Merge pull request #2631 from lsst-sqre/tickets/DM-41090

DM-41090: Set Gafaelfawr CronJob time limits, tweak resource limits
lsst-sqre · Oct 6, 2023 · 2ffa35c · 2ffa35c
2 parents 929f865 + 6dd27ba
commit 2ffa35c
Show file tree

Hide file tree

Showing 4 changed files with 15 additions and 1 deletion.
diff --git a/applications/gafaelfawr/README.md b/applications/gafaelfawr/README.md
@@ -83,6 +83,8 @@ Authentication and identity system
 | ingress.additionalHosts | list | `[]` | Defines additional FQDNs for Gafaelfawr.  This doesn't work for cookie or browser authentication, but for token-based services like git-lfs or the webdav server it does. |
 | maintenance.affinity | object | `{}` | Affinity rules for Gafaelfawr maintenance and audit pods |
 | maintenance.auditSchedule | string | `"30 3 * * *"` | Cron schedule string for Gafaelfawr data consistency audit (in UTC) |
+| maintenance.cleanupSeconds | int | 86400 (1 day) | How long to keep old jobs around before deleting them |
+| maintenance.deadlineSeconds | int | 300 (5 minutes) | How long the job is allowed to run before it will be terminated |
 | maintenance.maintenanceSchedule | string | `"5 * * * *"` | Cron schedule string for Gafaelfawr periodic maintenance (in UTC) |
 | maintenance.nodeSelector | object | `{}` | Node selection rules for Gafaelfawr maintenance and audit pods |
 | maintenance.podAnnotations | object | `{}` | Annotations for Gafaelfawr maintenance and audit pods |

diff --git a/applications/gafaelfawr/templates/cronjob-audit.yaml b/applications/gafaelfawr/templates/cronjob-audit.yaml
@@ -10,6 +10,8 @@ spec:
   concurrencyPolicy: "Forbid"
   jobTemplate:
     spec:
+      activeDeadlineSeconds: {{ .Values.maintenance.deadlineSeconds }}
+      ttlSecondsAfterFinished: {{ .Values.maintenance.cleanupSeconds }}
       template:
         metadata:
           {{- with .Values.maintenance.podAnnotations }}

diff --git a/applications/gafaelfawr/templates/cronjob-maintenance.yaml b/applications/gafaelfawr/templates/cronjob-maintenance.yaml
@@ -9,6 +9,8 @@ spec:
   concurrencyPolicy: "Forbid"
   jobTemplate:
     spec:
+      activeDeadlineSeconds: {{ .Values.maintenance.deadlineSeconds }}
+      ttlSecondsAfterFinished: {{ .Values.maintenance.cleanupSeconds }}
       template:
         metadata:
           {{- with .Values.maintenance.podAnnotations }}

diff --git a/applications/gafaelfawr/values.yaml b/applications/gafaelfawr/values.yaml
@@ -347,11 +347,19 @@ maintenance:
   # -- Cron schedule string for Gafaelfawr periodic maintenance (in UTC)
   maintenanceSchedule: "5 * * * *"
 
+  # -- How long the job is allowed to run before it will be terminated
+  # @default -- 300 (5 minutes)
+  deadlineSeconds: 300
+
+  # -- How long to keep old jobs around before deleting them
+  # @default -- 86400 (1 day)
+  cleanupSeconds: 86400
+
   # -- Resource limits and requests for Gafaelfawr maintenance and audit pods
   # @default -- See `values.yaml`
   resources:
     limits:
-      cpu: "100m"
+      cpu: "1"
       memory: "300Mi"
     requests:
       cpu: "100m"