import quincy beta 17.1.0

[ceph.git] / ceph / qa / standalone / osd / osd-recovery-prio.sh

diff --git a/ceph/qa/standalone/osd/osd-recovery-prio.sh b/ceph/qa/standalone/osd/osd-recovery-prio.sh
index fb386e265ab3525dce0b5cb0748f1a1a4c840964..02b65f67a27050d19fe7db874bbee9e1c1690ba5 100755 (executable)
--- a/ceph/qa/standalone/osd/osd-recovery-prio.sh
+++ b/ceph/qa/standalone/osd/osd-recovery-prio.sh
@@ -25,7 +25,10 @@ function run() {
     export CEPH_MON="127.0.0.1:7114" # git grep '\<7114\>' : there must be only one
     export CEPH_ARGS
     CEPH_ARGS+="--fsid=$(uuidgen) --auth-supported=none "
-    CEPH_ARGS+="--mon-host=$CEPH_MON --osd_max_backfills=1 --debug_reserver=20"
+    CEPH_ARGS+="--mon-host=$CEPH_MON --osd_max_backfills=1 --debug_reserver=20 "
+    # Set osd op queue = wpq for the tests. Recovery priority is not
+    # considered by mclock_scheduler leading to unexpected results.
+    CEPH_ARGS+="--osd-op-queue=wpq "
     export objects=200
     export poolprefix=test
     export FORCE_PRIO="255"    # See OSD_RECOVERY_PRIORITY_FORCED
@@ -125,8 +128,8 @@ function TEST_recovery_priority() {
       fi
     done
 
-    ceph osd pool set $pool2 size 1
-    ceph osd pool set $pool3 size 1
+    ceph osd pool set $pool2 size 1 --yes-i-really-mean-it
+    ceph osd pool set $pool3 size 1 --yes-i-really-mean-it
     wait_for_clean || return 1
 
     dd if=/dev/urandom of=$dir/data bs=1M count=10
@@ -401,9 +404,9 @@ function TEST_recovery_pool_priority() {
     pool1_prio=$(expr $NORMAL_PRIO + $pool1_extra_prio)
     pool2_prio=$(expr $NORMAL_PRIO + $pool2_extra_prio)
 
-    ceph osd pool set $pool1 size 1
+    ceph osd pool set $pool1 size 1 --yes-i-really-mean-it
     ceph osd pool set $pool1 recovery_priority $pool1_extra_prio
-    ceph osd pool set $pool2 size 1
+    ceph osd pool set $pool2 size 1 --yes-i-really-mean-it
     ceph osd pool set $pool2 recovery_priority $pool2_extra_prio
     wait_for_clean || return 1
 
@@ -425,7 +428,31 @@ function TEST_recovery_pool_priority() {
 
     ceph osd pool set $pool1 size 2
     ceph osd pool set $pool2 size 2
-    sleep 10
+
+    # Wait for both PGs to be in recovering state
+    ceph pg dump pgs
+
+    # Wait for recovery to start
+    set -o pipefail
+    count=0
+    while(true)
+    do
+      if test $(ceph --format json pg dump pgs |
+             jq '.pg_stats | .[] | .state | contains("recovering")' | grep -c true) == "2"
+      then
+        break
+      fi
+      sleep 2
+      if test "$count" -eq "10"
+      then
+        echo "Recovery never started on both PGs"
+        return 1
+      fi
+      count=$(expr $count + 1)
+    done
+    set +o pipefail
+    ceph pg dump pgs
+
     CEPH_ARGS='' ceph --admin-daemon $(get_asok_path osd.${chk_osd1_1}) dump_recovery_reservations > $dir/dump.${chk_osd1_1}.out
     echo osd.${chk_osd1_1}
     cat $dir/dump.${chk_osd1_1}.out