feat: llm container wait for indexer healthcheck

Author: pshenok

docker-compose-ollama.yml

@@ -46,14 +46,19 @@ services:
       - ${LOCAL_FILES_PATH}:/usr/src/app/local_files/
       - ./indexer:/usr/src/app
     ports:
-      - 8001:8000
+      - 8001:8001
     environment:
       - PYTHONPATH=/usr/src
       - PYTHONUNBUFFERED=TRUE
       - LOCAL_FILES_PATH=${LOCAL_FILES_PATH}
       - EMBEDDING_MODEL_ID=${EMBEDDING_MODEL_ID}
       - EMBEDDING_SIZE=${EMBEDDING_SIZE}
       - START_INDEXING=${START_INDEXING}
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8001/health"]
+      interval: 10s
+      timeout: 5s
+      retries: 3
     depends_on:
       - qdrant
 
@@ -66,7 +71,7 @@ services:
     volumes:
       - ./llm:/usr/src/app
     ports:
-      - 8003:8000
+      - 8003:8003
     environment:
       - PYTHONPATH=/usr/src
       - PYTHONUNBUFFERED=TRUE
@@ -76,6 +81,7 @@ services:
       - ollama
       - qdrant
       - indexer
+    entrypoint: ["./wait-for-it.sh", "http://indexer:8001/health", "60", "--", "python", "app.py"]
 
   chat:
     build: ./chat

indexer/Dockerfile

@@ -16,7 +16,7 @@ COPY . .
 ENV START_INDEXING=${START_INDEXING}
 RUN echo "START_INDEXING is $START_INDEXING"
 
-ENV PORT 8000
+ENV PORT 8001
 ENV CURRENT_HOST 0.0.0.0
 ENV WORKERS 1
 

indexer/app.py

@@ -5,6 +5,7 @@
 from pydantic import BaseModel
 from async_queue import AsyncQueue
 from fastapi import FastAPI, APIRouter
+from fastapi.responses import JSONResponse
 from contextlib import asynccontextmanager
 from async_loop import index_loop, crawl_loop
 
@@ -20,6 +21,34 @@
 class Query(BaseModel):
     query: str
 
+@router.get(
+    "/health",
+    response_description="Health check",
+    status_code=200,
+)
+async def health():
+    try:
+        result = indexer.find("Indexer healthy")
+        if not result:
+            return JSONResponse(
+                status_code=500,
+                content={"status": "error", "detail": "Indexer is not healthy"},
+            )
+        
+        return JSONResponse(
+            status_code=200,
+            content={"status": "ok", "detail": "Indexer is healthy"},
+        )
+    
+    except Exception as e:
+        logger.error(f"Error during health check: {e}")
+        
+        return JSONResponse(
+            status_code=500,
+            content={"status": "error", "detail": str(e)},
+        )
+    
+
 @router.post(
     "/query", 
     response_description='Query local data storage',

llm/Dockerfile

@@ -10,6 +10,9 @@ RUN pip install huggingface_hub
 RUN huggingface-cli download $RERANKER_MODEL --repo-type model
 RUN pip install --no-cache-dir -r requirements.txt
 COPY . .
+RUN apt-get update && apt-get install -y curl
+COPY ./wait-for-it.sh /usr/local/bin/wait-for-it.sh
+RUN chmod +x /usr/local/bin/wait-for-it.sh
 
 ENV PORT 8000
 ENV CURRENT_HOST 0.0.0.0

llm/wait-for-it.sh

@@ -0,0 +1,26 @@
+#!/bin/bash
+
+set -e
+
+HOST=$1
+TIMEOUT=$2
+
+if [ -z "$HOST" ] || [ -z "$TIMEOUT" ]; then
+  echo "Usage: $0 <host:port/path> <timeout>"
+  exit 1
+fi
+
+echo "Waiting for $HOST to return HTTP 200..."
+
+for ((i=1;i<=TIMEOUT;i++)); do
+  STATUS=$(curl -o /dev/null -s -w "%{http_code}" "$HOST" || echo "000")
+  if [ "$STATUS" -eq 200 ]; then
+    echo "Service $HOST is ready."
+    exit 0
+  fi
+  echo "Service $HOST not ready, retrying ($i/$TIMEOUT)..."
+  sleep 1
+done
+
+echo "Timeout reached while waiting for $HOST"
+exit 1