[LU-10632] recovery-small test 26a fails with ‘client not evicted from OST’

Duplicate — Sat, 4 Jun 2022 12:48:00 +0000

recovery-small test_26a fails with the error

recovery-small test_26a: @@@@@@ FAIL: client not evicted from OST

The following output comes from the failure at https://testing.hpdd.intel.com/test_sets/fe065d76-0baa-11e8-a7cd-52540065bddc

Looking at the test_log, we see that there is a problem getting osc.*.state parameter on the client

CMD: trevis-6vm6.trevis.hpdd.intel.com /usr/sbin/lctl get_param osc.lustre-OST0000-osc-ffff880039c89800.state
/usr/lib64/lustre/tests/test-framework.sh: line 8496: ((: > 1517964926: syntax error: operand expected (error token is "> 1517964926")

Test 26a calls the check_clients_evicted() routine to get the time of eviction and check it against the “before” time. If you can’t get the time the client osc was evicted, then you can’t compare it to the times and the test will fail.

8482 # check that clients "oscs" was evicted after "before"
8483 check_clients_evicted() {
8484         local before=$1
8485         shift
8486         local oscs=${@}
8487         local osc
8488         local rc=0
8489 
8490         for osc in $oscs; do
8491                 ((rc++))
8492                 echo "Check state for $osc"
8493                 local evicted=$(do_facet client $LCTL get_param osc.$osc.state |
8494                         tail -n 3 | awk -F"[ [,]" \
8495                         '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
8496                 if (($? == 0)) && (($evicted > $before)); then
8497                         echo "$osc is evicted at $evicted"
8498                         ((rc--))
8499                 fi
8500         done
8501 
8502         [ $rc -eq 0 ] || error "client not evicted from OST"
8503 }

you see that getting no value for $evicted, then we don’t decrement the return code and the test fails.

There’s nothing obviously wrong in the console and dmesg logs except in the client console log, we do see some LustreErrors

[12802.505434] Lustre: DEBUG MARKER: df
[12805.745429] Lustre: Evicted from MGS (at 10.9.4.63@tcp) after server handle changed from 0x6728fd52bd37c36a to 0x6728fd52bd37c913
[12805.745586] LustreError: 14144:0:(file.c:4097:ll_inode_revalidate_fini()) lustre: revalidate FID [0x200000007:0x1:0x0] error: rc = -5
[12805.746256] LustreError: 14144:0:(lmv_obd.c:1387:lmv_statfs()) can't stat MDS #0 (lustre-MDT0000-mdc-ffff880039c89800), error -108
[12805.746260] LustreError: 14144:0:(llite_lib.c:1785:ll_statfs_internal()) md_statfs fails: rc = -108
[12805.760747] Lustre: DEBUG MARKER: /usr/sbin/lctl get_param osc.lustre-OST0000-osc-ffff880039c89800.state

Here are logs for a few other recovery-small test 26a failures:
review-dne-part-1: https://testing.hpdd.intel.com/test_sets/4ec62074-f473-11e7-8c43-52540065bddc
failover: https://testing.hpdd.intel.com/test_sets/bd9510da-f6cd-11e7-bd00-52540065bddc
failover: https://testing.hpdd.intel.com/test_sets/01049780-ffef-11e7-a6ad-52540065bddc
failover: https://testing.hpdd.intel.com/test_sets/4b4b3144-06d6-11e8-a7cd-52540065bddc

Whamcloud Community JIRA

[LU-10632] recovery-small test 26a fails with ‘client not evicted from OST’