--- /dev/null
+<?xml version='1.0' encoding='UTF-8'?><chapter xmlns="http://docbook.org/ns/docbook" xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en-US" xml:id="lsom"
+ condition="l212">
+ <title xml:id="lsom.title">Lazy Size on MDT (LSoM)</title>
+ <para>This chapter describes Lazy Size on MDT (LSoM).</para>
+ <section xml:id="dbdoclet.lsomintro">
+ <title>
+ <indexterm>
+ <primary>lsom</primary>
+ <secondary>intro</secondary>
+ </indexterm>Introduction to Lazy Size on MDT (LSoM)</title>
+ <para>In the Lustre file system, MDSs store the ctime, mtime, owner,
+ and other file attributes. The OSSs store the size and number of
+ blocks used for each file. To obtain the correct file size, the client
+ must contact each OST that the file is stored across, which means
+ multiple RPCs to get the size and blocks for a file when a file is
+ striped over multiple OSTs. The Lazy Size on MDT (LSoM) feature stores
+ the file size on the MDS and avoids the need to fetch the file size
+ from the OST(s) in cases where the application understands that the
+ size may not be accurate. Lazy means there is no guarantee of the
+ accuracy of the attributes stored on the MDS.</para>
+ <para>Since many Lustre installations use SSD for MDT storage, the
+ motivation for the LSoM work is to speed up the time it takes to get
+ the size of a file from the Lustre file system by storing that data on
+ the MDTs. We expect this feature to be initially used by Lustre policy
+ engines that scan the backend MDT storage, make decisions based on broad
+ size categories, and do not depend on a totally accurate file size.
+ Examples include Lester, Robinhood, Zester, and DDN’s Lustre Integrated
+ Policy Engine (LiPE). Future improvements will allow the LSoM data to be
+ accessed by tools such as <literal>lfs find</literal>.</para>
+ </section>
+ <section xml:id="dbdoclet.enablelsom">
+ <title><indexterm><primary>lsom</primary>
+ <secondary>enablelsom</secondary></indexterm>Enable LSoM</title>
+ <para> LSoM is always enabled and nothing needs to be done to enable the
+ feature for fetching the LSoM data when scanning the MDT inodes with a
+ policy engine. It is also possible to access the LSoM data on the client
+ via the <literal>lfs getsom</literal> command. Because the LSoM data is
+ currently accessed on the client via the xattr interface, the
+ <literal>xattr_cache</literal> will cache the file size and block count on
+ the client as long as the inode is cached. In most cases this is
+ desirable, since it improves access to the LSoM data. However, it also
+ means that the LSoM data may be stale if the file size is changed after the
+ xattr is first accessed or if the xattr is accessed shortly after the file
+ is first created.</para>
+ <para>If it is necessary to access up-to-date LSoM data that has gone
+ stale, it is possible to flush the xattr cache from the client by
+ cancelling the MDC locks via
+ <literal>ldlm set_param ldlm.namespaces.*mdc*.lru_size=clear</literal>.
+ Otherwise, the file attributes will be dropped from the client cache if
+ the file has not been accessed before the LDLM lock timeout. The timeout
+ is stored via
+ <literal>lctl get_param ldlm.namespaces.*mdc*.lru_max_age</literal>.</para>
+ <para>If repeated access to LSoM attributes for files that are recently
+ created or frequently modified from a specific client, such as an HSM agent
+ node, it is possible to disable xattr caching on a client via:
+ <literal>lctl set_param llite.*.xattr_cache=0</literal>. This may cause
+ extra overhead when accessing files, and is not recommended for normal
+ usage.</para>
+ </section>
+ <section xml:id="dbdoclet.usercmds">
+ <title><indexterm><primary>lsom</primary>
+ <secondary>usercommands</secondary></indexterm>User Commands</title>
+ <para>Lustre provides the <literal>lfs getsom</literal> command to list
+ file attributes that are stored on the MDT.</para>
+ <para>The <literal>llsom_sync</literal> command allows the user to sync
+ the file attributes on the MDT with the valid/up-to-date data on the
+ OSTs. <literal>llsom_sync</literal> is called on the client with the
+ Lustre file system mount point. <literal>llsom_sync</literal> uses Lustre
+ MDS changelogs and, thus, a changelog user must be registered to use this
+ utility.</para>
+ <section xml:id="dbdoclet.lfsgetsom">
+ <title><indexterm><primary>lsom</primary>
+ <secondary>lfsgetsom</secondary></indexterm>lfs getsom for LSoM data
+ </title>
+ <para>The <literal>lfs getsom</literal> command lists file attributes
+ that are stored on the MDT. <literal>lfs getsom</literal> is called
+ with the full path and file name for a file on the Lustre file
+ system. If no flags are used, then all file attributes stored on the
+ MDS will be shown.</para>
+ <section><title>lfs getsom Command</title>
+ <para><screen>lfs getsom [-s] [-b] [-f] <filename></screen>
+ The various <literal>lfs getsom</literal> options are listed and
+ described below.</para>
+ <informaltable frame="all">
+ <tgroup cols="2">
+ <colspec colname="c1" colwidth="3*" />
+ <colspec colname="c2" colwidth="7*" />
+ <thead>
+ <row>
+ <entry>
+ <para>
+ <emphasis role="bold">Option</emphasis>
+ </para>
+ </entry>
+ <entry>
+ <para>
+ <emphasis role="bold">Description</emphasis>
+ </para>
+ </entry>
+ </row>
+ </thead>
+ <tbody>
+ <row>
+ <entry>
+ <para>
+ <literal>-s</literal>
+ </para>
+ </entry>
+ <entry>
+ <para>Only show the size value of the LSoM data for a given
+ file. This is an optional flag</para>
+ </entry>
+ </row>
+ <row>
+ <entry>
+ <para>
+ <literal>-b</literal>
+ </para>
+ </entry>
+ <entry>
+ <para>Only show the blocks value of the LSoM data for a
+ given file. This is an optional flag</para>
+ </entry>
+ </row>
+ <row>
+ <entry>
+ <para>
+ <literal>-f</literal>
+ </para>
+ </entry>
+ <entry>
+ <para>Only show the flag value of the LSoM data for a given
+ file. This is an optional flag. Valid flags are:</para>
+ <para>SOM_FL_UNKNOWN = 0x0000 - Unknown or no SoM data,
+ must get size from OSTs.</para>
+ <para>SOM_FL_STRICT = 0x0001 - Known strictly correct, FLR
+ file (SoM guaranteed)</para>
+ <para>SOM_FL_STALE = 0x0002 - Known stale -was right at
+ some point in the past, but it is known (or likely) to be
+ incorrect now (e.g. opened for write)</para>
+ <para>SOM_FL_LAZY= 0x0004 - Approximate, may never have
+ been strictly correct, need to sync SOM data to achieve
+ eventual consistency.</para>
+ </entry>
+ </row>
+ </tbody>
+ </tgroup>
+ </informaltable>
+ </section>
+ </section>
+ <section><title>Syncing LSoM data</title>
+ <para>The <literal>llsom_sync</literal> command allows the user to sync
+ the file attributes on the MDT with the valid/up-to-date data on the
+ OSTs. <literal>llsom_sync</literal> is called on the client with the
+ client mount point for the Lustre file system.
+ <literal>llsom_sync</literal> uses Lustre MDS changelogs and, thus, a
+ changelog user must be registered to use this utility.</para>
+ <section><title>llsom_sync Command</title>
+ <screen>llsom_sync --mdt|-m <mdt> --user|-u <user_id>
+ [--daemonize|-d] [--verbose|-v] [--interval|-i] [--min-age|-a]
+ [--max-cache|-c] [--sync|-s] <lustre_mount_point></screen>
+ <para>The various <literal>llsom_sync</literal> options are
+ listed and described below.</para>
+ <informaltable frame="all">
+ <tgroup cols="2">
+ <colspec colname="c1" colwidth="3*" />
+ <colspec colname="c2" colwidth="7*" />
+ <thead>
+ <row>
+ <entry>
+ <para><emphasis role="bold">Option</emphasis></para>
+ </entry>
+ <entry>
+ <para><emphasis role="bold">Description</emphasis></para>
+ </entry>
+ </row>
+ </thead>
+ <tbody>
+ <row>
+ <entry>
+ <para><literal>--mdt | -m <mdt></literal></para>
+ </entry>
+ <entry>
+ <para>The metadata device which need to be synced the LSoM
+ xattr of files. A changelog user must be registered for
+ this device.Required flag.</para>
+ </entry>
+ </row>
+ <row>
+ <entry>
+ <para><literal>--user | -u <user_id></literal></para>
+ </entry>
+ <entry>
+ <para>The changelog user id for the MDT device. Required
+ flag.</para>
+ </entry>
+ </row>
+ <row>
+ <entry>
+ <para><literal>--daemonize | -d</literal></para>
+ </entry>
+ <entry>
+ <para>Optional flag to “daemonize” the program. In daemon
+ mode, the utility will scan, process the changelog records
+ and sync the LSoM xattr for files periodically.</para>
+ </entry>
+ </row>
+ <row>
+ <entry>
+ <para><literal>--verbose | -v</literal></para>
+ </entry>
+ <entry>
+ <para>Optional flag to produce verbose output.</para>
+ </entry>
+ </row>
+ <row>
+ <entry>
+ <para><literal>--interval | -i</literal></para>
+ </entry>
+ <entry>
+ <para>Optional flag for the time interval to scan the
+ Lustre changelog and process the log record in daemon
+ mode.</para>
+ </entry>
+ </row>
+ <row>
+ <entry>
+ <para><literal>--min-age | -a</literal></para>
+ </entry>
+ <entry>
+ <para>Optional flag for the time that
+ <literal>llsom_sync</literal> tool will not try to sync the
+ LSoM data for any files closed less than this many seconds
+ old. The default min-age value is 600s(10 minutes).</para>
+ </entry>
+ </row>
+ <row>
+ <entry>
+ <para><literal>--max-cache | -c</literal></para>
+ </entry>
+ <entry>
+ <para>Optional flag for the total memory used for the FID
+ cache which can be with a suffix [KkGgMm].The default
+ max-cache value is 256MB. For the parameter value < 100,
+ it is taken as the percentage of total memory size used for
+ the FID cache instead of the cache size.</para>
+ </entry>
+ </row>
+ <row>
+ <entry>
+ <para><literal>--sync | -s</literal></para>
+ </entry>
+ <entry>
+ <para>Optional flag to sync file data to make the dirty
+ data out of cache to ensure the blocks count is correct
+ when update the file LSoM xattr. This option could hurt
+ server performance significantly if thousands of fsync
+ requests are sent.</para>
+ </entry>
+ </row>
+ </tbody>
+ </tgroup>
+ </informaltable>
+ </section>
+ </section>
+ </section>
+</chapter>
git://git.whamcloud.com / doc / manual.git / commitdiff