From 699ea1acbaf4eb1682539b2e8a0e2e594e333091 Mon Sep 17 00:00:00 2001
From: bw6938 <brian.wickersham@att.com>
Date: Thu, 28 Nov 2019 03:28:13 +0000
Subject: [PATCH] [ceph-client] Validate failure domain support for replica
 count per pool

Ensure each pool is configured with enough failure domains to
satisfy the pool's replica size requirements. If any pool does
not have enough failure domains to satisfy the pool's replica size,
then fail the ceph deployment.

Change-Id: I9dd1cafd05e81f145d1eb8c916591203946bc8f1
---
 ceph-client/templates/bin/_helm-tests.sh.tpl | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)

diff --git a/ceph-client/templates/bin/_helm-tests.sh.tpl b/ceph-client/templates/bin/_helm-tests.sh.tpl
index d3fe6ecdc..757c725cb 100755
--- a/ceph-client/templates/bin/_helm-tests.sh.tpl
+++ b/ceph-client/templates/bin/_helm-tests.sh.tpl
@@ -51,6 +51,24 @@ function check_osd_count() {
   fi
 }
 
+function check_failure_domain_count_per_pool() {
+  echo "#### Start: Checking failure domain count per pool ####"
+  pools=$(ceph osd pool ls)
+  for pool in ${pools}
+  do
+    crush_rule=$(ceph osd pool get ${pool} crush_rule | awk '{print $2}')
+    bucket_type=$(ceph osd crush rule dump ${crush_rule} | grep '"type":' | awk -F'"' 'NR==2 {print $4}')
+    num_failure_domains=$(ceph osd tree | grep ${bucket_type} | wc -l)
+    pool_replica_size=$(ceph osd pool get ${pool} size | awk '{print $2}')
+    if [[ ${num_failure_domains} -ge ${pool_replica_size} ]]; then
+      echo "--> Info: Pool ${pool} is configured with enough failure domains ${num_failure_domains} to satisfy pool replica size ${pool_replica_size}"
+    else
+      echo "--> Error : Pool ${pool} is NOT configured with enough failure domains ${num_failure_domains} to satisfy pool replica size ${pool_replica_size}"
+      exit 1
+    fi
+  done
+}
+
 function mgr_validation() {
   echo "#### Start: MGR validation ####"
   mgr_dump=$(ceph mgr dump -f json-pretty)
@@ -202,7 +220,9 @@ OSD_POOLS_DETAILS=$(ceph osd pool ls detail -f json-pretty)
 OSD_CRUSH_RULE_DUMP=$(ceph osd crush rule dump -f json-pretty)
 PG_STAT=$(ceph pg stat -f json-pretty)
 
+
 pg_validation
 pool_validation
 pool_failuredomain_validation
+check_failure_domain_count_per_pool
 check_cluster_status