From 5253cbfa010361c26a7073abb3872246615bd9aa Mon Sep 17 00:00:00 2001
From: Greg <2653109+glinton@users.noreply.github.com>
Date: Mon, 25 Feb 2019 17:02:30 -0700
Subject: [PATCH] Add ceph_health metrics to ceph input (#5482)

---
 plugins/inputs/ceph/README.md |  6 ++++++
 plugins/inputs/ceph/ceph.go   | 15 +++++++++++++++
 2 files changed, 21 insertions(+)

diff --git a/plugins/inputs/ceph/README.md b/plugins/inputs/ceph/README.md
index 88ca4aff7c171..33585b079562a 100644
--- a/plugins/inputs/ceph/README.md
+++ b/plugins/inputs/ceph/README.md
@@ -136,6 +136,11 @@ All admin measurements will have the following tags:
 
 *Cluster Stats*
 
++ ceph_health
+  - fields:
+    - status
+    - overall_status
+
 - ceph_osdmap
   - fields:
     - epoch (float)
@@ -209,6 +214,7 @@ ceph_pool_usage,name=telegraf kb_used=0,bytes_used=0,objects=0 15506589110000000
 ceph_pgmap_state,state=undersized+peered count=30 1550658910000000000
 ceph_pgmap bytes_total=10733223936,read_op_per_sec=0,write_op_per_sec=0,num_pgs=30,data_bytes=0,bytes_avail=9654697984,read_bytes_sec=0,write_bytes_sec=0,version=0,bytes_used=1078525952 1550658910000000000
 ceph_osdmap num_up_osds=1,num_in_osds=1,full=false,nearfull=false,num_remapped_pgs=0,epoch=34,num_osds=1 1550658910000000000
+ceph_health status="HEALTH_WARN",overall_status="HEALTH_WARN" 1550658910000000000
 ```
 
 *Admin Socket Stats*
diff --git a/plugins/inputs/ceph/ceph.go b/plugins/inputs/ceph/ceph.go
index d3911102d8011..e28f977d23777 100644
--- a/plugins/inputs/ceph/ceph.go
+++ b/plugins/inputs/ceph/ceph.go
@@ -319,6 +319,10 @@ func (c *Ceph) exec(command string) (string, error) {
 
 // CephStatus is used to unmarshal "ceph -s" output
 type CephStatus struct {
+	Health struct {
+		Status        string `json:"status"`
+		OverallStatus string `json:"overall_status"`
+	} `json:"health"`
 	OSDMap struct {
 		OSDMap struct {
 			Epoch          float64 `json:"epoch"`
@@ -357,6 +361,7 @@ func decodeStatus(acc telegraf.Accumulator, input string) error {
 	}
 
 	decoders := []func(telegraf.Accumulator, *CephStatus) error{
+		decodeStatusHealth,
 		decodeStatusOsdmap,
 		decodeStatusPgmap,
 		decodeStatusPgmapState,
@@ -371,6 +376,16 @@ func decodeStatus(acc telegraf.Accumulator, input string) error {
 	return nil
 }
 
+// decodeStatusHealth decodes the health portion of the output of 'ceph status'
+func decodeStatusHealth(acc telegraf.Accumulator, data *CephStatus) error {
+	fields := map[string]interface{}{
+		"status":         data.Health.Status,
+		"overall_status": data.Health.OverallStatus,
+	}
+	acc.AddFields("ceph_health", fields, map[string]string{})
+	return nil
+}
+
 // decodeStatusOsdmap decodes the OSD map portion of the output of 'ceph -s'
 func decodeStatusOsdmap(acc telegraf.Accumulator, data *CephStatus) error {
 	fields := map[string]interface{}{