From b01757b9e325fde604ddddc6405568f6aa728b0c Mon Sep 17 00:00:00 2001
From: Graham McIntire <graham@mcintire.me>
Date: Sat, 1 Aug 2026 12:24:05 -0500
Subject: [PATCH] fix: revert prop-grid-rs memory limit to 3Gi after OOM at
 1.5Gi

---
 k8s/deployment-grid-rs.yaml | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/k8s/deployment-grid-rs.yaml b/k8s/deployment-grid-rs.yaml
index fb3b3d15..b04cfb47 100644
--- a/k8s/deployment-grid-rs.yaml
+++ b/k8s/deployment-grid-rs.yaml
@@ -139,11 +139,12 @@ spec:
               # After the dense-grid rewrite (commit 63f25a96) the
               # per-task grid footprint dropped from ~200-400 MB of
               # nested HashMaps to ~18 MB of dense f32 planes.
-              # PARALLELISM went 1 → 3 in the same commit. The remaining
-              # large term is the f00 native-level decode (~530 MB GRIB2
-              # + wgrib2 working set). 1.5Gi leaves headroom for three
-              # concurrent chain steps plus one analysis step.
-              memory: 1.5Gi
+              # PARALLELISM went 1 → 3 in the same commit. 1.5Gi was
+              # attempted (commit db57b3a1) but the pod OOMed within
+              # ~20s at that limit — the f00 native-level decode still
+              # carries a ~530 MB GRIB2 + wgrib2 working set. Hold at
+              # 3Gi until RSS is profiled at steady state.
+              memory: 3Gi
           securityContext:
             allowPrivilegeEscalation: false
             runAsNonRoot: true