From edb92fe4e8792f998a677c127f1c8b0ee434214b Mon Sep 17 00:00:00 2001
From: Chris Grindstaff <chris@gstaff.org>
Date: Mon, 18 Nov 2024 15:08:37 -0500
Subject: [PATCH] feat: Harvest should monitor `wafl.dir.size.warning`

Fixes: #3243
---
 conf/ems/9.6.0/ems.yaml                  |  7 +++++++
 container/prometheus/ems_alert_rules.yml | 25 +++++++++++++++++++++++-
 docs/resources/ems-alert-runbook.md      | 12 ++++++++++++
 3 files changed, 43 insertions(+), 1 deletion(-)

diff --git a/conf/ems/9.6.0/ems.yaml b/conf/ems/9.6.0/ems.yaml
index c8b1014d4..4ebfe64ce 100644
--- a/conf/ems/9.6.0/ems.yaml
+++ b/conf/ems/9.6.0/ems.yaml
@@ -944,6 +944,13 @@ events:
       - parameters.mirror_config_id  => mirror_config_id
       - parameters.primary_config_id => primary_config_id
 
+  - name: wafl.dir.size.warning
+    exports:
+      - parameters.fileid    => directory_inum
+      - parameters.vol       => volume
+      - parameters.app       => app
+      - parameters.volident  => vol_ident
+
   - name: wafl.readdir.expired
     exports:
       - parameters.app            => app
diff --git a/container/prometheus/ems_alert_rules.yml b/container/prometheus/ems_alert_rules.yml
index e69d452a4..2319ef212 100644
--- a/container/prometheus/ems_alert_rules.yml
+++ b/container/prometheus/ems_alert_rules.yml
@@ -2008,4 +2008,27 @@ groups:
             {{- end -}}
         annotations:
           summary: "SnapMirror active sync planned failover operation completed for Destination path: \"{{ $labels.dstpath }}\"."
-          impact: "Protection"
\ No newline at end of file
+          impact: "Protection"
+
+      - alert: Directory size is approaching the maximum directory size (maxdirsize) limit
+        expr: last_over_time(ems_events{message="wafl.dir.size.warning"}[5m]) == 1
+        labels:
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
+        annotations:
+          summary: "Directory size for file ID \"{{ $labels.directory_inum }}\" in volume \"{{ $labels.volume }}\" is approaching the maximum directory size (maxdirsize) limit."
+          impact: "Availability"
diff --git a/docs/resources/ems-alert-runbook.md b/docs/resources/ems-alert-runbook.md
index ecc2fd5bd..74981f4ee 100644
--- a/docs/resources/ems-alert-runbook.md
+++ b/docs/resources/ems-alert-runbook.md
@@ -52,6 +52,18 @@ If you use Cloud Volumes ONTAP, perform the following corrective actions:
   2. Ensure that the login and connectivity information is still valid.
 Contact NetApp technical support if the issue persists.
 
+### Directory size is approaching the maximum directory size (maxdirsize) limit
+
+**Impact**: Availability
+
+**EMS Event**: `wafl.dir.size.warning`
+
+This message occurs when the size of a directory surpasses a configured percentage (default: 90%) of its current maximum directory size (maxdirsize) limit.
+
+**Remediation**
+
+Use the "volume file show-inode" command with the file ID and volume name information to find the file path. Reduce the number of files in the directory. If not possible, use the (privilege:advanced) option "volume modify -volume vol_name -maxdir-size new_value" to increase the maximum number of files per directory. However, doing so could impact system performance. If you need to increase the maximum directory size, contact NetApp technical support.
+
 ### Disk Out of Service
 
 **Impact**: Availability