Merge pull request #1052 from anarkiwi/hc

test with torchserve healthcheck.

.github/workflows/docker-test.yml

@@ -7,7 +7,7 @@ jobs:
     - uses: actions/checkout@v4
     - name: docker build
       run: |
-        ./tests/test_torchserve.sh
+        timeout 600s ./tests/test_torchserve.sh
   test-gamutrf-extra-images:
     runs-on: ubuntu-latest
     steps:

tests/test_torchserve.sh

@@ -16,7 +16,15 @@ torch-model-archiver --force --model-name yolov8n --version 1.0 --serialized-fil
 rm -rf model_store && mkdir model_store
 mv yolov8n.mar model_store/
 # TODO: --runtime nvidia is required for Orin, --gpus all for x86
-docker run -v $(pwd)/model_store:/model_store -p 8080:8080 --rm --name testts --entrypoint timeout -d iqtlabs/torchserve:v0.0.5 180s /torchserve/torchserve-entrypoint.sh --models yolov8n=yolov8n.mar
+docker run -v $(pwd)/model_store:/model_store -p 8080:8080 --rm --name testts --entrypoint timeout -d iqtlabs/torchserve:v0.0.6 600s /torchserve/torchserve-entrypoint.sh --models yolov8n=yolov8n.mar
+echo waiting for healthcheck
+HC=""
+while [ "$HC" != "healthy" ] ; do
+  sleep 1
+  HC=$(docker inspect --format='{{json .State.Health}}' testts |jq -r .Status)
+  echo $HC
+  docker logs testts
+done
 PRED=$(wget -q --retry-connrefused --retry-on-host-error --body-file=persons.jpg --method=PUT -O- --header='Content-Type: image/jpg' http://127.0.0.1:8080/predictions/yolov8n | jq)
 echo $PRED
 if [ "$PRED" = "" ] ; then

torchserve-cuda.yml

@@ -4,7 +4,7 @@ networks:
 services:
   torchserve:
     restart: always
-    image: iqtlabs/cuda-torchserve:v0.0.5
+    image: iqtlabs/cuda-torchserve:v0.0.6
     networks:
       - gamutrf
     ports:

torchserve.yml

@@ -4,7 +4,7 @@ networks:
 services:
   torchserve:
     restart: always
-    image: iqtlabs/torchserve:v0.0.5
+    image: iqtlabs/torchserve:v0.0.6
     networks:
       - gamutrf
     ports: