<html>
  <head>
    <meta content="text/html; charset=ISO-8859-1"
      http-equiv="Content-Type">
  </head>
  <body bgcolor="#FFFFFF" text="#000000">
    <br>
    <blockquote
cite="mid:CA+nfvwTXBpvRrpjLE03M-3BmjG4mMerOJMZ2ua-hsTOCkAesPQ@mail.gmail.com"
      type="cite">
      <div dir="ltr">
        <div class="gmail_extra">
          <div class="gmail_quote">
            <blockquote class="gmail_quote" style="margin:0px 0px 0px
0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex">Investigation:<br>
              ------------<br>
              When I looked at UNICAST3, I saw a lot of missing messages
              on the<br>
              receive side and unacked messages on the send side. This
              caused me to<br>
              look into the (mainly OOB) thread pools and - voila -
              maxed out !<br>
              <br>
              I learned from Pedro that the Infinispan internal thread
              pool (with a<br>
              default of 32 threads) can be configured, so I increased
              it to 300 and<br>
              increased the OOB pools as well.<br>
              <br>
              This mitigated the problem somewhat, but when I increased
              the requester<br>
              threads to 100, I had the same problem again. Apparently,
              the Infinispan<br>
              internal thread pool uses a rejection policy of "run" and
              thus uses the<br>
              JGroups (OOB) thread when exhausted.<br>
            </blockquote>
            <div><br>
            </div>
            <div>We can't use another rejection policy in the remote
              executor because the message won't be re-delivered by
              JGroups, and we can't use a queue either.<br>
            </div>
          </div>
        </div>
      </div>
    </blockquote>
    <br>
    Can't we just send response "Node is busy" and cancel the operation?
    (at least in cases where this is possible - we can't do that safely
    for CommitCommand, but usually it could be doable, right?) And
    what's the problem with queues, besides that these can grow out of
    memory?<br>
    <br>
    Radim<br>
    <br>
    <pre class="moz-signature" cols="72">-- 
Radim Vansa <a class="moz-txt-link-rfc2396E" href="mailto:rvansa@redhat.com">&lt;rvansa@redhat.com&gt;</a>
JBoss DataGrid QA
</pre>
  </body>
</html>