Link: https://openstack.nimeyo.com/60317/?show=60317#q60317<br />
From: lukas <lukas.l.rupprecht@gmail.com><br />
<br />
<p>I'm using the <code>hadoop-swiftfs</code> from the <code>sahara-extra</code> project to read input from Swift during my MapReduce jobs. The hadoop version I am using is 2.6.0 and this already came with the <code>hadoop-openstack</code> additions.</p>
<p>I set <code>fs.swift.service.sahara.location-aware</code> to <code>true</code> in my <code>core-site.xml</code> and the location of input splits is correctly identified. However, it still seems that reads from MapReduce go through the Swift proxy as the proxy's network is always fully utilised.</p>
<p>Is there a way of directly reading from the object servers and avoid sending everything to the proxy first?</p>