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/$NAME.sh}
28 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 5 14"
32 cleanup_and_setup_lustre
34 rm -rf $DIR/[df][0-9]*
36 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
40 replay_barrier $SINGLEMDS
44 checkstat $MOUNT2/a || return 1
45 checkstat $MOUNT1/b || return 2
46 rm $MOUNT2/a $MOUNT1/b
47 checkstat $MOUNT1/a && return 3
48 checkstat $MOUNT2/b && return 4
52 run_test 1 "|X| simple create"
56 replay_barrier $SINGLEMDS
60 checkstat $MOUNT2/adir || return 1
62 checkstat $MOUNT2/adir && return 2
65 run_test 2 "|X| mkdir adir"
68 replay_barrier $SINGLEMDS
70 mkdir $MOUNT2/adir/bdir
73 checkstat $MOUNT2/adir || return 1
74 checkstat $MOUNT1/adir/bdir || return 2
75 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
76 checkstat $MOUNT1/adir && return 3
77 checkstat $MOUNT2/adir/bdir && return 4
80 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
84 replay_barrier $SINGLEMDS
85 mkdir $MOUNT1/adir && return 1
86 mkdir $MOUNT2/adir/bdir
89 checkstat $MOUNT2/adir || return 2
90 checkstat $MOUNT1/adir/bdir || return 3
92 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
93 checkstat $MOUNT1/adir && return 4
94 checkstat $MOUNT2/adir/bdir && return 5
97 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
101 # multiclient version of replay_single.sh/test_8
103 multiop_bg_pause $MOUNT2/a o_tSc || return 1
106 replay_barrier $SINGLEMDS
108 wait $pid || return 1
111 [ -e $MOUNT2/a ] && return 2
114 run_test 5 "open, unlink |X| close"
119 multiop_bg_pause $MOUNT2/a o_c || return 1
121 multiop_bg_pause $MOUNT1/a o_c || return 1
124 replay_barrier $SINGLEMDS
126 wait $pid1 || return 1
130 wait $pid2 || return 1
131 [ -e $MOUNT2/a ] && return 2
134 run_test 6 "open1, open2, unlink |X| close1 [fail $SINGLEMDS] close2"
137 replay_barrier $SINGLEMDS
138 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
140 checkstat $MOUNT2/$tfile || return 2
141 rm $MOUNT1/$tfile || return 3
145 run_test 8 "replay of resent request"
148 replay_barrier $SINGLEMDS
149 mcreate $MOUNT1/$tfile-1
150 mcreate $MOUNT2/$tfile-2
151 # drop first reint reply
152 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
154 do_facet $SINGLEMDS lctl set_param fail_loc=0
156 rm $MOUNT1/$tfile-[1,2] || return 1
160 run_test 9 "resending a replayed create"
163 mcreate $MOUNT1/$tfile-1
164 replay_barrier $SINGLEMDS
165 munlink $MOUNT1/$tfile-1
166 mcreate $MOUNT2/$tfile-2
167 # drop first reint reply
168 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
170 do_facet $SINGLEMDS lctl set_param fail_loc=0
172 checkstat $MOUNT1/$tfile-1 && return 1
173 checkstat $MOUNT1/$tfile-2 || return 2
178 run_test 10 "resending a replayed unlink"
181 replay_barrier $SINGLEMDS
182 mcreate $MOUNT1/$tfile-1
183 mcreate $MOUNT2/$tfile-2
184 mcreate $MOUNT1/$tfile-3
185 mcreate $MOUNT2/$tfile-4
186 mcreate $MOUNT1/$tfile-5
187 # drop all reint replies for a while
188 do_facet $SINGLEMDS lctl set_param fail_loc=0x0119
189 # note that with this fail_loc set, facet_failover df will fail
190 facet_failover $SINGLEMDS
191 #sleep for while, let both clients reconnect and timeout
192 sleep $((TIMEOUT * 2))
193 do_facet $SINGLEMDS lctl set_param fail_loc=0
195 rm $MOUNT1/$tfile-[1-5] || return 1
199 run_test 11 "both clients timeout during replay"
202 replay_barrier $SINGLEMDS
204 multiop_bg_pause $DIR/$tfile mo_c || return 1
207 #define OBD_FAIL_LDLM_ENQUEUE 0x302
208 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000302
209 facet_failover $SINGLEMDS
210 do_facet $SINGLEMDS lctl set_param fail_loc=0
211 df $MOUNT || return 1
214 kill -USR1 $MULTIPID || return 3
215 wait $MULTIPID || return 4
216 $CHECKSTAT -t file $DIR/$tfile || return 2
221 run_test 12 "open resend timeout"
224 multiop_bg_pause $DIR/$tfile mo_c || return 1
227 replay_barrier $SINGLEMDS
229 kill -USR1 $MULTIPID || return 3
230 wait $MULTIPID || return 4
233 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000115
234 facet_failover $SINGLEMDS
235 do_facet $SINGLEMDS lctl set_param fail_loc=0
236 df $MOUNT || return 1
239 $CHECKSTAT -t file $DIR/$tfile || return 2
244 run_test 13 "close resend timeout"
247 replay_barrier $SINGLEMDS
248 createmany -o $MOUNT1/$tfile- 25
249 createmany -o $MOUNT2/$tfile-2- 1
250 createmany -o $MOUNT1/$tfile-3- 25
253 facet_failover $SINGLEMDS
254 # expect failover to fail due to missing client 2
255 df $MOUNT && return 1
258 # first 25 files should have been replayed
259 unlinkmany $MOUNT1/$tfile- 25 || return 2
261 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
264 run_test 14 "timeouts waiting for lost client during replay"
266 test_15a() { # was test_15
267 replay_barrier $SINGLEMDS
268 createmany -o $MOUNT1/$tfile- 25
269 createmany -o $MOUNT2/$tfile-2- 1
272 facet_failover $SINGLEMDS
273 df $MOUNT || return 1
275 unlinkmany $MOUNT1/$tfile- 25 || return 2
276 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
278 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
281 run_test 15a "timeout waiting for lost client during replay, 1 client completes"
284 replay_barrier $SINGLEMDS
285 for ((i = 0; i < 2000; i++)); do
286 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
290 facet_failover $SINGLEMDS
292 df $MOUNT || return 1
294 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
297 run_test 15c "remove multiple OST orphans"
300 replay_barrier $SINGLEMDS
301 createmany -o $MOUNT1/$tfile- 25
302 createmany -o $MOUNT2/$tfile-2- 1
305 facet_failover $SINGLEMDS
307 facet_failover $SINGLEMDS
308 df $MOUNT || return 1
310 unlinkmany $MOUNT1/$tfile- 25 || return 2
312 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
316 run_test 16 "fail MDS during recovery (3571)"
319 createmany -o $MOUNT1/$tfile- 25
320 createmany -o $MOUNT2/$tfile-2- 1
322 # Make sure the disconnect is lost
329 df $MOUNT || return 1
331 unlinkmany $MOUNT1/$tfile- 25 || return 2
333 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
337 run_test 17 "fail OST during recovery (3571)"
339 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
342 test_18() { # bug 3822 - evicting client with enqueued lock
344 mkdir -p $MOUNT1/$tdir
345 touch $MOUNT1/$tdir/f0
346 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
347 statmany -s $MOUNT1/$tdir/f 1 500 &
350 do_facet $SINGLEMDS lctl set_param fail_loc=0x8000030b # hold enqueue
352 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
353 do_facet client lctl set_param fail_loc=0x80000305 # drop cb, evict
355 usleep 500 # wait to ensure first client is one that will be evicted
356 openfile -f O_RDONLY $MOUNT2/$tdir/f0
358 dmesg | grep "entering recovery in server" && \
359 error "client not evicted" || true
360 do_facet client "lctl set_param fail_loc=0"
361 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
363 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
365 test_19() { # Bug 10991 - resend of open request does not fail assertion.
366 replay_barrier $SINGLEMDS
367 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
369 checkstat $DIR2/${tfile}0 || return 2
370 rm $DIR/${tfile}0 || return 3
374 run_test 19 "resend of open request"
376 equals_msg `basename $0`: test complete, cleaning up
377 SLEEP=$((`date +%s` - $NOW))
378 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
379 check_and_cleanup_lustre
380 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true