From b767be90cde468a8cfbec58520c63869b504839e Mon Sep 17 00:00:00 2001 From: Akanksha Gupta Date: Thu, 23 Jul 2026 15:51:07 -0700 Subject: [PATCH] Set max_slice_restarts to 1000000 in Shared Pathways Service deployment This change configures the max_slice_restarts parameter to 1000000 when deploying the shared Pathways service. Using a very high backOffLimit for the worker pods keeps the jobset from failing on multiple client reconnections. PiperOrigin-RevId: 952999777 --- .../shared_pathways_service/deploy_pathways_service.py | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/pathwaysutils/experimental/shared_pathways_service/deploy_pathways_service.py b/pathwaysutils/experimental/shared_pathways_service/deploy_pathways_service.py index 72903ee..0776b48 100644 --- a/pathwaysutils/experimental/shared_pathways_service/deploy_pathways_service.py +++ b/pathwaysutils/experimental/shared_pathways_service/deploy_pathways_service.py @@ -4,7 +4,6 @@ import dataclasses import logging import math -import os from typing import Any from absl import app from absl import flags @@ -142,7 +141,6 @@ def calculate_vms_per_slice(topology: str, chips_per_vm: int) -> int: ) from e - def deploy_jobset(jobset_yaml: dict[str, Any]) -> None: """Deploys the JobSet to the current Kubernetes cluster.""" try: @@ -185,6 +183,9 @@ def run_deployment( topology=topology, num_slices=num_slices, shared_pathways_service=True, + # TODO(b/496958026): Remove this once go/sps-worker-pod-stability is + # implemented + max_slice_restarts=1000000, ) # If custom server_image is provided, mutate the templates to use it.