: qdrant-pvc
-spec:
-  accessModes:
-    - ReadWriteOnce
-  resources:
-    requests:
-      storage: 1Gi
---
-apiVersion: v1
-kind: PersistentVolumeClaim
-metadata:
-  name: typesense-pvc
-spec:
-  accessModes:
-    - ReadWriteOnce
-  resources:
-    requests:
-      storage: 1Gi
--- a/deployment/kubernetes/postgres-service-deployment.yaml
+++ b/deployment/kubernetes/postgres-service-deployment.yaml
@ -9,13 +9,14 @@ spec:
    - protocol: TCP
      port: 5432
      targetPort: 5432
-  type: ClusterIP
+  clusterIP: None
 ---
 apiVersion: apps/v1
-kind: Deployment
+kind: StatefulSet
 metadata:
-  name: relational-db-deployment
+  name: relational-db-statefulset
 spec:
+  serviceName: relational-db-service
  replicas: 1
  selector:
    matchLabels:
@ -43,8 +44,13 @@ spec:
        - containerPort: 5432
        volumeMounts:
        - mountPath: /var/lib/postgresql/data
-          name: db-volume
-      volumes:
-      - name: db-volume
-        persistentVolumeClaim:
-          claimName: db-volume-claim
+          name: db-storage
+  volumeClaimTemplates:
+  - metadata:
+      name: db-storage
+    spec:
+      accessModes: ["ReadWriteOnce"]
+      resources:
+        requests:
+          # Adjust the storage request size as needed.
+          storage: 1Gi
--- a/deployment/kubernetes/qdrant-service-deployment.yaml
+++ b/deployment/kubernetes/qdrant-service-deployment.yaml
@ -1,43 +0,0 @@
-apiVersion: v1
-kind: Service
-metadata:
-  name: vector-db-service
-spec:
-  selector:
-    app: qdrant
-  ports:
-    - name: qdrant-port
-      protocol: TCP
-      port: 6333
-      targetPort: 6333
-  type: LoadBalancer
---
-apiVersion: apps/v1
-kind: Deployment
-metadata:
-  name: qdrant-deployment
-spec:
-  replicas: 1
-  selector:
-    matchLabels:
-      app: qdrant
-  template:
-    metadata:
-      labels:
-        app: qdrant
-    spec:
-      containers:
-      - name: qdrant
-        image: qdrant/qdrant:v1.1.3
-        ports:
-        - containerPort: 6333
-        volumeMounts:
-        - name: qdrant-storage
-          mountPath: /qdrant/storage
-        env:
-        - name: QDRANT__TELEMETRY_DISABLED
-          value: "true"
-      volumes:
-      - name: qdrant-storage
-        persistentVolumeClaim:
-          claimName: qdrant-pvc
--- a/deployment/kubernetes/secrets.yaml
+++ b/deployment/kubernetes/secrets.yaml
@ -7,6 +7,5 @@ type: Opaque
 data:
  postgres_user: cG9zdGdyZXM= # "postgres" base64 encoded
  postgres_password: cGFzc3dvcmQ= # "password" base64 encoded
-  typesense_api_key: dHlwZXNlbnNlX2FwaV9rZXk= # "typesense_api_key" base64 encoded
-  google_oauth_client_id: REPLACE-THIS # You will need to provide this, use echo -n "your-client-id" | base64
-  google_oauth_client_secret: REPLACE-THIS # You will need to provide this, use echo -n "your-client-id" | base64
+  google_oauth_client_id: MjcwNjk3ODEzMi1iMzZnb20wa2Fhb3I2MmlwYWt2dmRxdm91OGRic2d1cC5hcHBzLmdvb2dsZXVzZXJjb250ZW50LmNvbQ== # You will need to provide this, use echo -n "your-client-id" | base64
+  google_oauth_client_secret: R09DU1BYLWlZbDBSN1ZvYnk0cjZJRUFmekRqdjhad0pnOGI= # You will need to provide this, use echo -n "your-client-id" | base64
--- a/deployment/kubernetes/typesense-service-deployment.yaml
+++ b/deployment/kubernetes/typesense-service-deployment.yaml
@ -1,48 +0,0 @@
-apiVersion: v1
-kind: Service
-metadata:
-  name: search-engine-service
-spec:
-  selector:
-    app: typesense
-  ports:
-    - name: typesense-port
-      protocol: TCP
-      port: 8108
-      targetPort: 8108
-  type: LoadBalancer
---
-apiVersion: apps/v1
-kind: Deployment
-metadata:
-  name: typesense-deployment
-spec:
-  replicas: 1
-  selector:
-    matchLabels:
-      app: typesense
-  template:
-    metadata:
-      labels:
-        app: typesense
-    spec:
-      containers:
-      - name: typesense
-        image: typesense/typesense:0.24.1
-        ports:
-        - containerPort: 8108
-        env:
-        - name: TYPESENSE_API_KEY
-          valueFrom:
-            secretKeyRef:
-              name: danswer-secrets
-              key: typesense_api_key
-        - name: TYPESENSE_DATA_DIR
-          value: /typesense/storage
-        volumeMounts:
-        - name: typesense-storage
-          mountPath: /typesense/storage
-      volumes:
-      - name: typesense-storage
-        persistentVolumeClaim:
-          claimName: typesense-pvc
--- a/deployment/kubernetes/vespa-service-deployment.yaml
+++ b/deployment/kubernetes/vespa-service-deployment.yaml
@ -0,0 +1,63 @@
+apiVersion: v1
+kind: Service
+metadata:
+  name: document-index-service
+spec:
+  selector:
+    app: vespa
+  ports:
+    - name: vespa-tenant-port
+      protocol: TCP
+      port: 19071
+      targetPort: 19071
+    - name: vespa-port
+      protocol: TCP
+      port: 8081
+      targetPort: 8081
+  type: LoadBalancer
+---
+apiVersion: apps/v1
+kind: StatefulSet
+metadata:
+  name: vespa
+  labels:
+    app: vespa
+spec:
+  replicas: 1
+  serviceName: vespa
+  selector:
+    matchLabels:
+      app: vespa
+  template:
+    metadata:
+      labels:
+        app: vespa
+    spec:
+      containers:
+      - name: vespa
+        image: vespaengine/vespa:8
+        imagePullPolicy: IfNotPresent
+        securityContext:
+          privileged: true
+          runAsUser: 0
+        ports:
+        - containerPort: 19071
+        - containerPort: 8081
+        readinessProbe:
+          httpGet:
+            path: /state/v1/health
+            port: 19071
+            scheme: HTTP
+        volumeMounts:
+        - name: vespa-storage
+          mountPath: /opt/vespa/var/
+  volumeClaimTemplates:
+  - metadata:
+      name: vespa-storage
+    spec:
+      accessModes:
+      - ReadWriteOnce
+      resources:
+        requests:
+          # Adjust the storage request size as needed.
+          storage: 1Gi