test balancer: Backport specific fixes

Add "ceph balancer sleep" command to set balancer sleep_interval for testing
Remove unavailable "ceph balancer pool" part of testing
Remove setting of nonexistant osd_pool_default_pg_autoscale_mode
Improve balancer module log message
Fix log message test (no pg merging)
Pool balancing isn't grouped by rule, so results different here

Signed-off-by: David Zafman <dzafman@redhat.com>
(cherry picked from commit 61dedc249bb93a208e685c68a65a5c07d60870d1)
	From mimic backport

diff --git a/qa/standalone/mgr/balancer.sh b/qa/standalone/mgr/balancer.sh
index b8dffccbe8fa860aa2b692f85b11fc0275f67b5f..71acc0d2ad05a2cc93ad001e830fc46fd3f3cc8f 100755 (executable)
--- a/qa/standalone/mgr/balancer.sh
+++ b/qa/standalone/mgr/balancer.sh
@@ -64,18 +64,6 @@ function TEST_balancer() {
     test "$EVAL" = "current cluster score 0.000000 (lower is better)"
     ceph balancer eval-verbose || return 1
 
-    ceph balancer pool add $TEST_POOL1 || return 1
-    ceph balancer pool add $TEST_POOL2 || return 1
-    ceph balancer pool ls || return 1
-    eval POOL=$(ceph balancer pool ls | jq '.[0]')
-    test "$POOL" = "$TEST_POOL1" || return 1
-    eval POOL=$(ceph balancer pool ls | jq '.[1]')
-    test "$POOL" = "$TEST_POOL2" || return 1
-    ceph balancer pool rm $TEST_POOL1 || return 1
-    ceph balancer pool rm $TEST_POOL2 || return 1
-    ceph balancer pool ls || return 1
-    ceph balancer pool add $TEST_POOL1 || return 1
-
     ceph balancer mode crush-compat || return 1
     ceph balancer status || return 1
     eval MODE=$(ceph balancer status | jq '.mode')
@@ -104,7 +92,7 @@ function TEST_balancer() {
     ! ceph balancer optimize plan_upmap $TEST_POOL || return 1
     ceph balancer status || return 1
     eval RESULT=$(ceph balancer status | jq '.optimize_result')
-    test "$RESULT" = "Unable to find further optimization, or pool(s)' pg_num is decreasing, or distribution is already perfect" || return 1
+    test "$RESULT" = "Unable to find further optimization, or distribution is already perfect" || return 1
 
     ceph balancer on || return 1
     ACTIVE=$(ceph balancer status | jq '.active')
@@ -130,7 +118,6 @@ function TEST_balancer2() {
     # Integer average of PGS per OSD (150)
     FINAL_PER_OSD2=$(expr \( \( $TEST_PGS1 + $TEST_PGS2 \) \* $DEFAULT_REPLICAS \) / $OSDS)
 
-    CEPH_ARGS+="--osd_pool_default_pg_autoscale_mode=off "
     CEPH_ARGS+="--debug_osd=20 "
     setup $dir || return 1
     run_mon $dir a || return 1
@@ -143,7 +130,7 @@ function TEST_balancer2() {
     ceph osd set-require-min-compat-client luminous
     ceph balancer mode upmap || return 1
     ceph balancer on || return 1
-    ceph config set mgr mgr/balancer/sleep_interval 5
+    ceph balancer sleep 5
 
     create_pool $TEST_POOL1 $TEST_PGS1
 
@@ -195,17 +182,20 @@ function TEST_balancer2() {
     sleep 30
     ceph osd df
 
-    # FINAL_PER_OSD2 should distribute evenly
+    # We should be with plue or minus 1 of FINAL_PER_OSD2
+    # This is because here each pool is balanced independently
+    MIN=$(expr $FINAL_PER_OSD2 - 1)
+    MAX=$(expr $FINAL_PER_OSD2 + 1)
     PGS=$(ceph osd df --format=json-pretty | jq '.nodes[0].pgs')
-    test $PGS -eq $FINAL_PER_OSD2 || return 1
+    test $PGS -ge $MIN -a $PGS -le $MAX || return 1
     PGS=$(ceph osd df --format=json-pretty | jq '.nodes[1].pgs')
-    test $PGS -eq $FINAL_PER_OSD2 || return 1
+    test $PGS -ge $MIN -a $PGS -le $MAX || return 1
     PGS=$(ceph osd df --format=json-pretty | jq '.nodes[2].pgs')
-    test $PGS -eq $FINAL_PER_OSD2 || return 1
+    test $PGS -ge $MIN -a $PGS -le $MAX || return 1
     PGS=$(ceph osd df --format=json-pretty | jq '.nodes[3].pgs')
-    test $PGS -eq $FINAL_PER_OSD2 || return 1
+    test $PGS -ge $MIN -a $PGS -le $MAX || return 1
     PGS=$(ceph osd df --format=json-pretty | jq '.nodes[4].pgs')
-    test $PGS -eq $FINAL_PER_OSD2 || return 1
+    test $PGS -ge $MIN -a $PGS -le $MAX || return 1
 
     teardown $dir || return 1
 }

diff --git a/src/pybind/mgr/balancer/module.py b/src/pybind/mgr/balancer/module.py
index 6b1abc0ad37e69db7ac2c43c9ef553341204373a..612015bc1c2a353a27fcf5f9c629fe0d7abdea9b 100644 (file)
--- a/src/pybind/mgr/balancer/module.py
+++ b/src/pybind/mgr/balancer/module.py
@@ -269,6 +269,11 @@ class Module(MgrModule):
             "desc": "Execute an optimization plan",
             "perm": "rw",
         },
+        {
+            "cmd": "balancer sleep name=secs,type=CephString",
+            "desc": "Set balancer sleep interval",
+            "perm": "rw",
+        },
     ]
     active = False
     run = True
@@ -403,6 +408,9 @@ class Module(MgrModule):
             r, detail = self.execute(plan)
             self.plan_rm(command['plan'])
             return (r, '', detail)
+        elif command['prefix'] == 'balancer sleep':
+            self.set_config('sleep_interval', command['secs'])
+            return (0, "", '')
         else:
             return (-errno.EINVAL, '',
                     "Command not found '{0}'".format(command['prefix']))
@@ -771,7 +779,7 @@ class Module(MgrModule):
                 break
         self.log.info('prepared %d/%d changes' % (total_did, max_iterations))
         if total_did == 0:
-            return -errno.EALREADY, 'Unable to find further optimization,' \
+            return -errno.EALREADY, 'Unable to find further optimization, ' \
                                     'or distribution is already perfect'
         return 0, ''