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-3452 Lustre <p>1.8 client don't able to start with 2.4 now (and 2.3 if i correctly understand).</p> <div class="code panel" style="border-width: 1px;"><div class="codeContent panelContent"> <pre class="code-java">ustreError: 152-6: Ignoring deprecated mount option <span class="code-quote">'acl'</span>. Lustre: 5036:0:(obd_config.c:1127:class_config_llog_handler()) skipping <span class="code-quote">'lmv'</span> config: cmd=cf001,clilmv:lmv Lustre: 5036:0:(obd_config.c:875:class_process_config()) Ignoring unknown param jobid_var=procname_uid LustreError: 5036:0:(obd_config.c:1199:class_config_llog_handler()) Err -22 on cfg command: Lustre: cmd=cf00f 0:&lt;NULL&gt; 1:sys.jobid_var=procname_uid 2:procname_uid LustreError: 15b-f: MGC192.168.69.5@tcp: The configuration from log <span class="code-quote">'lustre-client'</span> failed (-22). Make sure <span class="code-keyword">this</span> client and the MGS are running compatible versions of Lustre. LustreError: 15c-8: MGC192.168.69.5@tcp: The configuration from log <span class="code-quote">'lustre-client'</span> failed (-22). This may be the result of communication errors between <span class="code-keyword">this</span> node and the MGS, a bad configuration, or other errors. See the syslog <span class="code-keyword">for</span> more information. </pre> </div></div> LU-3452

1.8 <> 2.4 interop broken due job id patch.

Bug Major Resolved Duplicate WC Triage Alexey Lyashkov Tue, 11 Jun 2013 12:16:50 +0000 Sat, 22 Nov 2014 00:32:32 +0000 Sat, 22 Nov 2014 00:32:32 +0000 Lustre 2.1.3 Lustre 1.8.x (1.8.0 - 1.8.5) Lustre 1.8.9 Lustre 2.1.4 Lustre 1.8.9 0 3 <p>This is only true if the JobID feature is enabled, which it isn't by default. The jobid feature is enabled by default by the test framework but will not be when the filesystem is formatted by a user, or during an upgrade. </p> <p>There is a patch for b1_8 and b2_1 to allow the client to skip unknown conf_param settings which should be applied to clients that need to interoperate with servers that have JobId enabled. There isn't a patch to backport JobId to those clients. </p> <p>Andreas, </p> <p>i know about it's patch, but it's broke an interoperability testing for now. and i'm understand about "why it's hit", my question more likely to be "why job_id parameter send to non 2.4 clients".<br/> Some customers have a lots a 1.8 or older 2.x clients without job id or ignore a invalid parameter patch, so it's should be fixed in lustre to have interoperability works.</p> <p>JobID is not a "required" feature for users (i.e. it won't break if it isn't enabled, and it can be fixed with a small patch in older clients).</p> <p>As with many new Lustre features, there is no easy way to add new fixes to older clients (e.g. DNE, large xattrs, HSM, etc), so those features should not be enabled by sysadmins if the clients do not support them, and JobID is not enabled by default. It is unfortunate that this interoperability problem was not found before 2.3.0 was released, but it is impossible for us to test every possible combination of features enabled and disabled with every combination of client version and server version, and in this case we didn't catch the problem in time.</p> <p>The conf_param setting in the MGS config log does not check for compatibility of llog entries between client versions. Adding the MGS understand Lustre feature compatibility is much more complex than having the client ignore configuration parameters that it doesn't understand. The client <em>should</em> have already been ignoring such unknown parameters, but there was a bug in that code (<a href="https://jira.whamcloud.com/browse/LU-2309" title="Uninteroperable conf_params not documented" class="issue-link" data-issue-key="LU-2309"><del>LU-2309</del></a> was fixed in 2.1.4 and 2.4.0, and landed to b1_8 though after the 1.8.9 release).</p> <p>If you can think of a simple way to have the MGS hide this config record from the client that might be useful. Clients already connect to the OSS and MDS with OBD_CONNECT_JOBSTATS, so that the client/server can decide what size RPC message they should use, but they don't pass this flag to the MGS. The difficulty is adding a check for OBD_CONNECT_JOBSTATS on the MGS to the clients now means that the JobID code would break for 2.3.0 and 2.4.0 clients as well. It <em>might</em> be enough to have the MGS check for OBD_CONNECT_FULL20, which would leave only the 2.1.0-2.1.3 clients unable to interoperate, but they are much more likely to be updated to a newer release that can ignore the conf_param errors than older 1.8 clients.</p> <p>Since a patch to fix this was landed to b1_8 (after 1.8.9) and for 2.1.4, and this problem is only hit if the sysadmin explicitly enables jobstats I'm going to mark this fixed.</p> Related LU-2309 Development Rank 1|hzvt1b: Rank (Obsolete) 8632 Severity