fix: ensure script fails on error by setting 'set -e'

Dockerfile

@@ -33,7 +33,7 @@ WORKDIR /work
 ADD ./src /work
 
 # Install runpod and its dependencies
-RUN pip install -r requirements.txt && chmod +x /work/start.sh
+RUN pip install -r ./requirements.txt && chmod +x /work/start.sh
 
 # Set the entrypoint
 ENTRYPOINT ["/bin/sh", "-c", "/work/start.sh"]

src/start.sh

@@ -1,5 +1,8 @@
 #!/bin/bash
 
+# fail on error:
+set -e
+
 # This script starts the llama-server with the command line arguments
 # specified in the environment variable LLAMA_SERVER_CMD_ARGS, ensuring
 # that the server listens on port 3098. It also starts the handler.py
@@ -16,8 +19,8 @@ if [[ "$LLAMA_SERVER_CMD_ARGS" != *"/workspace"* ]]; then
     echo "Tip: For reduced downloads and faster startup times, consider using a model stored in a network volume mounted to /workspace."
 fi
 
-# check if the substring -port is in LLAMA_SERVER_CMD_ARGS
-if [[ "$LLAMA_SERVER_CMD_ARGS" != *"-port"* ]]; then
+# check if the substring -port is in LLAMA_SERVER_CMD_ARGS and if yes, raise an error:
+if [[ "$LLAMA_SERVER_CMD_ARGS" == *"-port"* ]]; then
     echo "Error: You must not define -port in LLAMA_SERVER_CMD_ARGS, as port 3098 is required."
     exit 1
 fi
@@ -26,13 +29,13 @@ fi
 trap cleanup SIGINT SIGTERM
 
 # kill any existing llama-server processes
-pgrep llama-server | xargs kill
+pkill llama-server
 
 # we have a string with all the command line arguments in the env var LLAMA_SERVER_CMD_ARGS;
 # it contains a.e. "-hf modelname -ctx_size 4096".
 
 # We need to pass these arguments to llama-server verbatim.
-llama-server $LLAMA_SERVER_CMD_ARGS -port 3098 2>&1 | tee llama.server.log &
+/app/llama-server $LLAMA_SERVER_CMD_ARGS -port 3098 2>&1 | tee llama.server.log &
 
 LLAMA_SERVER_PID=$! # store the process ID (PID) of the background command