From 4a86012eadbc2b7da932fcdf3f02b06269c35d2e Mon Sep 17 00:00:00 2001 From: Travis Nielsen Date: Wed, 5 Oct 2022 15:36:02 -0600 Subject: [PATCH] osd: recommend removing memory limits from osd prepare job The osd prepare job is not recommended to have memory limits since it's a one-time operation depending on the size of the disk that must be allowed to complete successfully. Signed-off-by: Travis Nielsen --- Documentation/CRDs/Cluster/ceph-cluster-crd.md | 11 +++++------ deploy/charts/rook-ceph-cluster/values.yaml | 5 ++--- 2 files changed, 7 insertions(+), 9 deletions(-) diff --git a/Documentation/CRDs/Cluster/ceph-cluster-crd.md b/Documentation/CRDs/Cluster/ceph-cluster-crd.md index b11e679fc..de87a7fa0 100755 --- a/Documentation/CRDs/Cluster/ceph-cluster-crd.md +++ b/Documentation/CRDs/Cluster/ceph-cluster-crd.md @@ -513,16 +513,15 @@ If a user configures a limit or request value that is too low, Rook will still r * `mon`: 1024MB * `mgr`: 512MB * `osd`: 2048MB -* `prepareosd`: 1200MB (see the note below) * `crashcollector`: 60MB * `mgr-sidecar`: 100MB limit, 40MB requests +* `prepareosd`: no limits (see the note) !!! note - * The OSD prepare job may burst its memory usage during the OSD provisioning. - We either recommend not setting memory limits on the OSD prepare job, or setting limits over 1GB to prevent OSD - provisioning failure due to memory constraints that are difficult to troubleshoot. The OSD prepare job - only bursts a single time per OSD. All future runs of the OSD prepare job will detect the OSD is already - provisioned and not require a burst. + We recommend not setting memory limits on the OSD prepare job to prevent OSD provisioning failure due to memory constraints. + The OSD prepare job bursts memory usage during the OSD provisioning depending on the size of the device, typically + 1-2Gi for large disks. The OSD prepare job only bursts a single time per OSD. + All future runs of the OSD prepare job will detect the OSD is already provisioned and skip the provisioning. !!! hint The resources for MDS daemons are not configured in the Cluster. Refer to the [Ceph Filesystem CRD](../Shared-Filesystem/ceph-filesystem-crd.md) instead. diff --git a/deploy/charts/rook-ceph-cluster/values.yaml b/deploy/charts/rook-ceph-cluster/values.yaml index 67a61813d..fc8f8ce76 100644 --- a/deploy/charts/rook-ceph-cluster/values.yaml +++ b/deploy/charts/rook-ceph-cluster/values.yaml @@ -257,9 +257,8 @@ cephClusterSpec: cpu: "1000m" memory: "4Gi" prepareosd: - limits: - cpu: "500m" - memory: "1200Mi" + # limits: It is not recommended to set limits on the OSD prepare job since it's a one-time burst for memory + # that must be allowed to complete without an OOM kill requests: cpu: "500m" memory: "50Mi"