Merge pull request #1053 from anarkiwi/hc99

add healthcheck config for torchserve.

orchestrator.yml

@@ -43,12 +43,17 @@ services:
     # nvidia container toolkit etc must be installed.
     # Comment out "deploy" if on non-x86 platform (e.g. Pi4)
     deploy:
-       resources:
-         reservations:
-            devices:
-              - driver: nvidia
-                count: 1
-                capabilities: [gpu]
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: 1
+              capabilities: [gpu]
+    # if torchserve is not being used, comment out the
+    # dependency.
+    depends_on:
+      torchserve:
+        condition: service_healthy
     volumes:
       - '${VOL_PREFIX}:/logs'
     command:

torchserve-cuda.yml

@@ -5,14 +5,16 @@ services:
   torchserve:
     restart: always
     image: iqtlabs/cuda-torchserve:v0.0.6
+    healthcheck:
+      start_period: 120s
     networks:
       - gamutrf
     ports:
       - '8080:8080'
       - '8081:8081'
     volumes:
       - '${VOL_PREFIX}/model_store:/model_store'
-      - '/root/.cache:/root/.cache'
+      - '/root/.cache/pip:/root/.cache/pip'
     deploy:
       resources:
         reservations:

torchserve.yml

@@ -5,13 +5,15 @@ services:
   torchserve:
     restart: always
     image: iqtlabs/torchserve:v0.0.6
+    healthcheck:
+      start_period: 120s
     networks:
       - gamutrf
     ports:
       - '8080:8080'
       - '8081:8081'
     volumes:
       - '${VOL_PREFIX}/model_store:/model_store'
-      - '/root/.cache:/root/.cache'
+      - '/root/.cache/pip:/root/.cache/pip'
     command:
       - --models mini2_snr=mini2_snr.mar