[LU-11619] sanity test 155 fail for ZFS

• Status: Open
• Project: Lustre
• Affects Version/s: Lustre 2.12.0
• Type: Bug
• Priority: Major
• Reporter: James A Simmons
• Assignee: Sarah Liu
• Resolution: Unresolved

Environment

lustre 2.12 on clients and servers with ZFS backend.

• Severity: 3

Description

While doing testing using ZFS I see sanity test 155 failing. The reason for this is that parametes like osd-..writethrough_cache_enable don't exist for ZFS. Before it silently failed since it was accessing a broken symlink but now an error is reported back to user land if you attempt to set these parameters on ZFS.

Comments

[Peter Jones]

jamesanunez

Are we seeing this in our testing?

Peter

[James Nunez (Inactive)]

Yes, we do see this error, but the test does not fail because of it since we are not checking that the call to 'lctl get/set_param' succeeds. Looking at a recent full test session for b2_10 at https://testing.whamcloud.com/test_sets/16603dd4-df08-11e8-a871-52540065bddc, we see

== sanity test 155a: Verify small file correctness: read cache:on write_cache:on ===================== 17:10:46 (1541092246)
CMD: trevis-11vm8 params=\$(/usr/sbin/lctl get_param obdfilter.*.writethrough_cache_enable);
			 [[ -z \"lustre-OST0000\" ]] && param= ||
			 param=\$(grep lustre-OST0000 <<< \"\$params\");
			 [[ -z \$param ]] && param=\"\$params\";
			 while read s; do echo ost1 \$s;
			 done <<< \"\$param\"
trevis-11vm8: error: get_param: param_path 'obdfilter/*/writethrough_cache_enable': No such file or directory

From sanity.sh

11874 save_writethrough() {
11875         local facets=$(get_facets OST)
11876 
11877         save_lustre_params $facets "osd-*.*.writethrough_cache_enable" > $1
11878 }
11879 
11880 test_155a() {
11881         [ $PARALLEL == "yes" ] && skip "skip parallel run"
11882 
11883         local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
11884 
11885         save_writethrough $p
11886 
11887         set_cache read on
11888         set_cache writethrough on
11889         test_155_small_load
11890         restore_lustre_params < $p
11891         rm -f $p
11892 }
11893 run_test 155a "Verify small file correctness: read cache:on write_cache:on"

[Andreas Dilger]

Alex, is there any reason we would want the ability to turn writethrough cache OFF for ZFS? AFAIK, this would completely kill ZFS performance and is undesirable. In that case, we should just skip tests that turn writethrough cache off for ZFS.

[Alex Zhuravlev]

adilgerwell, I guess it's impossible technically. even with ZIL it's still going though ARC, AFAIK.

[James Nunez (Inactive)]

As Andreas stated above, we should skip all tests that turn writethrough cache off for ZFS.

Since the writethrough_cache_enable parameter doesn't exist for OSSs with ZFS file systems, should we skip all tests that try to turn on, off or query writethrough_cache_enable?