<div dir="ltr"><p class="MsoNormal">Hi all, </p><p class="MsoNormal"><br></p><p class="MsoNormal">I have a three-node cluster containing a durable
queue that is mirrored across all three nodes. Persistent messages are being published to
an exchange routed to this queue (delivery mode = 2). I have intentionally disabled all consumers of this queue in
order to see how memory and disk usage are affected. I'd expect the memory
usage to remain fairly low, while the disk usage increases (since all the
messages for this queue should be persisted to disk). </p>
<p class="MsoNormal"> </p>
<p class="MsoNormal">However, after letting several hundred thousand messages
accumulate in this queue, I was surprised to notice that the memory usage on
the primary node for this queue is significantly higher (roughly 3x) than the
other two nodes in the cluster. Shouldn't the memory usage be roughly
equivalent, since the queue is mirrored across all three nodes in the cluster?</p>
<p class="MsoNormal"> </p>
<p class="MsoNormal">Additionally, when I view the "Memory details" for
the primary node for this queue, the majority of the memory is occupied by the
"Binaries" category of the breakdown. The documentation page states
that "In-memory message bodies show up" in this category. But
shouldn't the message bodies all be on the disk, and not in memory and affecting the report in this way? </p><p class="MsoNormal"><br></p><p class="MsoNormal">Any clarification regarding these issues would be greatly appreciated. Thank you.</p>
</div>