docs/build/plain-html/topics/impala_data_cache.html - impala - Git at Google

 <?xml version="1.0" encoding="UTF-8"?>
 <!DOCTYPE html
   PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
 <html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en" lang="en">
 <head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />

 <meta name="copyright" content="(C) Copyright 2025" />
 <meta name="DC.rights.owner" content="(C) Copyright 2025" />
 <meta name="DC.Type" content="concept" />
 <meta name="DC.Title" content="Data Cache for Remote Reads" />
 <meta name="DC.Relation" scheme="URI" content="../topics/impala_performance.html" />
 <meta name="prodname" content="Impala" />
 <meta name="version" content="Impala 3.4.x" />
 <meta name="DC.Format" content="XHTML" />
 <meta name="DC.Identifier" content="data_cache" />
 <link rel="stylesheet" type="text/css" href="../commonltr.css" />
 <title>Data Cache for Remote Reads</title>
 </head>
 <body id="data_cache">


   <h1 class="title topictitle1" id="ariaid-title1">Data Cache for Remote Reads</h1>


   <div class="body conbody">

     <p class="p">
       When Impala compute nodes and its storage are not co-located, the network bandwidth
       requirement goes up as the network traffic includes the data fetch as well as the
       shuffling exchange traffic of intermediate results.
     </p>


     <p class="p">
       To mitigate the pressure on the network, you can enable the compute nodes to cache the
       working set read from remote filesystems, such as, remote HDFS data node, S3, ABFS, ADLS.
     </p>


     <p class="p">
       To enable remote data cache, set the <code class="ph codeph">--data_cache</code> Impala Daemon start-up
       flag as below:
     </p>


 <pre class="pre codeblock"><code>--data_cache=<var class="keyword varname">dir1</var>,<var class="keyword varname">dir2</var>,<var class="keyword varname">dir3</var>,...:<var class="keyword varname">quota</var></code></pre>

     <p class="p">
       The flag is set to a list of directories, separated by <code class="ph codeph">,</code>, followed by a
       <code class="ph codeph">:</code>, and a capacity <code class="ph codeph"><var class="keyword varname">quota</var></code> per
       directory.
     </p>


     <p class="p">
       If set to an empty string, data caching is disabled.
     </p>


     <p class="p">
       Cached data is stored in the specified directories.
     </p>


     <p class="p">
       The specified directories must exist in the local filesystem of each Impala Daemon, or
       Impala will fail to start.
     </p>


     <p class="p">
       In addition, the filesystem which the directory resides in must support hole punching.
     </p>


     <p class="p">
       The cache can consume up to the <code class="ph codeph">quota</code> bytes for each of the directories
       specified.
     </p>


     <p class="p">
       The default setting for <code class="ph codeph">--data_cache</code> is an empty string.
     </p>


     <p class="p">
       For example, with the following setting, the data cache may use up to 1 TB, with 500 GB
       max in <code class="ph codeph">/data/0</code> and <code class="ph codeph">/data/1</code> respectively.
     </p>


 <pre class="pre codeblock"><code>--data_cache=/data/0,/data/1:500GB</code></pre>

     <div class="p"> In Impala 3.4 and higher, you can configure one of the following cache eviction policies for
       the data cache: <ul class="ul">
         <li class="li">LRU (Least Recently Used--the default)</li>

         <li class="li">LIRS (Inter-referenece Recency Set)</li>

       </ul>
  LIRS is a scan-resistent, low performance-overhead policy. You configure a cache
       eviction policy using the <code class="ph codeph">--data_cache_eviction_policy</code> Impala Daemon start-up
       flag: </div>

     <div class="p">
 <pre class="pre codeblock"><code>--data_cache_eviction_policy=<var class="keyword varname">policy</var>
 </code></pre>
     </div>

     <div class="note note"><span class="notetitle">Note:</span> The cache item will not expire as long as the same file metadata is used in the query.
       This is because the cache key consists of the filename, mtime (last modified time of the
       file), and file offset. If the mtime in the file metadata remains unchanged, the scan request
       will consistently access the cache (provided that there is enough capacity).</div>

   </div>


 <div class="related-links">
 <div class="familylinks">
 <div class="parentlink"><strong>Parent topic:</strong> <a class="link" href="../topics/impala_performance.html">Tuning Impala for Performance</a></div>
 </div>
 </div></body>
 </html>
	<?xml version="1.0" encoding="UTF-8"?>
	<!DOCTYPE html
	PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
	<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en" lang="en">
	<head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />

	<meta name="copyright" content="(C) Copyright 2025" />
	<meta name="DC.rights.owner" content="(C) Copyright 2025" />
	<meta name="DC.Type" content="concept" />
	<meta name="DC.Title" content="Data Cache for Remote Reads" />
	<meta name="DC.Relation" scheme="URI" content="../topics/impala_performance.html" />
	<meta name="prodname" content="Impala" />
	<meta name="version" content="Impala 3.4.x" />
	<meta name="DC.Format" content="XHTML" />
	<meta name="DC.Identifier" content="data_cache" />
	<link rel="stylesheet" type="text/css" href="../commonltr.css" />
	<title>Data Cache for Remote Reads</title>
	</head>
	<body id="data_cache">


	<h1 class="title topictitle1" id="ariaid-title1">Data Cache for Remote Reads</h1>


	<div class="body conbody">

	<p class="p">
	When Impala compute nodes and its storage are not co-located, the network bandwidth
	requirement goes up as the network traffic includes the data fetch as well as the
	shuffling exchange traffic of intermediate results.
	</p>


	<p class="p">
	To mitigate the pressure on the network, you can enable the compute nodes to cache the
	working set read from remote filesystems, such as, remote HDFS data node, S3, ABFS, ADLS.
	</p>


	<p class="p">
	To enable remote data cache, set the <code class="ph codeph">--data_cache</code> Impala Daemon start-up
	flag as below:
	</p>


	<pre class="pre codeblock"><code>--data_cache=<var class="keyword varname">dir1</var>,<var class="keyword varname">dir2</var>,<var class="keyword varname">dir3</var>,...:<var class="keyword varname">quota</var></code></pre>

	<p class="p">
	The flag is set to a list of directories, separated by <code class="ph codeph">,</code>, followed by a
	<code class="ph codeph">:</code>, and a capacity <code class="ph codeph"><var class="keyword varname">quota</var></code> per
	directory.
	</p>


	<p class="p">
	If set to an empty string, data caching is disabled.
	</p>


	<p class="p">
	Cached data is stored in the specified directories.
	</p>


	<p class="p">
	The specified directories must exist in the local filesystem of each Impala Daemon, or
	Impala will fail to start.
	</p>


	<p class="p">
	In addition, the filesystem which the directory resides in must support hole punching.
	</p>


	<p class="p">
	The cache can consume up to the <code class="ph codeph">quota</code> bytes for each of the directories
	specified.
	</p>


	<p class="p">
	The default setting for <code class="ph codeph">--data_cache</code> is an empty string.
	</p>


	<p class="p">
	For example, with the following setting, the data cache may use up to 1 TB, with 500 GB
	max in <code class="ph codeph">/data/0</code> and <code class="ph codeph">/data/1</code> respectively.
	</p>


	<pre class="pre codeblock"><code>--data_cache=/data/0,/data/1:500GB</code></pre>

	<div class="p"> In Impala 3.4 and higher, you can configure one of the following cache eviction policies for
	the data cache: <ul class="ul">
	<li class="li">LRU (Least Recently Used--the default)</li>

	<li class="li">LIRS (Inter-referenece Recency Set)</li>

	</ul>
	LIRS is a scan-resistent, low performance-overhead policy. You configure a cache
	eviction policy using the <code class="ph codeph">--data_cache_eviction_policy</code> Impala Daemon start-up
	flag: </div>

	<div class="p">
	<pre class="pre codeblock"><code>--data_cache_eviction_policy=<var class="keyword varname">policy</var>
	</code></pre>
	</div>

	<div class="note note"><span class="notetitle">Note:</span> The cache item will not expire as long as the same file metadata is used in the query.
	This is because the cache key consists of the filename, mtime (last modified time of the
	file), and file offset. If the mtime in the file metadata remains unchanged, the scan request
	will consistently access the cache (provided that there is enough capacity).</div>

	</div>


	<div class="related-links">
	<div class="familylinks">
	<div class="parentlink"><strong>Parent topic:</strong> <a class="link" href="../topics/impala_performance.html">Tuning Impala for Performance</a></div>
	</div>
	</div></body>
	</html>