Merge branch 'master' into features/worker

Tiledesk · May 27, 2024 · 084ed8e · 084ed8e
2 parents f651faa + cd5fa6c
commit 084ed8e
Show file tree

Hide file tree

Showing 8 changed files with 51 additions and 44 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -5,6 +5,11 @@
     *Andrea Sponziello* 
 ### **Copyrigth**: *Tiledesk SRL*
 
+## [2024-05-20]
+
+### 0.1.17
+- added: PIENCONE_TYPE = "serverless|pod"
+
 ## [2024-05-18]
 
 ### 0.1.16

diff --git a/README.md b/README.md
@@ -13,6 +13,7 @@ pip install -e .
 
 ```commandline
 export REDIS_URL="redis://localhost:6379/0"
+export PINECONE_TYPE="serverless|pod"
 export PINECONE_API_KEY="pinecone api key"
 export PINECONE_TEXT_KEY="pinecone field for text - default text in pod content"
 export PINECONE_INDEX="pinecone index name"

diff --git a/entrypoint.sh b/entrypoint.sh
@@ -53,5 +53,5 @@ fi
 
 echo "start gunicorn with $ENVIRON --workers $WORKERS --timeout $TIMEOUT --max-requests $MAXREQUESTS --max-requests-jitter $MAXRJITTER --graceful-timeout $GRACEFULTIMEOUT"
 
-gunicorn --bind 0.0.0.0:8000  --workers $WORKERS --timeout $TIMEOUT --max-requests $MAXREQUESTS --max-requests-jitter $MAXRJITTER --graceful-timeout $GRACEFULTIMEOUT --env ENVIRON="$environment" --log-config-json log_conf.json --worker-class uvicorn.workers.UvicornWorker tilellm.__main__:app
+gunicorn --bind 0.0.0.0:8000  --workers $WORKERS --timeout $TIMEOUT --max-requests $MAXREQUESTS --max-requests-jitter $MAXRJITTER --graceful-timeout $GRACEFULTIMEOUT --env PINECONE_TYPE="$environment" --log-config-json log_conf.json --worker-class uvicorn.workers.UvicornWorker tilellm.__main__:app
 
diff --git a/log_conf.json b/log_conf.json
@@ -25,11 +25,13 @@
       "stream": "ext://sys.stderr"
     }
   },
+
   "root": {
     "level": "INFO",
     "handlers": [
       "stdout","stderr"
-    ]
+    ],
+     "propagate": false
   },
   "formatters": {
     "simple": {

diff --git a/pyproject.toml b/pyproject.toml
@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "tilellm"
-version = "0.1.16"
+version = "0.1.17"
 description = "tiledesk for RAG"
 authors = ["Gianluca Lorenzo <gianluca.lorenzo@gmail.com>"]
 repository = "https://github.com/Tiledesk/tiledesk-llm"

diff --git a/tilellm/__main__.py b/tilellm/__main__.py
@@ -46,7 +46,7 @@
 # args = parser.parse_args()
 
 ENVIRONMENTS = {
-    'dev': '.environ',
+    'serverless': '.environ',
     'prod': '.environ.prod',
 }
 
@@ -55,7 +55,7 @@
 logger = logging.getLogger(__name__)
 
 
-environment = os.environ.get("ENVIRON", "dev")
+environment = os.environ.get("PINECONE_TYPE", "serverless")
 # environment = "prod"
 load_dotenv(ENVIRONMENTS.get(environment) or '.environ')
 
@@ -270,18 +270,18 @@ async def create_scrape_item_single(item: ItemSingle, redis_client: aioredis.cli
 
         logger.info(f"webhook: {webhook}, token: {token}")
 
-        if webhook:
-            res = PineconeIndexingResult(id=item.id, status=200)
-            try:
-                async with aiohttp.ClientSession() as session:
-                    res = await session.post(webhook,
-                                             json=res.model_dump(exclude_none=True),
-                                             headers={"Content-Type": "application/json",
-                                                      "X-Auth-Token": token})
-                    logger.info(f"200 {await res.json()}")
-            except Exception as ewh:
-                logger.error(ewh)
-                pass
+        # if webhook:
+        #    res = PineconeIndexingResult(id=item.id, status=200)
+        #    try:
+        #        async with aiohttp.ClientSession() as session:
+        #            res = await session.post(webhook,
+        #                                     json=res.model_dump(exclude_none=True),
+        #                                     headers={"Content-Type": "application/json",
+        #                                              "X-Auth-Token": token})
+        #            logger.info(f"200 {await res.json()}")
+        #    except Exception as ewh:
+        #        logger.error(ewh)
+        #        pass
 
         pc_result = await add_pc_item(item)
         # import datetime
@@ -299,20 +299,20 @@ async def create_scrape_item_single(item: ItemSingle, redis_client: aioredis.cli
                                          scrape_status_response.model_dump_json(),
                                          ex=expiration_in_seconds)
 
-        logger.debug(f"End {add_to_queue}")
-        if webhook:
-            try:
-                async with aiohttp.ClientSession() as session:
-                    res = await session.post(webhook,
-                                             json=pc_result.model_dump(exclude_none=True),
-                                             headers={"Content-Type": "application/json",
-                                                      "X-Auth-Token": token})
-                    logger.info(f"300 {await res.json()}")
-            except Exception as ewh:
-                logger.error(ewh)
-                pass
-
-        return JSONResponse(content={"message": f"Item {item.id} created successfully"})
+        # logger.debug(f"End {add_to_queue}")
+        # if webhook:
+        #    try:
+        #        async with aiohttp.ClientSession() as session:
+        #            res = await session.post(webhook,
+        #                                     json=pc_result.model_dump(exclude_none=True),
+        #                                     headers={"Content-Type": "application/json",
+        #                                              "X-Auth-Token": token})
+        #            logger.info(f"300 {await res.json()}")
+        #    except Exception as ewh:
+        #        logger.error(ewh)
+        #        pass
+
+        return JSONResponse(content=pc_result.model_dump(exclude_none=True)) # {"message": f"Item {item.id} created successfully"})
 
     except Exception as e:
         scrape_status_response = ScrapeStatusResponse(status_message="Error",
@@ -324,14 +324,14 @@ async def create_scrape_item_single(item: ItemSingle, redis_client: aioredis.cli
 
         logger.error(f"Error {add_to_queue}")
         import traceback
-        if webhook:
-            res = PineconeIndexingResult(id=item.id, status=400, error=repr(e))
-            async with aiohttp.ClientSession() as session:
-                response = await session.post(webhook, json=res.model_dump(exclude_none=True),
-                                              headers={"Content-Type": "application/json", "X-Auth-Token": token})
-                logger.error(response)
-                logger.error(f"{await response.json()}")
-            logger.error(f"Error {e}, webhook: {webhook}")
+        # if webhook:
+        #    res = PineconeIndexingResult(id=item.id, status=400, error=repr(e))
+        #    async with aiohttp.ClientSession() as session:
+        #        response = await session.post(webhook, json=res.model_dump(exclude_none=True),
+        #                                      headers={"Content-Type": "application/json", "X-Auth-Token": token})
+        #        logger.error(response)
+        #        logger.error(f"{await response.json()}")
+        #    logger.error(f"Error {e}, webhook: {webhook}")
         traceback.print_exc()
         logger.error(e)
         raise HTTPException(status_code=400, detail=repr(e))

diff --git a/tilellm/controller/openai_controller.py b/tilellm/controller/openai_controller.py
@@ -320,6 +320,7 @@ async def delete_id_from_namespace(metadata_id:str, namespace:str):
     try:
         return await delete_pc_ids_namespace(metadata_id=metadata_id, namespace=namespace)
     except Exception as ex:
+        logger.error(ex)
         raise ex
 
 

diff --git a/tilellm/store/pinecone_repository.py b/tilellm/store/pinecone_repository.py
@@ -33,7 +33,7 @@ async def add_pc_item(item):
     try:
         await delete_pc_ids_namespace(metadata_id=metadata_id, namespace=namespace)
     except Exception as ex:
-        logger.error(ex)
+        logger.warning(ex)
         pass
 
     emb_dimension = get_embeddings_dimension(embedding)
@@ -196,9 +196,7 @@ async def delete_pc_ids_namespace(metadata_id: str, namespace: str):
                 offset += len(ids)
 
     except Exception as ex:
-
-        logger.error(ex)
-
+        # logger.error(ex)
         raise ex
 
 
@@ -456,7 +454,7 @@ async def create_pc_index(embeddings, emb_dimension):
     else:
         logger.debug(f'Create index {const.PINECONE_INDEX} and embeddings ...')
 
-        if os.environ.get("ENVIRON") == "dev":
+        if os.environ.get("PINECONE_TYPE") == "serverless":
             pc.create_index(const.PINECONE_INDEX,
                             dimension=emb_dimension,
                             metric='cosine',
Original file line number	Diff line number	Diff line change
Expand Up		@@ -53,5 +53,5 @@ fi

		echo "start gunicorn with $ENVIRON --workers $WORKERS --timeout $TIMEOUT --max-requests $MAXREQUESTS --max-requests-jitter $MAXRJITTER --graceful-timeout $GRACEFULTIMEOUT"

		gunicorn --bind 0.0.0.0:8000 --workers $WORKERS --timeout $TIMEOUT --max-requests $MAXREQUESTS --max-requests-jitter $MAXRJITTER --graceful-timeout $GRACEFULTIMEOUT --env ENVIRON="$environment" --log-config-json log_conf.json --worker-class uvicorn.workers.UvicornWorker tilellm.__main__:app
		gunicorn --bind 0.0.0.0:8000 --workers $WORKERS --timeout $TIMEOUT --max-requests $MAXREQUESTS --max-requests-jitter $MAXRJITTER --graceful-timeout $GRACEFULTIMEOUT --env PINECONE_TYPE="$environment" --log-config-json log_conf.json --worker-class uvicorn.workers.UvicornWorker tilellm.__main__:app