5 # bug number: 13129 13129 10124
6 ALWAYS_EXCEPT="2 3 15c $REPLAY_DUAL_EXCEPT"
9 PTLDEBUG=${PTLDEBUG:--1}
10 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
12 CLEANUP=${CLEANUP:-""}
13 MOUNT_2=${MOUNT_2:-"yes"}
14 . $LUSTRE/tests/test-framework.sh
16 if [ "$FAILURE_MODE" = "HARD" ] && mixed_ost_devs; then
17 CONFIG_EXCEPTIONS="17"
18 echo -n "Several ost services on one ost node are used with FAILURE_MODE=$FAILURE_MODE. "
19 echo "Except the tests: $CONFIG_EXCEPTIONS"
20 ALWAYS_EXCEPT="$ALWAYS_EXCEPT $CONFIG_EXCEPTIONS"
25 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
27 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
29 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 5 14"
33 cleanup_and_setup_lustre
35 rm -rf $DIR/[df][0-9]*
37 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
41 replay_barrier $SINGLEMDS
45 checkstat $MOUNT2/a || return 1
46 checkstat $MOUNT1/b || return 2
47 rm $MOUNT2/a $MOUNT1/b
48 checkstat $MOUNT1/a && return 3
49 checkstat $MOUNT2/b && return 4
53 run_test 1 "|X| simple create"
57 replay_barrier $SINGLEMDS
61 checkstat $MOUNT2/adir || return 1
63 checkstat $MOUNT2/adir && return 2
66 run_test 2 "|X| mkdir adir"
69 replay_barrier $SINGLEMDS
71 mkdir $MOUNT2/adir/bdir
74 checkstat $MOUNT2/adir || return 1
75 checkstat $MOUNT1/adir/bdir || return 2
76 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
77 checkstat $MOUNT1/adir && return 3
78 checkstat $MOUNT2/adir/bdir && return 4
81 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
85 replay_barrier $SINGLEMDS
86 mkdir $MOUNT1/adir && return 1
87 mkdir $MOUNT2/adir/bdir
90 checkstat $MOUNT2/adir || return 2
91 checkstat $MOUNT1/adir/bdir || return 3
93 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
94 checkstat $MOUNT1/adir && return 4
95 checkstat $MOUNT2/adir/bdir && return 5
98 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
102 # multiclient version of replay_single.sh/test_8
104 multiop_bg_pause $MOUNT2/a o_tSc || return 1
107 replay_barrier $SINGLEMDS
109 wait $pid || return 1
112 [ -e $MOUNT2/a ] && return 2
115 run_test 5 "open, unlink |X| close"
120 multiop_bg_pause $MOUNT2/a o_c || return 1
122 multiop_bg_pause $MOUNT1/a o_c || return 1
125 replay_barrier $SINGLEMDS
127 wait $pid1 || return 1
131 wait $pid2 || return 1
132 [ -e $MOUNT2/a ] && return 2
135 run_test 6 "open1, open2, unlink |X| close1 [fail $SINGLEMDS] close2"
138 replay_barrier $SINGLEMDS
139 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
141 checkstat $MOUNT2/$tfile || return 2
142 rm $MOUNT1/$tfile || return 3
146 run_test 8 "replay of resent request"
149 replay_barrier $SINGLEMDS
150 mcreate $MOUNT1/$tfile-1
151 mcreate $MOUNT2/$tfile-2
152 # drop first reint reply
153 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
155 do_facet $SINGLEMDS lctl set_param fail_loc=0
157 rm $MOUNT1/$tfile-[1,2] || return 1
161 run_test 9 "resending a replayed create"
164 mcreate $MOUNT1/$tfile-1
165 replay_barrier $SINGLEMDS
166 munlink $MOUNT1/$tfile-1
167 mcreate $MOUNT2/$tfile-2
168 # drop first reint reply
169 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
171 do_facet $SINGLEMDS lctl set_param fail_loc=0
173 checkstat $MOUNT1/$tfile-1 && return 1
174 checkstat $MOUNT1/$tfile-2 || return 2
179 run_test 10 "resending a replayed unlink"
182 replay_barrier $SINGLEMDS
183 mcreate $MOUNT1/$tfile-1
184 mcreate $MOUNT2/$tfile-2
185 mcreate $MOUNT1/$tfile-3
186 mcreate $MOUNT2/$tfile-4
187 mcreate $MOUNT1/$tfile-5
188 # drop all reint replies for a while
189 do_facet $SINGLEMDS lctl set_param fail_loc=0x0119
190 # note that with this fail_loc set, facet_failover df will fail
191 facet_failover $SINGLEMDS
192 #sleep for while, let both clients reconnect and timeout
193 sleep $((TIMEOUT * 2))
194 do_facet $SINGLEMDS lctl set_param fail_loc=0
196 rm $MOUNT1/$tfile-[1-5] || return 1
200 run_test 11 "both clients timeout during replay"
203 replay_barrier $SINGLEMDS
205 multiop_bg_pause $DIR/$tfile mo_c || return 1
208 #define OBD_FAIL_LDLM_ENQUEUE 0x302
209 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000302
210 facet_failover $SINGLEMDS
211 do_facet $SINGLEMDS lctl set_param fail_loc=0
212 df $MOUNT || return 1
215 kill -USR1 $MULTIPID || return 3
216 wait $MULTIPID || return 4
217 $CHECKSTAT -t file $DIR/$tfile || return 2
222 run_test 12 "open resend timeout"
225 multiop_bg_pause $DIR/$tfile mo_c || return 1
228 replay_barrier $SINGLEMDS
230 kill -USR1 $MULTIPID || return 3
231 wait $MULTIPID || return 4
234 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000115
235 facet_failover $SINGLEMDS
236 do_facet $SINGLEMDS lctl set_param fail_loc=0
237 df $MOUNT || return 1
240 $CHECKSTAT -t file $DIR/$tfile || return 2
245 run_test 13 "close resend timeout"
248 replay_barrier $SINGLEMDS
249 createmany -o $MOUNT1/$tfile- 25
250 createmany -o $MOUNT2/$tfile-2- 1
251 createmany -o $MOUNT1/$tfile-3- 25
254 facet_failover $SINGLEMDS
255 # expect failover to fail due to missing client 2
256 df $MOUNT && return 1
259 # first 25 files should have been replayed
260 unlinkmany $MOUNT1/$tfile- 25 || return 2
262 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
265 run_test 14 "timeouts waiting for lost client during replay"
267 test_15a() { # was test_15
268 replay_barrier $SINGLEMDS
269 createmany -o $MOUNT1/$tfile- 25
270 createmany -o $MOUNT2/$tfile-2- 1
273 facet_failover $SINGLEMDS
274 df $MOUNT || return 1
276 unlinkmany $MOUNT1/$tfile- 25 || return 2
277 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
279 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
282 run_test 15a "timeout waiting for lost client during replay, 1 client completes"
285 replay_barrier $SINGLEMDS
286 for ((i = 0; i < 2000; i++)); do
287 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
291 facet_failover $SINGLEMDS
293 df $MOUNT || return 1
295 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
298 run_test 15c "remove multiple OST orphans"
301 replay_barrier $SINGLEMDS
302 createmany -o $MOUNT1/$tfile- 25
303 createmany -o $MOUNT2/$tfile-2- 1
306 facet_failover $SINGLEMDS
308 facet_failover $SINGLEMDS
309 df $MOUNT || return 1
311 unlinkmany $MOUNT1/$tfile- 25 || return 2
313 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
317 run_test 16 "fail MDS during recovery (3571)"
320 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
322 createmany -o $MOUNT1/$tfile- 25
323 createmany -o $MOUNT2/$tfile-2- 1
325 # Make sure the disconnect is lost
332 df $MOUNT || return 1
334 unlinkmany $MOUNT1/$tfile- 25 || return 2
336 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
340 run_test 17 "fail OST during recovery (3571)"
342 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
345 test_18() { # bug 3822 - evicting client with enqueued lock
347 mkdir -p $MOUNT1/$tdir
348 touch $MOUNT1/$tdir/f0
349 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
350 statmany -s $MOUNT1/$tdir/f 1 500 &
353 do_facet $SINGLEMDS lctl set_param fail_loc=0x8000030b # hold enqueue
355 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
356 do_facet client lctl set_param fail_loc=0x80000305 # drop cb, evict
358 usleep 500 # wait to ensure first client is one that will be evicted
359 openfile -f O_RDONLY $MOUNT2/$tdir/f0
361 dmesg | grep "entering recovery in server" && \
362 error "client not evicted" || true
363 do_facet client "lctl set_param fail_loc=0"
364 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
366 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
368 test_19() { # Bug 10991 - resend of open request does not fail assertion.
369 replay_barrier $SINGLEMDS
370 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
372 checkstat $DIR2/${tfile}0 || return 2
373 rm $DIR/${tfile}0 || return 3
377 run_test 19 "resend of open request"
381 replay_barrier $SINGLEMDS
385 facet_failover $SINGLEMDS
386 df $MOUNT1 || return 1
388 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
389 TIER1=$((`date +%s` - BEFORE))
391 replay_barrier $SINGLEMDS
395 facet_failover $SINGLEMDS
396 df $MOUNT1 || return 1
398 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
399 TIER2=$((`date +%s` - BEFORE))
400 [ $TIER2 -ge $((TIER1 * 2)) ] && \
401 error "recovery time is growing $TIER2 > $TIER1"
404 run_test 20 "recovery time is not increasing"
407 local param_file=$TMP/$tfile-params
409 save_lustre_params $(facet_active_host $SINGLEMDS) "mdt.*.commit_on_sharing" > $param_file
410 do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=1
411 touch $MOUNT1/$tfile-1
412 mv $MOUNT2/$tfile-1 $MOUNT2/$tfile-2
413 mv $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
414 replay_barrier_nosync $SINGLEMDS
417 facet_failover $SINGLEMDS
419 # all renames are replayed
420 unlink $MOUNT1/$tfile-3 || return 2
422 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
424 do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=0
425 rm -rf $MOUNT1/$tfile-*
426 restore_lustre_params < $param_file
430 run_test 21 "commit on sharing"
432 equals_msg `basename $0`: test complete, cleaning up
433 SLEEP=$((`date +%s` - $NOW))
434 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
435 check_and_cleanup_lustre
436 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true