<html>
  <head>
    <meta content="text/html; charset=ISO-8859-1"
      http-equiv="Content-Type">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <div class="moz-cite-prefix">Hi Hugo,<br>
          I know the tombstone mechanism. In my opinion after the
      reclaim time, the object of xxx.tombstone will be deleted at all.
      Is that right? Maybe I misunderstand the doc :( ...<br>
          We try to "colddown" the swift system ( just wait for the
      reclaiming) and test, but the result is not satisfying.<br>
      <br>
      Thanks,<br>
      Jonathan Lu<br>
      <br>
      On 2013/6/18 11:04, Kuo Hugo wrote:<br>
    </div>
    <blockquote
cite="mid:CA++_uhttqs++A7Rj=eSwgVDjDu5uoZTZ8iX1=0yfmz_255KGaw@mail.gmail.com"
      type="cite">
      <div dir="ltr">Hi Jonathan , 
        <div><br>
        </div>
        <div style="">How did you perform "<span
            style="font-family:arial,sans-serif;font-size:14px"><font
              color="#ff9900">delete all the objects in the storage</font>"
            ?  Those deleted objects still consume inodes in tombstone
            status until the reclaim time. </span></div>
        <div style=""><font face="arial, sans-serif"><span
              style="font-size:14px">Would you mind to compare the
              result of $> </span></font><span
            style="font-family:arial,sans-serif;font-size:14px">sudo cat
            /proc/slabinfo | grep xfs   ,  before/after set the
            vfs_cache_pressure   </span></div>
        <div style=""><span
            style="font-family:arial,sans-serif;font-size:14px"><br>
          </span></div>
        <div style=""><font face="arial, sans-serif">
            <div style="font-size:14px"><font color="#f1c232">spongebob</font>@<font
                color="#0000ff">patrick1:~$</font> sudo cat
              /proc/slabinfo | grep xfs</div>
            <div style="font-size:14px"><span
                style="font-family:arial;font-size:small"><font
                  color="#999999">xfs_ili                </font><font
                  color="#ff0000">70153  70182</font><font
                  color="#999999">    216   18    1 : tunables    0    0
                     0 : slabdata   3899   3899      0</font></span><br>
            </div>
          </font><font color="#999999">xfs_inode         </font><font
            color="#ff0000">169738 170208</font><font color="#999999">  
            1024   16    4 : tunables    0    0    0 : slabdata  10638
             10638      0<br>
            xfs_efd_item          60     60    400   20    2 : tunables
               0    0    0 : slabdata      3      3      0<br>
            xfs_buf_item         234    234    224   18    1 : tunables
               0    0    0 : slabdata     13     13      0<br>
            xfs_trans             28     28    280   14    1 : tunables
               0    0    0 : slabdata      2      2      0<br>
            xfs_da_state          32     32    488   16    2 : tunables
               0    0    0 : slabdata      2      2      0<br>
            xfs_btree_cur         38     38    208   19    1 : tunables
               0    0    0 : slabdata      2      2      0<br>
            xfs_log_ticket        40     40    200   20    1 : tunables
               0    0    0 : slabdata      2      2      0</font></div>
        <div style=""><span
            style="font-family:arial,sans-serif;font-size:14px"><br>
          </span></div>
        <div style=""><br>
        </div>
        <div style=""><font face="arial, sans-serif"><span
              style="font-size:14px">Hi Robert, </span></font></div>
        <div style=""><font face="arial, sans-serif"><span
              style="font-size:14px">The performance degradation still
              there even only main swift workers are running in storage
              node. ( stop replicator/updater/auditor ). In my knowing. </span></font></div>
        <div style=""><font face="arial, sans-serif"><span
              style="font-size:14px">I'll check xs_dir_lookup and
              xs_ig_missed here. Thanks</span></font></div>
        <div style=""><font face="arial, sans-serif"><span
              style="font-size:14px"><br>
            </span></font></div>
        <div style=""><font face="arial, sans-serif"><span
              style="font-size:14px"><br>
            </span></font></div>
        <div style=""><font face="arial, sans-serif"><span
              style="font-size:14px"><br>
            </span></font></div>
        <div style="">
          <span style="font-family:arial,sans-serif;font-size:14px"><br>
          </span></div>
        <div style=""><span
            style="font-family:arial,sans-serif;font-size:14px"><br>
          </span></div>
        <div style=""><span
            style="font-family:arial,sans-serif;font-size:14px"> </span></div>
      </div>
      <div class="gmail_extra"><br clear="all">
        <div>
          <div dir="ltr">
            <div>+Hugo Kuo+</div>
            <div><a moz-do-not-send="true"
                href="mailto:hugo@swiftstack.com" target="_blank">hugo@swiftstack.com</a><br>
            </div>
            <div><a moz-do-not-send="true"
                href="mailto:tonytkdk@gmail.com" target="_blank">tonytkdk@gmail.com<br>
              </a></div>
            <div>+886 935004793<br>
            </div>
          </div>
        </div>
        <br>
        <br>
        <div class="gmail_quote">2013/6/18 Jonathan Lu <span dir="ltr"><<a
              moz-do-not-send="true" href="mailto:jojokururu@gmail.com"
              target="_blank">jojokururu@gmail.com</a>></span><br>
          <blockquote class="gmail_quote" style="margin:0 0 0
            .8ex;border-left:1px #ccc solid;padding-left:1ex">
            <div class="im">On 2013/6/17 18:59, Robert van Leeuwen
              wrote:<br>
              <blockquote class="gmail_quote" style="margin:0 0 0
                .8ex;border-left:1px #ccc solid;padding-left:1ex">
                <blockquote class="gmail_quote" style="margin:0 0 0
                  .8ex;border-left:1px #ccc solid;padding-left:1ex">
                  I'm facing the issue about the performance
                  degradation, and once I glanced that changing the
                  value in /proc/sys<br>
                  /vm/vfs_cache_pressure will do a favour.<br>
                  Can anyone explain to me whether and why it is useful?<br>
                </blockquote>
                Hi,<br>
                <br>
                When this is set to a lower value the kernel will try to
                keep the inode/dentry cache longer in memory.<br>
                Since the swift replicator is scanning the filesystem
                continuously it will eat up a lot of iops if those are
                not in memory.<br>
                <br>
                To see if a lot of cache misses are happening, for xfs,
                you can look at xs_dir_lookup and xs_ig_missed.<br>
                ( look at <a moz-do-not-send="true"
                  href="http://xfs.org/index.php/Runtime_Stats"
                  target="_blank">http://xfs.org/index.php/Runtime_Stats</a>
                )<br>
                <br>
                We greatly benefited from setting this to a low value
                but we have quite a lot of files on a node ( 30 million)<br>
                Note that setting this to zero will result in the OOM
                killer killing the machine sooner or later.<br>
                (especially if files are moved around due to a cluster
                change ;)<br>
                <br>
                Cheers,<br>
                Robert van Leeuwen<br>
              </blockquote>
              <br>
            </div>
            Hi,<br>
                We set this to a low value(20) and the performance is
            better than before. It seems quite useful.<br>
            <br>
                According to your description, this issue is related
            with the object quantity in the storage. We delete all the
            objects in the storage but it doesn't help anything. The
            only method to recover is to format and re-mount the storage
            node. We try to install swift on different environment but
            this degradation problem seems to be an inevitable one.<br>
            <br>
            Cheers,<br>
            Jonathan Lu
            <div class="HOEnZb">
              <div class="h5"><br>
                <br>
                _______________________________________________<br>
                Mailing list: <a moz-do-not-send="true"
                  href="https://launchpad.net/%7Eopenstack"
                  target="_blank">https://launchpad.net/~openstack</a><br>
                Post to     : <a moz-do-not-send="true"
                  href="mailto:openstack@lists.launchpad.net"
                  target="_blank">openstack@lists.launchpad.net</a><br>
                Unsubscribe : <a moz-do-not-send="true"
                  href="https://launchpad.net/%7Eopenstack"
                  target="_blank">https://launchpad.net/~openstack</a><br>
                More help   : <a moz-do-not-send="true"
                  href="https://help.launchpad.net/ListHelp"
                  target="_blank">https://help.launchpad.net/ListHelp</a><br>
              </div>
            </div>
          </blockquote>
        </div>
        <br>
      </div>
    </blockquote>
    <br>
  </body>
</html>