Merge pull request #60777 from VallariAg/wip-nvmeof-prometheus-rbd-image-reused

monitoring: Add prometheus alert NVMeoFMultipleNamespacesOfRBDImage

Reviewed-by: Afreen Misbah <afreen@ibm.com>

3 files changed, 67 insertions, 0 deletions

diff --git a/monitoring/ceph-mixin/prometheus_alerts.libsonnet b/monitoring/ceph-mixin/prometheus_alerts.libsonnet
index 143e65f20e7..5d1ab49b533 100644
--- a/monitoring/ceph-mixin/prometheus_alerts.libsonnet
+++ b/monitoring/ceph-mixin/prometheus_alerts.libsonnet
@@ -856,6 +856,16 @@
           },
         },
         {
+          alert: 'NVMeoFMultipleNamespacesOfRBDImage',
+          'for': '1m',
+          expr: 'count by(pool_name, rbd_name) (count by(bdev_name, pool_name, rbd_name) (ceph_nvmeof_bdev_metadata and on (bdev_name) ceph_nvmeof_subsystem_namespace_metadata)) > 1',
+          labels: { severity: 'warning', type: 'ceph_default' },
+          annotations: {
+            summary: 'RBD image {{ $labels.pool_name }}/{{ $labels.rbd_name }} cannot be reused for multiple NVMeoF namespace ',
+            description: 'Each NVMeoF namespace must have a unique RBD pool and image, across all different gateway groups.',
+          },
+        },
+        {
           alert: 'NVMeoFTooManyGateways',
           'for': '1m',
           expr: 'count(ceph_nvmeof_gateway_info) by (cluster) > %.2f' % [$._config.NVMeoFMaxGatewaysPerCluster],
diff --git a/monitoring/ceph-mixin/prometheus_alerts.yml b/monitoring/ceph-mixin/prometheus_alerts.yml
index 3eb8a8db4fa..3440d761351 100644
--- a/monitoring/ceph-mixin/prometheus_alerts.yml
+++ b/monitoring/ceph-mixin/prometheus_alerts.yml
@@ -765,6 +765,15 @@ groups:
         labels:
           severity: "warning"
           type: "ceph_default"
+      - alert: "NVMeoFMultipleNamespacesOfRBDImage"
+        annotations:
+          description: "Each NVMeoF namespace must have a unique RBD pool and image, across all different gateway groups."
+          summary: "RBD image {{ $labels.pool_name }}/{{ $labels.rbd_name }} cannot be reused for multiple NVMeoF namespace "
+        expr: "count by(pool_name, rbd_name) (count by(bdev_name, pool_name, rbd_name) (ceph_nvmeof_bdev_metadata and on (bdev_name) ceph_nvmeof_subsystem_namespace_metadata)) > 1"
+        for: "1m"
+        labels:
+          severity: "warning"
+          type: "ceph_default"
       - alert: "NVMeoFTooManyGateways"
         annotations:
           description: "You may create many gateways, but 4 is the tested limit"
diff --git a/monitoring/ceph-mixin/tests_alerts/test_alerts.yml b/monitoring/ceph-mixin/tests_alerts/test_alerts.yml
index a4e63bbcf73..b3b29308d08 100644
--- a/monitoring/ceph-mixin/tests_alerts/test_alerts.yml
+++ b/monitoring/ceph-mixin/tests_alerts/test_alerts.yml
@@ -2270,6 +2270,54 @@ tests:
           summary: "wah subsystem has reached its maximum number of namespaces on cluster mycluster"
           description: "Subsystems have a max namespace limit defined at creation time. This alert means that no more namespaces can be added to wah"
 
+# NVMeoFMultipleNamespacesOfRBDImage
+ - interval: 1m
+   input_series:
+    - series: 'ceph_nvmeof_bdev_metadata{bdev_name="bdev1", instance="ceph-nvme-vm1", pool_name="mypool", rbd_name="myimage1"}'
+      values: '1x10'
+    - series: 'ceph_nvmeof_bdev_metadata{bdev_name="bdev1", instance="ceph-nvme-vm2", pool_name="mypool", rbd_name="myimage1"}'
+      values: '1x10'
+    - series: 'ceph_nvmeof_bdev_metadata{bdev_name="bdev2", instance="ceph-nvme-vm1", pool_name="mypool", rbd_name="myimage2"}'
+      values: '1x10'
+    - series: 'ceph_nvmeof_bdev_metadata{bdev_name="bdev2", instance="ceph-nvme-vm2", pool_name="mypool", rbd_name="myimage2"}'
+      values: '1x10'
+    - series: 'ceph_nvmeof_bdev_metadata{bdev_name="bdev3", instance="ceph-nvme-vm1", pool_name="mypool", rbd_name="myimage1"}'
+      values: '1x10'
+    - series: 'ceph_nvmeof_bdev_metadata{bdev_name="bdev3", instance="ceph-nvme-vm2", pool_name="mypool", rbd_name="myimage1"}'
+      values: '1x10'
+    - series: 'ceph_nvmeof_bdev_metadata{bdev_name="bdev4", instance="ceph-nvme-vm1", pool_name="mypool", rbd_name="myimage1"}' # bdev with no ns
+      values: '1x10'
+    - series: 'ceph_nvmeof_subsystem_namespace_metadata{nqn="nqn1", nsid="1", bdev_name="bdev1", instance="ceph-nvme-vm1", cluster="mycluster"}'
+      values: '1x10'
+    - series: 'ceph_nvmeof_subsystem_namespace_metadata{nqn="nqn1", nsid="1", bdev_name="bdev1", instance="ceph-nvme-vm2", cluster="mycluster"}'
+      values: '1x10'
+    - series: 'ceph_nvmeof_subsystem_namespace_metadata{nqn="nqn1", nsid="2", bdev_name="bdev2", instance="ceph-nvme-vm1", cluster="mycluster"}'
+      values: '1x10'
+    - series: 'ceph_nvmeof_subsystem_namespace_metadata{nqn="nqn1", nsid="2", bdev_name="bdev2", instance="ceph-nvme-vm2", cluster="mycluster"}'
+      values: '1x10'
+    - series: 'ceph_nvmeof_subsystem_namespace_metadata{nqn="nqn2", nsid="1", bdev_name="bdev3", instance="ceph-nvme-vm1", cluster="mycluster"}'
+      values: '1x10'
+    - series: 'ceph_nvmeof_subsystem_namespace_metadata{nqn="nqn2", nsid="1", bdev_name="bdev3", instance="ceph-nvme-vm2", cluster="mycluster"}'
+      values: '1x10'
+   promql_expr_test:
+     - expr: count by(pool_name, rbd_name) (count by(bdev_name, pool_name, rbd_name) (ceph_nvmeof_bdev_metadata and on (bdev_name) ceph_nvmeof_subsystem_namespace_metadata)) > 1
+       eval_time: 1m
+       exp_samples:
+         - labels: '{pool_name="mypool", rbd_name="myimage1"}'
+           value: 2
+   alert_rule_test:
+    - eval_time: 5m
+      alertname: NVMeoFMultipleNamespacesOfRBDImage
+      exp_alerts:
+      - exp_labels:
+          pool_name: mypool
+          rbd_name: myimage1 
+          severity: warning
+          type: ceph_default
+        exp_annotations:
+          summary: "RBD image mypool/myimage1 cannot be reused for multiple NVMeoF namespace "
+          description: "Each NVMeoF namespace must have a unique RBD pool and image, across all different gateway groups."
+
  # NVMeoFTooManyGateways
  - interval: 1m
    input_series: