refactor(scripts): enhance error handling and logging for robustness (#…

…21) Improve error handling by using 'set -e -o pipefail' to ensure scripts exit on errors. Refactor error handling functions for clarity and consistency. Enhance logging by redirecting error messages to the appropriate log files. Simplify main function execution and ensure only JSON output is sent to stdout. Remove hardcoded environment name in Azure template for flexibility.
truefoundry · Dec 26, 2024 · c1ec897 · c1ec897
1 parent 58cd52c
commit c1ec897
Show file tree

Hide file tree

Showing 3 changed files with 54 additions and 58 deletions.
diff --git a/scripts/get_environment.sh b/scripts/get_environment.sh
@@ -1,5 +1,5 @@
 #!/bin/bash
-
+set -e -o pipefail
 # Read input from stdin (Terraform external data source)
 eval "$(jq -r '@sh "
 CONTROL_PLANE_URL=\(.control_plane_url)

diff --git a/scripts/setup_truefoundry_cluster.sh b/scripts/setup_truefoundry_cluster.sh
@@ -1,5 +1,5 @@
 #!/bin/bash
-set -e
+set -e -o pipefail
 
 # Read input from stdin (Terraform external data source)
 eval "$(jq -r '@sh "
@@ -13,28 +13,27 @@ TRUEFOUNDRY_STDOUT_FILE=\(.stdout_log_file)
 TRUEFOUNDRY_STDERR_FILE=\(.stderr_log_file)
 "')"
 
-
-# Validate required parameters
-[ -z "${CONTROL_PLANE_URL}" ] && handle_error "CONTROL_PLANE_URL is required" && return 1
-[ -z "${API_KEY}" ] && handle_error "API_KEY is required" && return 1
-[ -z "${CLUSTER_NAME}" ] && handle_error "CLUSTER_NAME is required" && return 1
-[ -z "${CLUSTER_TYPE}" ] && handle_error "CLUSTER_TYPE is required" && return 1
+# Error handling
+function handle_error() {
+    local error_message="$1"
+    echo "[ERROR] $error_message" >&2
+    exit 1
+}
 
 # Logging functions
 function log_info() {
-    echo "[INFO] $1" >> $TRUEFOUNDRY_STDOUT_FILE
+    echo "[INFO] $1" >> "$TRUEFOUNDRY_STDOUT_FILE"
 }
 
 function log_error() {
-    echo "[ERROR] $1" >> $TRUEFOUNDRY_STDERR_FILE
+    echo "[ERROR] $1" >> "$TRUEFOUNDRY_STDERR_FILE"
 }
 
-# Error handling
-function handle_error() {
-    log_error "$1"
-    jq -n --arg error "$1" '{"error": $error}'
-    exit 1
-}
+# Validate required parameters
+[ -z "${CONTROL_PLANE_URL}" ] && handle_error "CONTROL_PLANE_URL is required"
+[ -z "${API_KEY}" ] && handle_error "API_KEY is required"
+[ -z "${CLUSTER_NAME}" ] && handle_error "CLUSTER_NAME is required"
+[ -z "${CLUSTER_TYPE}" ] && handle_error "CLUSTER_TYPE is required"
 
 # HTTP request handler
 function make_request() {
@@ -44,7 +43,6 @@ function make_request() {
     local expected_status_codes="$4"
     local response_file=$(mktemp)
     local http_code_file=$(mktemp)
-
     log_info "make_request: Making ${method} request to ${url}"
 
     local curl_cmd="curl -s -X ${method} \
@@ -54,129 +52,127 @@ function make_request() {
 
     [ -n "$body" ] && curl_cmd="$curl_cmd -d '${body}'"
 
-    # Execute request and capture response, redirect all output to stderr
-    eval "${curl_cmd} '${url}'" > "${response_file}" 2>&2
+    # Execute request and capture response
+    eval "${curl_cmd} '${url}'" > "${response_file}" 2>>"$TRUEFOUNDRY_STDERR_FILE"
     local curl_exit_code=$?
 
-    # Get HTTP status code, redirect all output to stderr
-    eval "${curl_cmd} -o /dev/null -w '%{http_code}' '${url}'" > "${http_code_file}" 2>&2
+    # Get HTTP status code
+    eval "${curl_cmd} -o /dev/null -w '%{http_code}' '${url}'" > "${http_code_file}" 2>>"$TRUEFOUNDRY_STDERR_FILE"
     local http_code=$(<"${http_code_file}")
 
     # Cleanup
     rm -f "${http_code_file}"
 
     # Validate response
     if [ $curl_exit_code -ne 0 ]; then
+        local error_msg="Curl command failed with exit code: ${curl_exit_code}"
         rm -f "${response_file}"
+        handle_error "$error_msg"
         return 1
     fi
 
     if [[ ! ",$expected_status_codes," =~ ,"${http_code}", ]]; then
-        log_error "Request failed with status code: ${http_code}"
+        local error_msg="Request failed with status code: ${http_code}"
+        log_error "$error_msg"
         log_error "Response body:"
-        cat "${response_file}" >&2
+        cat "${response_file}" >> "$TRUEFOUNDRY_STDERR_FILE"
         rm -f "${response_file}"
+        handle_error "$error_msg"
         return 1
     fi
 
-    # Output response to stdout but only within a function call
+    # Output response
     cat "${response_file}"
     rm -f "${response_file}"
     return 0
 }
 
-
 function create_cluster() {
     log_info "create_cluster: Creating cluster..."
 
-    [-z "$CLUSTER_CONFIG_BASE64"] && handle_error "$CLUSTER_CONFIG_BASE64 is required" && return 1
+    [ -z "$CLUSTER_CONFIG_BASE64" ] && handle_error "CLUSTER_CONFIG_BASE64 is required"
 
     local manifest=$(echo "$CLUSTER_CONFIG_BASE64" | base64 -d)
+    [ $? -ne 0 ] && handle_error "Failed to decode CLUSTER_CONFIG_BASE64"
 
     log_info "create_cluster: cluster manifest ${manifest}"
-    local response=$(make_request "PUT" "${CONTROL_PLANE_URL}/api/svc/v1/cluster/" "${manifest}" "200,201") || \
-        handle_error "create_cluster: Failed to create cluster"
+    local response=$(make_request "PUT" "${CONTROL_PLANE_URL}/api/svc/v1/cluster/" "${manifest}" "200,201")
+    [ $? -ne 0 ] && handle_error "Failed to create cluster"
 
     log_info "create_cluster: Response $response"
 
-    local cluster_id=$(echo "${response}" | jq -r '.id') || \
-        handle_error "create_cluster: Failed to parse cluster response"
+    local cluster_id=$(echo "${response}" | jq -r '.id')
+    [ $? -ne 0 ] && handle_error "Failed to parse cluster response"
 
     log_info "create_cluster: Cluster ID $cluster_id"
-
-    [ -z "${cluster_id}" ] && handle_error "create_cluster: No cluster ID found in response" && return 1
+    [ -z "${cluster_id}" ] && handle_error "No cluster ID found in response"
 
     echo "${cluster_id}"
 }
 
 function get_cluster_token() {
     local cluster_id="$1"
+    [ -z "$cluster_id" ] && handle_error "Cluster ID is required for token generation"
     log_info "get_cluster_token: Getting cluster token..."
 
     local response=$(make_request "GET" \
         "${CONTROL_PLANE_URL}/api/svc/v1/cluster/${cluster_id}/token" \
-        "" "200") || handle_error "Failed to get cluster token"
+        "" "200")
+    [ $? -ne 0 ] && handle_error "Failed to get cluster token"
 
-    local token=$(echo "${response}" | jq -r '.clusterToken') || \
-        handle_error "Failed to parse cluster token response"
-
-    [ -z "${token}" ] && handle_error "No cluster token found in response" && return 1
+    local token=$(echo "${response}" | jq -r '.clusterToken')
+    [ $? -ne 0 ] && handle_error "Failed to parse cluster token response"
+    [ -z "${token}" ] && handle_error "No cluster token found in response"
 
     echo "${token}"
 }
 
-# Provider operations
 function setup_provider_account() {
+    [ -z "$PROVIDER_CONFIG_BASE64" ] && handle_error "PROVIDER_CONFIG_BASE64 is required"
 
-    [ -z "$PROVIDER_CONFIG_BASE64" ] && handle_error "$CLUSTER_CONFIG_BASE64 is required" && return 1
 
     log_info "setup_provider_account: Creating provider account..."
     local provider_manifest=$(echo "$PROVIDER_CONFIG_BASE64" | base64 -d)
+    [ $? -ne 0 ] && handle_error "Failed to decode PROVIDER_CONFIG_BASE64"
 
     log_info "setup_provider_account: provider account manifest ${provider_manifest}"
     local response=$(make_request "PUT" \
         "${CONTROL_PLANE_URL}/api/svc/v1/provider-accounts/" \
         "${provider_manifest}" \
-        "200,201") || handle_error "setup_provider_account: Failed to create provider account"
+        "200,201")
+    [ $? -ne 0 ] && handle_error "Failed to create provider account"
 
     log_info "setup_provider_account: Provider account created successfully"
 }
 
-# Main execution
 function main() {
-    # Capture all output in variables to prevent unwanted stdout
-    local environment_name
-    local cluster_manifest
-    local cluster_id
-    local cluster_token
-    local tenant_name
-
     # Verify platform health
     log_info "main: Checking platform health..."
-    make_request "GET" "${CONTROL_PLANE_URL}/api/svc/" "" "200" >/dev/null || \
-        handle_error "Platform health check failed"
+    make_request "GET" "${CONTROL_PLANE_URL}/api/svc/" "" "200" >/dev/null || handle_error "Platform health check failed"
 
     # Setup provider account if configured and cluster type is not generic
     if [ "${CLUSTER_TYPE}" != "generic" ]; then
         setup_provider_account
     fi
 
-    cluster_id=$(create_cluster)
-    cluster_token=$(get_cluster_token "${cluster_id}")
+    local cluster_id
+    cluster_id=$(create_cluster) || handle_error "Failed to create cluster"
+
+    local cluster_token
+    cluster_token=$(get_cluster_token "${cluster_id}") || handle_error "Failed to get cluster token"
 
     log_info "main: cluster_id=$cluster_id"
     log_info "main: cluster_token=$cluster_token"
 
-    # Output only the final JSON result to stdout
+    # Output the final JSON result to stdout
     jq -n \
         --arg cluster_id "$cluster_id" \
         --arg cluster_token "$cluster_token" \
         '{
             cluster_id: $cluster_id,
-            cluster_token: $cluster_token,
+            cluster_token: $cluster_token
         }'
 }
 
-# Execute main function and ensure only JSON output goes to stdout
-output=$(main)
-echo "$output"
+# Execute main function directly
+main
diff --git a/templates/cluster/azure.json.tpl b/templates/cluster/azure.json.tpl
@@ -13,7 +13,7 @@
         }
     ],
     "cluster_type": "${cluster_type}",
-    "environment_names": ["${env_name}", "prod"]
+    "environment_names": ["${env_name}"]
     %{ if container_registry_enabled },
     "default_registry_fqn": "${tenant_name}:${account_type}:${cluster_name}:docker-registry:registry"
     %{ endif }