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-10588 Lustre <p>Just wanted to report these kernel messages on the MDS seen shortly after having started lfsck_namespace and oi_scrub. I know they clearly lack some information but seem be related to lfsck.</p> <div class="code panel" style="border-width: 1px;"><div class="codeContent panelContent"> <pre class="code-java">Jan 30 14:11:36 oak-md1-s2 kernel: list passed to list_sort() too <span class="code-object">long</span> <span class="code-keyword">for</span> efficiency </pre> </div></div> <p>Stephane</p> 3.10.0-693.2.2.el7_lustre.pl1.x86_64 LU-10588

lfsck generates "kernel: list passed to list_sort() too long for efficiency"

Improvement Trivial Resolved Cannot Reproduce WC Triage Stephane Thiell Tue, 30 Jan 2018 22:37:06 +0000 Wed, 13 Feb 2019 08:03:13 +0000 Wed, 13 Feb 2019 08:03:08 +0000 0 5 <p>Stephane,<br/> There is no direct caller for <tt>list_sort</tt> in the Lustre tree, not any use of "sort" in LFSCK at all, so there is no way to know what is causing this message. </p> <p>One option would be to change, <tt>list_sort()</tt> to call <tt>WARN_ONCE()</tt>, like:</p> <div class="code panel" style="border-width: 1px;"><div class="codeContent panelContent"> <pre class="code-java"> <span class="code-keyword">if</span> (lev &gt; max_lev) { <span class="code-keyword">if</span> (WARN_ONCE(lev &gt;= AREAY_SIZE(part) - 1, <span class="code-quote">"list too <span class="code-object">long</span> <span class="code-keyword">for</span> efficiency (%d &gt;= %d)\n"</span>, lev, MAX_LIST_LENGTH_BITS)) lev--; max_lev = lev; } </pre> </div></div> <p>This will dump the stack trace of the thread that hits this problem. </p> <p>Ah thanks much Andreas! Will do when possible and update the ticket if I got something. This might take some time though.</p> <p>Stephane</p> <p>Fan Yong</p> <p>Anything else to suggest here?</p> <p>Peter</p> <p>In Lustre, the unique caller of list_sort() is __ldiskfs_es_shrink() that will be called by system under the case of system RAM pressure to release some RAM from extent status tree.</p> <p>Currently, only with the message "Jan 30 14:11:36 oak-md1-s2 kernel: list passed to list_sort() too long for efficiency", I can NOT say whether it is the __ldiskfs_es_shrink() triggered such message or not.</p> <p>But by default, Lustre only enables file extents for OST, NOT for MDT. But in this case, the message was printed on the MDS. That is strange. Means even if LFSCK caused too many inodes cached (that is controlled by system), the extent status tree on the MDT would be almost empty and should not cause list_sort() warning.</p> <p>Anyway, we need the stack trace to know who triggered the list_sort() warning.</p> <p><a href="https://jira.whamcloud.com/secure/ViewProfile.jspa?name=sthiell" class="user-hover" rel="sthiell">sthiell</a>,</p> <p>Have you got the stack trace for the list_sort() sponsor? Have you checked the RAM usage on MDT when list_sort() trouble happened?</p> <p>Hi nasf,</p> <p>Thanks for this useful information! We're indeed using ldiskfs. But this is currently low priority for us and as we're testing a patched kernel already for another important issue, we decided not to include any other change for now. Reproducibility, you know... <img class="emoticon" src="https://jira.whamcloud.com/images/icons/emoticons/smile.png" height="16" width="16" align="absmiddle" alt="" border="0"/></p> <p>When things get settled for us with 2.10.x, I'll patch the kernel to get more debugging from this warning as suggested by Andreas.</p> <p>Thanks again.</p> <p>Stephane</p> <p>Closing this old issue since there is no information, please re-open if more information becomes available.</p> Development Rank 1|hzzrxz: Rank (Obsolete) 9223372036854775807