feat: Add temporal client rpc timeout

deployments/aws/fargate/locals.tf

@@ -26,31 +26,33 @@ locals {
 
   api_env = [
     for k, v in merge({
-      LOG_LEVEL                   = var.log_level
-      TRACECAT__API_URL           = local.internal_api_url
-      TRACECAT__API_ROOT_PATH     = "/api"
-      TRACECAT__APP_ENV           = var.tracecat_app_env
-      TRACECAT__PUBLIC_RUNNER_URL = local.public_api_url
-      TRACECAT__PUBLIC_APP_URL    = local.public_app_url
-      TRACECAT__ALLOW_ORIGINS     = local.allow_origins
-      TRACECAT__AUTH_TYPES        = var.auth_types
-      TEMPORAL__CLUSTER_URL       = local.temporal_cluster_url
-      TEMPORAL__CLUSTER_QUEUE     = local.temporal_cluster_queue
-      SAML_SP_ACS_URL             = local.saml_acs_url
-      RUN_MIGRATIONS              = "true"
+      LOG_LEVEL                    = var.log_level
+      TRACECAT__API_URL            = local.internal_api_url
+      TRACECAT__API_ROOT_PATH      = "/api"
+      TRACECAT__APP_ENV            = var.tracecat_app_env
+      TRACECAT__PUBLIC_RUNNER_URL  = local.public_api_url
+      TRACECAT__PUBLIC_APP_URL     = local.public_app_url
+      TRACECAT__ALLOW_ORIGINS      = local.allow_origins
+      TRACECAT__AUTH_TYPES         = var.auth_types
+      TEMPORAL__CLUSTER_URL        = local.temporal_cluster_url
+      TEMPORAL__CLUSTER_QUEUE      = local.temporal_cluster_queue
+      TEMPORAL__CLIENT_RPC_TIMEOUT = var.temporal_client_rpc_timeout
+      SAML_SP_ACS_URL              = local.saml_acs_url
+      RUN_MIGRATIONS               = "true"
     }, local.tracecat_db_configs) :
     { name = k, value = tostring(v) }
   ]
 
   worker_env = [
     for k, v in merge({
-      LOG_LEVEL                   = var.log_level
-      TRACECAT__API_URL           = local.internal_api_url
-      TRACECAT__API_ROOT_PATH     = "/api"
-      TRACECAT__APP_ENV           = var.tracecat_app_env
-      TRACECAT__PUBLIC_RUNNER_URL = local.public_api_url
-      TEMPORAL__CLUSTER_URL       = local.temporal_cluster_url
-      TEMPORAL__CLUSTER_QUEUE     = local.temporal_cluster_queue
+      LOG_LEVEL                    = var.log_level
+      TRACECAT__API_URL            = local.internal_api_url
+      TRACECAT__API_ROOT_PATH      = "/api"
+      TRACECAT__APP_ENV            = var.tracecat_app_env
+      TRACECAT__PUBLIC_RUNNER_URL  = local.public_api_url
+      TEMPORAL__CLUSTER_URL        = local.temporal_cluster_url
+      TEMPORAL__CLUSTER_QUEUE      = local.temporal_cluster_queue
+      TEMPORAL__CLIENT_RPC_TIMEOUT = var.temporal_client_rpc_timeout
     }, local.tracecat_db_configs) :
     { name = k, value = tostring(v) }
   ]

deployments/aws/fargate/variables.tf

@@ -177,6 +177,11 @@ variable "temporal_memory" {
   default = "512"
 }
 
+variable "temporal_client_rpc_timeout" {
+  type        = string
+  description = "RPC timeout for Temporal client in seconds"
+}
+
 variable "caddy_cpu" {
   type    = string
   default = "256"

tracecat/config.py

@@ -123,6 +123,8 @@
 TEMPORAL__TLS_ENABLED = os.environ.get("TEMPORAL__TLS_ENABLED", False)
 TEMPORAL__TLS_CLIENT_CERT = os.environ.get("TEMPORAL__TLS_CLIENT_CERT")
 TEMPORAL__TLS_CLIENT_PRIVATE_KEY = os.environ.get("TEMPORAL__TLS_CLIENT_PRIVATE_KEY")
+TEMPORAL__CLIENT_RPC_TIMEOUT = os.environ.get("TEMPORAL__CLIENT_RPC_TIMEOUT")
+"""RPC timeout for Temporal workflows in seconds."""
 
 # Secrets manager config
 TRACECAT__UNSAFE_DISABLE_SM_MASKING = os.environ.get(

tracecat/workflow/executions/service.py

@@ -378,6 +378,8 @@ async def _dispatch_workflow(
             role=self.role,
             wf_exec_id=wf_exec_id,
         )
+        if rpc_timeout := config.TEMPORAL__CLIENT_RPC_TIMEOUT:
+            kwargs["rpc_timeout"] = datetime.timedelta(seconds=float(rpc_timeout))
         try:
             result = await self._client.execute_workflow(
                 DSLWorkflow.run,