From 2b20e1232b6542e780a61e1832e72cc2babc8989 Mon Sep 17 00:00:00 2001
From: Paul Buetow <paul@buetow.org>
Date: Sat, 1 Aug 2026 20:58:40 +0300
Subject: shuriken CronJob: bound memory + ImageMagick disk spill (OOM fix)

The first run OOM-killed (exit 137) on irregular.ninja: Alpine's
ImageMagick is Q16-HDRI (32 bytes/pixel), so a 37 MP AI-upscaled PNG
decodes to ~1.2 GiB per buffer and -auto-orient doubles it -- well
over the 1 Gi pod limit. The image now ships an ImageMagick policy
(memory/map 512 MiB, disk 8 GiB) that spills oversized decodes to a
disk cache instead of OOMing. Match it here: bump the pod memory
limit to 2 Gi for the working set, pin OMP_NUM_THREADS=1 (the build
is OpenMP-enabled; extra resample threads only multiply HDRI memory
pressure), and add a disk-backed /tmp emptyDir (sizeLimit 8 Gi) for
the pixel-cache spill.
---
 f3s/shuriken/helm-chart/templates/cronjob.yaml | 22 ++++++++++++++++++++--
 1 file changed, 20 insertions(+), 2 deletions(-)

diff --git a/f3s/shuriken/helm-chart/templates/cronjob.yaml b/f3s/shuriken/helm-chart/templates/cronjob.yaml
index ddb97b8..2641ce0 100644
--- a/f3s/shuriken/helm-chart/templates/cronjob.yaml
+++ b/f3s/shuriken/helm-chart/templates/cronjob.yaml
@@ -65,13 +65,24 @@ spec:
                 # to shuriken as --image-jobs.
                 - name: SHURIKEN_IMAGE_JOBS
                   value: "1"
+                # ImageMagick 7 is Q16-HDRI and built with OpenMP. Pin OpenMP
+                # to a single thread: the pod is cpu-limited to 1 core anyway,
+                # and extra resample threads would only multiply the HDRI pixel
+                # buffer's memory pressure (the OOM trigger for large PNGs).
+                - name: OMP_NUM_THREADS
+                  value: "1"
               resources:
                 requests:
                   cpu: 100m
                   memory: 256Mi
                 limits:
                   cpu: "1"
-                  memory: 1Gi
+                  # 2 Gi: the ImageMagick policy (baked into the image) caps
+                  # the in-RAM pixel cache at 512 MiB and spills oversized
+                  # decodes to the /tmp disk cache, so this headroom covers the
+                  # working set + bash + the HDRI buffers of one photo without
+                  # the pod being OOM-killed on a multi-megapixel source image.
+                  memory: 2Gi
               securityContext:
                 allowPrivilegeEscalation: false
                 runAsUser: 0
@@ -82,10 +93,17 @@ spec:
                 - name: configs
                   mountPath: /configs
                   readOnly: true
+                - name: tmp
+                  mountPath: /tmp
           volumes:
             - name: data
               persistentVolumeClaim:
                 claimName: shuriken-data-pvc
             - name: configs
               configMap:
-                name: shuriken-config
\ No newline at end of file
+                name: shuriken-config
+            # Disk-backed scratch for ImageMagick's disk pixel cache (the
+            # policy spills oversized HDRI decodes here instead of OOMing).
+            - name: tmp
+              emptyDir:
+                sizeLimit: 8Gi
\ No newline at end of file
-- 
cgit v1.2.3