From bce388ebbe4f854f9d3bdcbc85de0d1920dd049e Mon Sep 17 00:00:00 2001 From: Jiffin Tony Thottan Date: Wed, 15 Jun 2022 22:37:16 +0530 Subject: [PATCH] docs: mention about reconciliation issue in hpa docs During reconciliation of cephobjectstore, Rook operator resets the value of replica count by the HPA and HPA brings back it up when it notices the desired count for replica change. Signed-off-by: Jiffin Tony Thottan (cherry picked from commit 823c06da44749732c1ae1fe940a2969c1430a165) --- .../Storage-Configuration/Monitoring/ceph-monitoring.md | 3 +++ 1 file changed, 3 insertions(+) diff --git a/Documentation/Storage-Configuration/Monitoring/ceph-monitoring.md b/Documentation/Storage-Configuration/Monitoring/ceph-monitoring.md index c3e16d9b1..d2391d573 100644 --- a/Documentation/Storage-Configuration/Monitoring/ceph-monitoring.md +++ b/Documentation/Storage-Configuration/Monitoring/ceph-monitoring.md @@ -300,3 +300,6 @@ spec: sum(rate(ceph_rgw_put[2m])) # prometheus query used for autoscaling threshold: "90" ``` + +!!! warning + During reconciliation of a `CephObjectStore`, the Rook Operator will reset the replica count for RGW which was set by horizontal pod scaler. The horizontal pod autoscaler will change the again once it re-evaluates the rule. This can result in a performance hiccup of several seconds after a reconciliation. This is briefly discussed (here)[https://github.com/rook/rook/issues/10001]