From 5428b3bf6a463cf0537568b8aaad3aea02bd8e24 Mon Sep 17 00:00:00 2001
From: Sujeeth Jinesh <sujinesh@google.com>
Date: Fri, 19 Apr 2024 00:56:26 +0000
Subject: [PATCH] Prevent Pathways SIGTERMs from counting against backoffLimit

---
 xpk.py | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/xpk.py b/xpk.py
index 7e98e3b8..d095d92d 100644
--- a/xpk.py
+++ b/xpk.py
@@ -265,10 +265,20 @@ class CapacityType(enum.Enum):
           xpk.google.com/workload: {args.workload}
       spec:
         backoffLimit: 4
+        podFailurePolicy:
+          rules:
+          - action: Ignore          # one of: Ignore, FailJob, Count
+            onExitCodes:
+              containerName: "pathways-worker"
+              operator: In          # one of: In, NotIn
+              # TODO: verify this is the correct exit code
+              values: [143]         # Don't count SIGTERMed workers against the backoffLimit
+            onPodConditions: []
         completions: {system.vms_per_slice}
         parallelism: {system.vms_per_slice}
         template:
           spec:
+            restartPolicy: Never
             terminationGracePeriodSeconds: {args.termination_grace_period_seconds}
             containers:
             - args: