[LU-6167] endless loop in lustre_rsync

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.7.0
• Affects Version/s: Lustre 2.5.3
• Labels: patch
• Environment: CentOS
• Severity: 3
• Assignee: Yang Sheng
• Reporter: wu libin (Inactive)

Description

When test lustre_rsync combined with racer, some strange behavior come out.
1. follw is the test case i use:

1. Test 13 - lustre_rsync, use racer test suite
   test_13() {
   init_src
   init_changelog

local rrc=0
local rc=0
local clients=$CLIENTS
local RDIRS
local i

1. 1. init racer directories
   for d in ${RACERDIRS}; do
   is_mounted $d || continue

RDIRS="$RDIRS $d/racer"
mkdir -p $d/racer

1. lfs setstripe $d/racer -c -1
   if [ $MDSCOUNT -ge 2 ]; then
   for i in $(seq $((MDSCOUNT - 1))); do
   RDIRS="$RDIRS $d/racer$i"
   if [ ! -e $d/racer$i ]; then
   $LFS mkdir -i $i $d/racer$i ||
   error "lfs mkdir $i failed"
   fi
   done
   fi
   done

1. 2. racer start
   local rpids=""
   for rdir in $RDIRS; do
   do_nodes $clients "DURATION=$DURATION MDSCOUNT=$MDSCOUNT \
   $racer $rdir $NUM_RACER_THREADS" &
   pid=$!
   rpids="$rpids $pid"
   done
   for pid in $rpids; do
   wait $pid
   rc=$?
   echo "pid=$pid rc=$rc"
   if [ $rc != 0 ]; then
   rrc=$((rrc + 1))
   fi
   done
2. 8. Replicate the changes to $TGT and TGT2
   $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
   -D $LRSYNC_LOG $EXTRA_FLAGS

1. 9. check difference
   check_diff $DIR $TGT
   check_diff $DIR $TGT2
   echo "check difference on target dir"
   sleep 120
   fini_changelog
   cleanup_src_tgt
   return 0
   }
   run_test 13 "lustre_rsync, use racer test suite"
   It will cause lustre_rsync run in a endless loop, and never come out.

Comments

[Peter Jones]

Landed for 2.7

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/13545/
Subject: LU-6167 utils: fix bugs in lustre_sync
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: dcf2a82d148797b4ac204a65ec795cde141e1d3b

[Peter Jones]

Yang Sheng

Could you please take care of this patch?

Thanks

Peter

[wu libin (Inactive)]

I think this patch can solve this problem, but really not sure the root cause.
http://review.whamcloud.com/#/c/13545/

[Gerrit Updater]

Wu Libin ([e-mail]) uploaded a new patch: http://review.whamcloud.com/13545
Subject: LU-6167 utils: fix bugs in lustre_sync
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: bb7871a24da289df3b0755508c3f307b21fdbdb0

[wu libin (Inactive)]

also core dump, the stack is like:
bt
#0 0x0000003566232925 in raise () from /lib64/libc.so.6
#1 0x0000003566234105 in abort () from /lib64/libc.so.6
#2 0x0000003566270837 in __libc_message () from /lib64/libc.so.6
#3 0x0000003566276166 in malloc_printerr () from /lib64/libc.so.6
#4 0x0000003566278f81 in _int_free () from /lib64/libc.so.6
#5 0x0000000000404375 in lr_cascade_move (fid=0x250a630 "[0x200000401:0x37a:0x0]", dest=0x2511d20 "/home/target/racer/11/3/11", info=0x24e0340) at lustre_rsync.c:682
#6 0x000000000040435a in lr_cascade_move (fid=0x250d960 "[0x200000400:0x366:0x0]", dest=0x2512e30 "/home/target/racer/11/3", info=0x24e0340) at lustre_rsync.c:677
#7 0x000000000040435a in lr_cascade_move (fid=0x24e0454 "[0x200000400:0x37a:0x0]", dest=0x24e1755 "/home/target/racer/11", info=0x24e0340) at lustre_rsync.c:677
#8 0x0000000000405369 in lr_move (info=0x24e0340) at lustre_rsync.c:964
#9 0x0000000000406eb8 in lr_replicate () at lustre_rsync.c:1552
#10 0x000000000040751b in main (argc=18, argv=<value optimized out>) at lustre_rsync.c:1776

the attached file is the test script i used.

[wu libin (Inactive)]

The script i used to test.