Update healthcheck configuration for GPU services to improve startup reliability

Author: chrisdengso

compose/docker-compose-multi-gpu.yml

@@ -22,10 +22,12 @@ services:
               capabilities: [gpu]
     healthcheck:
       test: [ "CMD", "curl", "-f", "http://localhost:18080/health" ]
-      interval: 300s
+      interval: 30s
       timeout: 10s
       retries: 3
-      start_period: 30s
+      # Must exceed model download + TRT engine warmup on first boot, otherwise
+      # the container is killed mid-startup and never reaches "healthy".
+      start_period: 240s
 
   ifr-trt-gpu1:
     image: insightface-rest:${IFR_VERSION}
@@ -48,10 +50,12 @@ services:
               capabilities: [gpu]
     healthcheck:
       test: [ "CMD", "curl", "-f", "http://localhost:18080/health" ]
-      interval: 300s
+      interval: 30s
       timeout: 10s
       retries: 3
-      start_period: 30s
+      # Must exceed model download + TRT engine warmup on first boot, otherwise
+      # the container is killed mid-startup and never reaches "healthy".
+      start_period: 240s
 
 
   nginx:

compose/docker-compose-v2.yml

@@ -9,10 +9,12 @@ x-base_service: &base_service
       - &v_nginx ../misc/nginx_conf/conf.d:/etc/nginx/conf.d
     healthcheck:
       test: [ "CMD", "curl", "-f", "http://localhost:18080/health" ]
-      interval: 300s
+      interval: 30s
       timeout: 10s
       retries: 3
-      start_period: 30s
+      # Must exceed model download + TRT engine warmup on first boot, otherwise
+      # the container is killed mid-startup and never reaches "healthy".
+      start_period: 240s
 
 services:
   ifr-trt-gpu0:

compose/docker-compose.yml

@@ -23,7 +23,9 @@ services:
               capabilities: [gpu]
     healthcheck:
       test: [ "CMD", "curl", "-f", "http://localhost:18080/health" ]
-      interval: 300s
+      interval: 30s
       timeout: 10s
       retries: 3
-      start_period: 30s
+      # Must exceed model download + TRT engine warmup on first boot, otherwise
+      # the container is killed mid-startup and never reaches "healthy".
+      start_period: 240s