glusterfs

1
#!/bin/bash
2

3
#Create gfid split-brain of directory and check if conservative merge
4
#completes successfully.
5

6
. $(dirname $0)/../../include.rc
7
. $(dirname $0)/../../volume.rc
8
. $(dirname $0)/../../afr.rc
9

10
function check_sh_entries() {
11
        local expected="$1"
12
        local count=
13
        local good="0"
14
        shift
15

16
        for i in $*; do
17
                count="$(count_sh_entries $i)"
18
                if [[ "x${count}" == "x${expected}" ]]; then
19
                        good="$((good + 1))"
20
                fi
21
        done
22
        if [[ "x${good}" != "x${last_good}" ]]; then
23
                last_good="${good}"
24
# This triggers a sweep of the heal index. However if more than one brick
25
# tries to heal the same directory at the same time, one of them will take
26
# the lock and the other will give up, waiting for the next heal cycle, which
27
# is set to 60 seconds (the minimum valid value). So, each time we detect
28
# that one brick has completed the heal, we trigger another heal.
29
                $CLI volume heal $V0
30
        fi
31

32
        echo "${good}"
33
}
34

35
cleanup;
36

37
TEST glusterd
38
TEST pidof glusterd
39

40
TEST $CLI volume create $V0 replica 2 $H0:$B0/brick{0,1}
41
TEST $CLI volume set $V0 cluster.heal-timeout 60
42
TEST $CLI volume set $V0 cluster.self-heal-daemon off
43
TEST $CLI volume set $V0 stat-prefetch off
44
TEST $CLI volume start $V0
45
TEST $GFS --volfile-id=$V0 --volfile-server=$H0 $M0;
46

47
#Create files with alternate brick down. One file has gfid mismatch.
48
TEST mkdir $M0/DIR
49

50
TEST kill_brick $V0 $H0 $B0/brick1
51
TEST touch $M0/DIR/FILE
52
TEST touch $M0/DIR/file{1..5}
53
TEST $CLI volume start $V0 force
54
EXPECT_WITHIN $CHILD_UP_TIMEOUT "1" afr_child_up_status $V0 1
55

56
TEST kill_brick $V0 $H0 $B0/brick0
57
TEST touch $M0/DIR/FILE
58
TEST touch $M0/DIR/file{6..10}
59
TEST $CLI volume start $V0 force
60
EXPECT_WITHIN $CHILD_UP_TIMEOUT "1" afr_child_up_status $V0 0
61

62
#Trigger heal and verify number of entries in backend
63
TEST $CLI volume set $V0 cluster.self-heal-daemon on
64
EXPECT_WITHIN $PROCESS_UP_TIMEOUT "Y" glustershd_up_status
65
EXPECT_WITHIN $CHILD_UP_TIMEOUT "1" afr_child_up_status_in_shd $V0 0
66
EXPECT_WITHIN $CHILD_UP_TIMEOUT "1" afr_child_up_status_in_shd $V0 1
67

68
TEST $CLI volume heal $V0
69

70
last_good=""
71

72
EXPECT_WITHIN $HEAL_TIMEOUT "2" check_sh_entries 2 $B0/brick{0,1}
73

74
#Two entries for DIR and two for FILE
75
EXPECT_WITHIN $HEAL_TIMEOUT "4" get_pending_heal_count $V0
76
TEST diff <(ls $B0/brick0/DIR) <(ls $B0/brick1/DIR)
77
cleanup
78