1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
|
"""
Backfill_toofull
"""
import logging
import time
from tasks import ceph_manager
from tasks.util.rados import rados
from teuthology import misc as teuthology
log = logging.getLogger(__name__)
def wait_for_pg_state(manager, pgid, state, to_osd):
log.debug("waiting for pg %s state is %s" % (pgid, state))
for i in range(300):
time.sleep(5)
manager.flush_pg_stats([0, 1, 2, 3])
pgs = manager.get_pg_stats()
pg = next((pg for pg in pgs if pg['pgid'] == pgid), None)
log.info('pg=%s' % pg);
assert pg
status = pg['state'].split('+')
if 'active' not in status:
log.debug('not active')
continue
if state not in status:
log.debug('not %s' % state)
continue
assert to_osd in pg['up']
return
assert False, '%s not in %s' % (pgid, state)
def task(ctx, config):
"""
Test backfill reservation calculates "toofull" condition correctly.
A pretty rigid cluster is brought up and tested by this task
"""
if config is None:
config = {}
assert isinstance(config, dict), \
'backfill_toofull task only accepts a dict for configuration'
first_mon = teuthology.get_first_mon(ctx, config)
(mon,) = ctx.cluster.only(first_mon).remotes.keys()
manager = ceph_manager.CephManager(
mon,
ctx=ctx,
logger=log.getChild('ceph_manager'),
)
profile = config.get('erasure_code_profile', {
'k': '2',
'm': '1',
'crush-failure-domain': 'osd'
})
profile_name = profile.get('name', 'backfill_toofull')
manager.create_erasure_code_profile(profile_name, profile)
pool = manager.create_pool_with_unique_name(
pg_num=1,
erasure_code_profile_name=profile_name,
min_size=2)
manager.raw_cluster_cmd('osd', 'pool', 'set', pool,
'pg_autoscale_mode', 'off')
manager.flush_pg_stats([0, 1, 2, 3])
manager.wait_for_clean()
pool_id = manager.get_pool_num(pool)
pgid = '%d.0' % pool_id
pgs = manager.get_pg_stats()
acting = next((pg['acting'] for pg in pgs if pg['pgid'] == pgid), None)
log.debug("acting=%s" % acting)
assert acting
primary = acting[0]
target = acting[1]
log.debug("write some data")
rados(ctx, mon, ['-p', pool, 'bench', '120', 'write', '--no-cleanup'])
df = manager.get_osd_df(target)
log.debug("target osd df: %s" % df)
total_kb = df['kb']
used_kb = df['kb_used']
log.debug("pause recovery")
manager.raw_cluster_cmd('osd', 'set', 'noout')
manager.raw_cluster_cmd('osd', 'set', 'nobackfill')
manager.raw_cluster_cmd('osd', 'set', 'norecover')
log.debug("stop tartget osd %s" % target)
manager.kill_osd(target)
manager.wait_till_active()
pgs = manager.get_pg_stats()
pg = next((pg for pg in pgs if pg['pgid'] == pgid), None)
log.debug('pg=%s' % pg)
assert pg
log.debug("re-write data")
rados(ctx, mon, ['-p', pool, 'cleanup'])
time.sleep(10)
rados(ctx, mon, ['-p', pool, 'bench', '60', 'write', '--no-cleanup'])
df = manager.get_osd_df(primary)
log.debug("primary osd df: %s" % df)
primary_used_kb = df['kb_used']
log.info("test backfill reservation rejected with toofull")
# We set backfillfull ratio less than new data size and expect the pg
# entering backfill_toofull state.
#
# We also need to update nearfull ratio to prevent "full ratio(s) out of order".
backfillfull = 0.9 * primary_used_kb / total_kb
nearfull = backfillfull * 0.9
log.debug("update nearfull ratio to %s and backfillfull ratio to %s" %
(nearfull, backfillfull))
manager.raw_cluster_cmd('osd', 'set-nearfull-ratio',
'{:.3f}'.format(nearfull + 0.001))
manager.raw_cluster_cmd('osd', 'set-backfillfull-ratio',
'{:.3f}'.format(backfillfull + 0.001))
log.debug("start tartget osd %s" % target)
manager.revive_osd(target)
manager.wait_for_active()
manager.wait_till_osd_is_up(target)
wait_for_pg_state(manager, pgid, 'backfill_toofull', target)
log.info("test pg not enter backfill_toofull after restarting backfill")
# We want to set backfillfull ratio to be big enough for the target to
# successfully backfill new data but smaller than the sum of old and new
# data, so if the osd backfill reservation incorrectly calculates "toofull"
# the test will detect this (fail).
#
# Note, we need to operate with "uncompressed" bytes because currently
# osd backfill reservation does not take compression into account.
#
# We also need to update nearfull ratio to prevent "full ratio(s) out of order".
pdf = manager.get_pool_df(pool)
log.debug("pool %s df: %s" % (pool, pdf))
assert pdf
compress_ratio = 1.0 * pdf['compress_under_bytes'] / pdf['compress_bytes_used'] \
if pdf['compress_bytes_used'] > 0 else 1.0
log.debug("compress_ratio: %s" % compress_ratio)
backfillfull = (used_kb + primary_used_kb) * compress_ratio / total_kb
assert backfillfull < 0.9
nearfull_min = max(used_kb, primary_used_kb) * compress_ratio / total_kb
assert nearfull_min < backfillfull
delta = backfillfull - nearfull_min
nearfull = nearfull_min + delta * 0.1
backfillfull = nearfull_min + delta * 0.2
log.debug("update nearfull ratio to %s and backfillfull ratio to %s" %
(nearfull, backfillfull))
manager.raw_cluster_cmd('osd', 'set-nearfull-ratio',
'{:.3f}'.format(nearfull + 0.001))
manager.raw_cluster_cmd('osd', 'set-backfillfull-ratio',
'{:.3f}'.format(backfillfull + 0.001))
wait_for_pg_state(manager, pgid, 'backfilling', target)
pgs = manager.get_pg_stats()
pg = next((pg for pg in pgs if pg['pgid'] == pgid), None)
log.debug('pg=%s' % pg)
assert pg
log.debug("interrupt %s backfill" % target)
manager.mark_down_osd(target)
# after marking the target osd down it will automatically be
# up soon again
log.debug("resume recovery")
manager.raw_cluster_cmd('osd', 'unset', 'noout')
manager.raw_cluster_cmd('osd', 'unset', 'nobackfill')
manager.raw_cluster_cmd('osd', 'unset', 'norecover')
# wait for everything to peer, backfill and recover
manager.wait_for_clean()
pgs = manager.get_pg_stats()
pg = next((pg for pg in pgs if pg['pgid'] == pgid), None)
log.info('pg=%s' % pg)
assert pg
assert 'clean' in pg['state'].split('+')
|
