https://jira.whamcloud.com This file is an XML representation of an issue en-us 9.4.14 940014 05-12-2023 https://jira.whamcloud.com/browse/LU-10514 Lustre <div class="preformatted panel" style="border-width: 1px;"><div class="preformattedContent panelContent"> <pre>n:lustre-release# bash $LUSTRE/tests/llmount.sh ... n:lustre-release# time touch /mnt/lustre/f0 real    1m30.060s user    0m0.001s sys    0m0.004s n:lustre-release# time touch /mnt/lustre/f1 real 2m0.058s user 0m0.000s sys 0m0.004s </pre> </div></div> <p>libtoolizing <tt>lustre/utils</tt> creates a wrapper script for <tt>l_getidentity</tt> that doesn't work.</p> <p>When invoked the wrapped script prints the following to stderr</p> <div class="preformatted panel" style="border-width: 1px;"><div class="preformattedContent panelContent"> <pre>/root/lustre-release/lustre/utils/l_getidentity: line 151: ls: command not found /root/lustre-release/lustre/utils/l_getidentity: line 198: rm: command not found /root/lustre-release/lustre/utils/l_getidentity: line 212: rm: command not found /root/lustre-release/lustre/utils/l_getidentity: line 213: mv: command not found /root/lustre-release/lustre/utils/l_getidentity: line 214: rm: command not found /root/lustre-release/lustre/utils/l_getidentity: error: `/root/lustre-release/lustre/utils/.libs/lt-l_getidentity' does not exist This script is just a wrapper for lt-l_getidentity. See the libtool documentation for more information. </pre> </div></div> <p>But this doesn't go anywhere because stderr is to connected to anything when l_getidentity is run.</p> <p><tt>l_getidentity</tt> should not depend on liblustreapi. We should factor out whatever it needs in separate .c files and add them to l_getidentity dependencies.</p> <p>Also why do it take 2 minutes to for the operation to complete. It seems like we're not handling failure from the identity downcall very well. It sets stuck at:</p> <div class="preformatted panel" style="border-width: 1px;"><div class="preformattedContent panelContent"> <pre>n:~# stack1 mdt 8856 mdt00_003 [&lt;ffffffffc0921e80&gt;] upcall_cache_get_entry+0x1d0/0x8f0 [obdclass] [&lt;ffffffffc10120c7&gt;] mdt_identity_get+0x17/0x50 [mdt] [&lt;ffffffffc0ff36eb&gt;] old_init_ucred_common+0xdb/0x290 [mdt] [&lt;ffffffffc0ff39c7&gt;] old_init_ucred+0x127/0x240 [mdt] [&lt;ffffffffc0ff5405&gt;] mdt_init_ucred_intent_getattr+0x85/0xa0 [mdt] [&lt;ffffffffc0ff04f5&gt;] mdt_intent_getattr+0xc5/0x470 [mdt] [&lt;ffffffffc0fe60b2&gt;] mdt_intent_opc+0x442/0xad0 [mdt] [&lt;ffffffffc0fedc73&gt;] mdt_intent_policy+0x1a3/0x360 [mdt] [&lt;ffffffffc0d042fa&gt;] ldlm_lock_enqueue+0x38a/0x970 [ptlrpc] [&lt;ffffffffc0d2da33&gt;] ldlm_handle_enqueue0+0x8f3/0x1400 [ptlrpc] [&lt;ffffffffc0db3752&gt;] tgt_enqueue+0x62/0x210 [ptlrpc] [&lt;ffffffffc0dbb965&gt;] tgt_request_handle+0x925/0x13b0 [ptlrpc] [&lt;ffffffffc0d5fc7e&gt;] ptlrpc_server_handle_request+0x24e/0xab0 [ptlrpc] [&lt;ffffffffc0d63422&gt;] ptlrpc_main+0xa92/0x1e40 [ptlrpc] [&lt;ffffffff810b252f&gt;] kthread+0xcf/0xe0 [&lt;ffffffff816b8798&gt;] ret_from_fork+0x58/0x90 [&lt;ffffffffffffffff&gt;] 0xffffffffffffffff </pre> </div></div> <p>It's sleeping at <tt>left = schedule_timeout(expiry)</tt> in <tt>upcall_cache_get_entry()</tt>.</p> <p>BTW, libtool is terrible.</p> LU-10514

all metadata operations take 1+ minutes thanks to libtool's l_getidentity

Bug Blocker Resolved Fixed John Hammond John Hammond Mon, 15 Jan 2018 18:18:27 +0000 Sat, 28 Dec 2019 09:06:01 +0000 Thu, 25 Jan 2018 04:52:42 +0000 Lustre 2.11.0 Lustre 2.11.0 1 9 <p>Sorry, forgot to mention that the libtoolizing of <tt>lustre/utils/</tt> was done by <a href="https://review.whamcloud.com/30562" class="external-link" target="_blank" rel="nofollow noopener">https://review.whamcloud.com/30562</a> <a href="https://jira.whamcloud.com/browse/LU-5541" title="improve liblustreapi build" class="issue-link" data-issue-key="LU-5541"><del>LU-5541</del></a> build: move libcfs and liblustreapi over to libtool.</p> <p>None of the tests are showing this regression. This looks like a corner case. What are the exact steps you did to reproduce this problem?</p> <p>/usr/src/lustre-2.10.56/lustre/utils$ ./l_getidentity</p> <p>usage: l_getidentity {-d|mdtname} {uid}<br/> Normally invoked as an upcall from Lustre, set via:<br/> lctl set_param mdt.${mdtname}.identity_upcall={path to upcall}<br/> -d: debug, print values to stdout instead of Lustre</p> <p>/usr/src/lustre-2.10.56/lustre/utils$ ldd ./l_getidentity<br/> not a dynamic executable</p> <p>Well the other option besides libtool is cmake <img class="emoticon" src="https://jira.whamcloud.com/images/icons/emoticons/smile.png" height="16" width="16" align="absmiddle" alt="" border="0"/></p> <p>I wouldn't consider this a blocker since under normal conditions things do work.</p> <p>James,</p> <p>Anything that breaks when <tt>./lustre/tests/llmount.sh</tt> is run from the lustre-release directory should be a blocker. Many developers depend on this working to reduce turn around time for testing locally.</p> <p>John L. Hammond (john.hammond@intel.com) uploaded a new patch: <a href="https://review.whamcloud.com/30872" class="external-link" target="_blank" rel="nofollow noopener">https://review.whamcloud.com/30872</a><br/> Subject: <a href="https://jira.whamcloud.com/browse/LU-10514" title="all metadata operations take 1+ minutes thanks to libtool&#39;s l_getidentity" class="issue-link" data-issue-key="LU-10514"><del>LU-10514</del></a> utils: statically link l_getidentity with libcfs.a<br/> Project: fs/lustre-release<br/> Branch: master<br/> Current Patch Set: 1<br/> Commit: e53f0067d9f8b4ab2d7bf8df75b5fb02aae03727</p> <p>You could also do l_getidentity_LDFLAGS := -static. We should look at making l_getidentity independent of libcfs.a as well. Currently we have to have libcfs.a due to need for libcfs_nid2str() used by /etc/perm.conf support. Do we need /etc/perm.conf anymore since we have nodemap which is way better. Second libcfs.a use is cfs_get_param() but we could just test for /proc and /sys indentity_info file and write to it directly.</p> <p>James, getting rid of <tt>/etc/perm.conf</tt> would be great. As you write, we have nodemap already. That was part of the goal of the <a href="https://jira.whamcloud.com/browse/LU-7533" title="remove obsolete idmap functionality" class="issue-link" data-issue-key="LU-7533"><del>LU-7533</del></a> cleanup.</p> <p>Oleg Drokin (oleg.drokin@intel.com) merged in patch <a href="https://review.whamcloud.com/30872/" class="external-link" target="_blank" rel="nofollow noopener">https://review.whamcloud.com/30872/</a><br/> Subject: <a href="https://jira.whamcloud.com/browse/LU-10514" title="all metadata operations take 1+ minutes thanks to libtool&#39;s l_getidentity" class="issue-link" data-issue-key="LU-10514"><del>LU-10514</del></a> utils: statically link l_getidentity with libcfs.a<br/> Project: fs/lustre-release<br/> Branch: master<br/> Current Patch Set: <br/> Commit: 14e8e8e8a6b1bfe370e3e1f9fef6e37ccfc23290</p> <p>Landed for 2.11</p> Duplicate LU-10545 Related LU-5541 LU-10627 Development Rank 1|hzzr4v: Rank (Obsolete) 9223372036854775807 Severity