From 60b7416df42c4f7ec6305894e94632afdac61f70 Mon Sep 17 00:00:00 2001
From: jgrusewski <jgrusewski@localhost>
Date: Sun, 24 May 2026 23:22:08 +0200
Subject: [PATCH] fix(argo): bump fxcache memory to 96Gi (OOM at 56Gi with 28
 chunks)

Alpha feature pipeline runs 28 parallel chunks across 17.8M imbalance
bars, peaking at ~70GB RSS. Previous 56Gi limit caused OOMKill.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
---
 infra/k8s/argo/train-multi-seed-template.yaml | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/infra/k8s/argo/train-multi-seed-template.yaml b/infra/k8s/argo/train-multi-seed-template.yaml
index 8ee36257c..eee064b67 100644
--- a/infra/k8s/argo/train-multi-seed-template.yaml
+++ b/infra/k8s/argo/train-multi-seed-template.yaml
@@ -353,10 +353,10 @@ spec:
         resources:
           requests:
             cpu: "4"
-            memory: 16Gi
+            memory: 32Gi
           limits:
             cpu: "28"
-            memory: 56Gi
+            memory: 96Gi
         volumeMounts:
           - name: training-data
             mountPath: /data