---
layout: docpage

title: "Documentation"

is_homepage: false
is_sphinx_doc: true

doc-parent: "Operating Cassandra"

doc-title: "Compaction"
doc-header-links: '
  <link rel="top" title="Apache Cassandra Documentation v3.10" href="../index.html"/>
      <link rel="up" title="Operating Cassandra" href="index.html"/>
      <link rel="next" title="Bloom Filters" href="bloom_filters.html"/>
      <link rel="prev" title="Hints" href="hints.html"/>
'
doc-search-path: "../search.html"

extra-footer: '
<script type="text/javascript">
    var DOCUMENTATION_OPTIONS = {
      URL_ROOT:    "",
      VERSION:     "",
      COLLAPSE_INDEX: false,
      FILE_SUFFIX: ".html",
      HAS_SOURCE:  false,
      SOURCELINK_SUFFIX: ""
    };
</script>
'

---
<div class="container-fluid">
  <div class="row">
    <div class="col-md-2">
      <div class="doc-navigation">
        <div class="doc-menu" role="navigation">
          <div class="navbar-header">
            <button type="button" class="pull-left navbar-toggle" data-toggle="collapse" data-target=".sidebar-navbar-collapse">
              <span class="sr-only">Toggle navigation</span>
              <span class="icon-bar"></span>
              <span class="icon-bar"></span>
              <span class="icon-bar"></span>
            </button>
          </div>
          <div class="navbar-collapse collapse sidebar-navbar-collapse">
            <form id="doc-search-form" class="navbar-form" action="../search.html" method="get" role="search">
              <div class="form-group">
                <input type="text" size="30" class="form-control input-sm" name="q" placeholder="Search docs">
                <input type="hidden" name="check_keywords" value="yes" />
                <input type="hidden" name="area" value="default" />
              </div>
            </form>
            
            
            <ul class="current">
<li class="toctree-l1"><a class="reference internal" href="../getting_started/index.html">Getting Started</a></li>
<li class="toctree-l1"><a class="reference internal" href="../architecture/index.html">Architecture</a></li>
<li class="toctree-l1"><a class="reference internal" href="../data_modeling/index.html">Data Modeling</a></li>
<li class="toctree-l1"><a class="reference internal" href="../cql/index.html">The Cassandra Query Language (CQL)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../configuration/index.html">Configuring Cassandra</a></li>
<li class="toctree-l1 current"><a class="reference internal" href="index.html">Operating Cassandra</a><ul class="current">
<li class="toctree-l2"><a class="reference internal" href="snitch.html">Snitch</a></li>
<li class="toctree-l2"><a class="reference internal" href="topo_changes.html">Adding, replacing, moving and removing nodes</a></li>
<li class="toctree-l2"><a class="reference internal" href="repair.html">Repair</a></li>
<li class="toctree-l2"><a class="reference internal" href="read_repair.html">Read repair</a></li>
<li class="toctree-l2"><a class="reference internal" href="hints.html">Hints</a></li>
<li class="toctree-l2 current"><a class="current reference internal" href="#">Compaction</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#types-of-compaction">Types of compaction</a></li>
<li class="toctree-l3"><a class="reference internal" href="#when-is-a-minor-compaction-triggered">When is a minor compaction triggered?</a></li>
<li class="toctree-l3"><a class="reference internal" href="#merging-sstables">Merging sstables</a></li>
<li class="toctree-l3"><a class="reference internal" href="#tombstones-and-garbage-collection-gc-grace">Tombstones and Garbage Collection (GC) Grace</a></li>
<li class="toctree-l3"><a class="reference internal" href="#ttl">TTL</a></li>
<li class="toctree-l3"><a class="reference internal" href="#fully-expired-sstables">Fully expired sstables</a></li>
<li class="toctree-l3"><a class="reference internal" href="#repaired-unrepaired-data">Repaired/unrepaired data</a></li>
<li class="toctree-l3"><a class="reference internal" href="#data-directories">Data directories</a></li>
<li class="toctree-l3"><a class="reference internal" href="#single-sstable-tombstone-compaction">Single sstable tombstone compaction</a></li>
<li class="toctree-l3"><a class="reference internal" href="#common-options">Common options</a></li>
<li class="toctree-l3"><a class="reference internal" href="#compaction-nodetool-commands">Compaction nodetool commands</a></li>
<li class="toctree-l3"><a class="reference internal" href="#switching-the-compaction-strategy-and-options-using-jmx">Switching the compaction strategy and options using JMX</a></li>
<li class="toctree-l3"><a class="reference internal" href="#more-detailed-compaction-logging">More detailed compaction logging</a></li>
<li class="toctree-l3"><a class="reference internal" href="#size-tiered-compaction-strategy">Size Tiered Compaction Strategy</a></li>
<li class="toctree-l3"><a class="reference internal" href="#leveled-compaction-strategy">Leveled Compaction Strategy</a></li>
<li class="toctree-l3"><a class="reference internal" href="#time-window-compactionstrategy">Time Window CompactionStrategy</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="bloom_filters.html">Bloom Filters</a></li>
<li class="toctree-l2"><a class="reference internal" href="compression.html">Compression</a></li>
<li class="toctree-l2"><a class="reference internal" href="cdc.html">Change Data Capture</a></li>
<li class="toctree-l2"><a class="reference internal" href="backups.html">Backups</a></li>
<li class="toctree-l2"><a class="reference internal" href="bulk_loading.html">Bulk Loading</a></li>
<li class="toctree-l2"><a class="reference internal" href="metrics.html">Monitoring</a></li>
<li class="toctree-l2"><a class="reference internal" href="security.html">Security</a></li>
<li class="toctree-l2"><a class="reference internal" href="hardware.html">Hardware Choices</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../tools/index.html">Cassandra Tools</a></li>
<li class="toctree-l1"><a class="reference internal" href="../troubleshooting/index.html">Troubleshooting</a></li>
<li class="toctree-l1"><a class="reference internal" href="../development/index.html">Cassandra Development</a></li>
<li class="toctree-l1"><a class="reference internal" href="../faq/index.html">Frequently Asked Questions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../bugs.html">Reporting Bugs and Contributing</a></li>
<li class="toctree-l1"><a class="reference internal" href="../contactus.html">Contact us</a></li>
</ul>

            
          </div><!--/.nav-collapse -->
        </div>
      </div>
    </div>
    <div class="col-md-8">
      <div class="content doc-content">
        <div class="container">
          
  <div class="section" id="compaction">
<span id="id1"></span><h1>Compaction<a class="headerlink" href="#compaction" title="Permalink to this headline">¶</a></h1>
<div class="section" id="types-of-compaction">
<h2>Types of compaction<a class="headerlink" href="#types-of-compaction" title="Permalink to this headline">¶</a></h2>
<p>The concept of compaction is used for different kinds of operations in Cassandra, the common thing about these
operations is that it takes one or more sstables and output new sstables. The types of compactions are;</p>
<dl class="docutils">
<dt>Minor compaction</dt>
<dd>triggered automatically in Cassandra.</dd>
<dt>Major compaction</dt>
<dd>a user executes a compaction over all sstables on the node.</dd>
<dt>User defined compaction</dt>
<dd>a user triggers a compaction on a given set of sstables.</dd>
<dt>Scrub</dt>
<dd>try to fix any broken sstables. This can actually remove valid data if that data is corrupted, if that happens you
will need to run a full repair on the node.</dd>
<dt>Upgradesstables</dt>
<dd>upgrade sstables to the latest version. Run this after upgrading to a new major version.</dd>
<dt>Cleanup</dt>
<dd>remove any ranges this node does not own anymore, typically triggered on neighbouring nodes after a node has been
bootstrapped since that node will take ownership of some ranges from those nodes.</dd>
<dt>Secondary index rebuild</dt>
<dd>rebuild the secondary indexes on the node.</dd>
<dt>Anticompaction</dt>
<dd>after repair the ranges that were actually repaired are split out of the sstables that existed when repair started.</dd>
<dt>Sub range compaction</dt>
<dd>It is possible to only compact a given sub range - this could be useful if you know a token that has been
misbehaving - either gathering many updates or many deletes. (<code class="docutils literal"><span class="pre">nodetool</span> <span class="pre">compact</span> <span class="pre">-st</span> <span class="pre">x</span> <span class="pre">-et</span> <span class="pre">y</span></code>) will pick
all sstables containing the range between x and y and issue a compaction for those sstables. For STCS this will
most likely include all sstables but with LCS it can issue the compaction for a subset of the sstables. With LCS
the resulting sstable will end up in L0.</dd>
</dl>
</div>
<div class="section" id="when-is-a-minor-compaction-triggered">
<h2>When is a minor compaction triggered?<a class="headerlink" href="#when-is-a-minor-compaction-triggered" title="Permalink to this headline">¶</a></h2>
<p>#  When an sstable is added to the node through flushing/streaming etc.
#  When autocompaction is enabled after being disabled (<code class="docutils literal"><span class="pre">nodetool</span> <span class="pre">enableautocompaction</span></code>)
#  When compaction adds new sstables.
#  A check for new minor compactions every 5 minutes.</p>
</div>
<div class="section" id="merging-sstables">
<h2>Merging sstables<a class="headerlink" href="#merging-sstables" title="Permalink to this headline">¶</a></h2>
<p>Compaction is about merging sstables, since partitions in sstables are sorted based on the hash of the partition key it
is possible to efficiently merge separate sstables. Content of each partition is also sorted so each partition can be
merged efficiently.</p>
</div>
<div class="section" id="tombstones-and-garbage-collection-gc-grace">
<h2>Tombstones and Garbage Collection (GC) Grace<a class="headerlink" href="#tombstones-and-garbage-collection-gc-grace" title="Permalink to this headline">¶</a></h2>
<div class="section" id="why-tombstones">
<h3>Why Tombstones<a class="headerlink" href="#why-tombstones" title="Permalink to this headline">¶</a></h3>
<p>When a delete request is received by Cassandra it does not actually remove the data from the underlying store. Instead
it writes a special piece of data known as a tombstone. The Tombstone represents the delete and causes all values which
occurred before the tombstone to not appear in queries to the database. This approach is used instead of removing values
because of the distributed nature of Cassandra.</p>
</div>
<div class="section" id="deletes-without-tombstones">
<h3>Deletes without tombstones<a class="headerlink" href="#deletes-without-tombstones" title="Permalink to this headline">¶</a></h3>
<p>Imagine a three node cluster which has the value [A] replicated to every node.:</p>
<div class="highlight-none"><div class="highlight"><pre><span></span>[A], [A], [A]
</pre></div>
</div>
<p>If one of the nodes fails and and our delete operation only removes existing values we can end up with a cluster that
looks like:</p>
<div class="highlight-none"><div class="highlight"><pre><span></span>[], [], [A]
</pre></div>
</div>
<p>Then a repair operation would replace the value of [A] back onto the two
nodes which are missing the value.:</p>
<div class="highlight-none"><div class="highlight"><pre><span></span>[A], [A], [A]
</pre></div>
</div>
<p>This would cause our data to be resurrected even though it had been
deleted.</p>
</div>
<div class="section" id="deletes-with-tombstones">
<h3>Deletes with Tombstones<a class="headerlink" href="#deletes-with-tombstones" title="Permalink to this headline">¶</a></h3>
<p>Starting again with a three node cluster which has the value [A] replicated to every node.:</p>
<div class="highlight-none"><div class="highlight"><pre><span></span>[A], [A], [A]
</pre></div>
</div>
<p>If instead of removing data we add a tombstone record, our single node failure situation will look like this.:</p>
<div class="highlight-none"><div class="highlight"><pre><span></span>[A, Tombstone[A]], [A, Tombstone[A]], [A]
</pre></div>
</div>
<p>Now when we issue a repair the Tombstone will be copied to the replica, rather than the deleted data being
resurrected.:</p>
<div class="highlight-none"><div class="highlight"><pre><span></span>[A, Tombstone[A]], [A, Tombstone[A]], [A, Tombstone[A]]
</pre></div>
</div>
<p>Our repair operation will correctly put the state of the system to what we expect with the record [A] marked as deleted
on all nodes. This does mean we will end up accruing Tombstones which will permanently accumulate disk space. To avoid
keeping tombstones forever we have a parameter known as <code class="docutils literal"><span class="pre">gc_grace_seconds</span></code> for every table in Cassandra.</p>
</div>
<div class="section" id="the-gc-grace-seconds-parameter-and-tombstone-removal">
<h3>The gc_grace_seconds parameter and Tombstone Removal<a class="headerlink" href="#the-gc-grace-seconds-parameter-and-tombstone-removal" title="Permalink to this headline">¶</a></h3>
<p>The table level <code class="docutils literal"><span class="pre">gc_grace_seconds</span></code> parameter controls how long Cassandra will retain tombstones through compaction
events before finally removing them. This duration should directly reflect the amount of time a user expects to allow
before recovering a failed node. After <code class="docutils literal"><span class="pre">gc_grace_seconds</span></code> has expired the tombstone may be removed (meaning there will
no longer be any record that a certain piece of data was deleted), but as a tombstone can live in one sstable and the
data it covers in another, a compaction must also include both sstable for a tombstone to be removed. More precisely, to
be able to drop an actual tombstone the following needs to be true;</p>
<ul class="simple">
<li>The tombstone must be older than <code class="docutils literal"><span class="pre">gc_grace_seconds</span></code></li>
<li>If partition X contains the tombstone, the sstable containing the partition plus all sstables containing data older
than the tombstone containing X must be included in the same compaction. We don&#8217;t need to care if the partition is in
an sstable if we can guarantee that all data in that sstable is newer than the tombstone. If the tombstone is older
than the data it cannot shadow that data.</li>
<li>If the option <code class="docutils literal"><span class="pre">only_purge_repaired_tombstones</span></code> is enabled, tombstones are only removed if the data has also been
repaired.</li>
</ul>
<p>If a node remains down or disconnected for longer than <code class="docutils literal"><span class="pre">gc_grace_seconds</span></code> it&#8217;s deleted data will be repaired back to
the other nodes and re-appear in the cluster. This is basically the same as in the &#8220;Deletes without Tombstones&#8221; section.
Note that tombstones will not be removed until a compaction event even if <code class="docutils literal"><span class="pre">gc_grace_seconds</span></code> has elapsed.</p>
<p>The default value for <code class="docutils literal"><span class="pre">gc_grace_seconds</span></code> is 864000 which is equivalent to 10 days. This can be set when creating or
altering a table using <code class="docutils literal"><span class="pre">WITH</span> <span class="pre">gc_grace_seconds</span></code>.</p>
</div>
</div>
<div class="section" id="ttl">
<h2>TTL<a class="headerlink" href="#ttl" title="Permalink to this headline">¶</a></h2>
<p>Data in Cassandra can have an additional property called time to live - this is used to automatically drop data that has
expired once the time is reached. Once the TTL has expired the data is converted to a tombstone which stays around for
at least <code class="docutils literal"><span class="pre">gc_grace_seconds</span></code>. Note that if you mix data with TTL and data without TTL (or just different length of the
TTL) Cassandra will have a hard time dropping the tombstones created since the partition might span many sstables and
not all are compacted at once.</p>
</div>
<div class="section" id="fully-expired-sstables">
<h2>Fully expired sstables<a class="headerlink" href="#fully-expired-sstables" title="Permalink to this headline">¶</a></h2>
<p>If an sstable contains only tombstones and it is guaranteed that that sstable is not shadowing data in any other sstable
compaction can drop that sstable. If you see sstables with only tombstones (note that TTL:ed data is considered
tombstones once the time to live has expired) but it is not being dropped by compaction, it is likely that other
sstables contain older data. There is a tool called <code class="docutils literal"><span class="pre">sstableexpiredblockers</span></code> that will list which sstables are
droppable and which are blocking them from being dropped. This is especially useful for time series compaction with
<code class="docutils literal"><span class="pre">TimeWindowCompactionStrategy</span></code> (and the deprecated <code class="docutils literal"><span class="pre">DateTieredCompactionStrategy</span></code>).</p>
</div>
<div class="section" id="repaired-unrepaired-data">
<h2>Repaired/unrepaired data<a class="headerlink" href="#repaired-unrepaired-data" title="Permalink to this headline">¶</a></h2>
<p>With incremental repairs Cassandra must keep track of what data is repaired and what data is unrepaired. With
anticompaction repaired data is split out into repaired and unrepaired sstables. To avoid mixing up the data again
separate compaction strategy instances are run on the two sets of data, each instance only knowing about either the
repaired or the unrepaired sstables. This means that if you only run incremental repair once and then never again, you
might have very old data in the repaired sstables that block compaction from dropping tombstones in the unrepaired
(probably newer) sstables.</p>
</div>
<div class="section" id="data-directories">
<h2>Data directories<a class="headerlink" href="#data-directories" title="Permalink to this headline">¶</a></h2>
<p>Since tombstones and data can live in different sstables it is important to realize that losing an sstable might lead to
data becoming live again - the most common way of losing sstables is to have a hard drive break down. To avoid making
data live tombstones and actual data are always in the same data directory. This way, if a disk is lost, all versions of
a partition are lost and no data can get undeleted. To achieve this a compaction strategy instance per data directory is
run in addition to the compaction strategy instances containing repaired/unrepaired data, this means that if you have 4
data directories there will be 8 compaction strategy instances running. This has a few more benefits than just avoiding
data getting undeleted:</p>
<ul class="simple">
<li>It is possible to run more compactions in parallel - leveled compaction will have several totally separate levelings
and each one can run compactions independently from the others.</li>
<li>Users can backup and restore a single data directory.</li>
<li>Note though that currently all data directories are considered equal, so if you have a tiny disk and a big disk
backing two data directories, the big one will be limited the by the small one. One work around to this is to create
more data directories backed by the big disk.</li>
</ul>
</div>
<div class="section" id="single-sstable-tombstone-compaction">
<h2>Single sstable tombstone compaction<a class="headerlink" href="#single-sstable-tombstone-compaction" title="Permalink to this headline">¶</a></h2>
<p>When an sstable is written a histogram with the tombstone expiry times is created and this is used to try to find
sstables with very many tombstones and run single sstable compaction on that sstable in hope of being able to drop
tombstones in that sstable. Before starting this it is also checked how likely it is that any tombstones will actually
will be able to be dropped how much this sstable overlaps with other sstables. To avoid most of these checks the
compaction option <code class="docutils literal"><span class="pre">unchecked_tombstone_compaction</span></code> can be enabled.</p>
</div>
<div class="section" id="common-options">
<span id="compaction-options"></span><h2>Common options<a class="headerlink" href="#common-options" title="Permalink to this headline">¶</a></h2>
<p>There is a number of common options for all the compaction strategies;</p>
<dl class="docutils">
<dt><code class="docutils literal"><span class="pre">enabled</span></code> (default: true)</dt>
<dd>Whether minor compactions should run. Note that you can have &#8216;enabled&#8217;: true as a compaction option and then do
&#8216;nodetool enableautocompaction&#8217; to start running compactions.</dd>
<dt><code class="docutils literal"><span class="pre">tombstone_threshold</span></code> (default: 0.2)</dt>
<dd>How much of the sstable should be tombstones for us to consider doing a single sstable compaction of that sstable.</dd>
<dt><code class="docutils literal"><span class="pre">tombstone_compaction_interval</span></code> (default: 86400s (1 day))</dt>
<dd>Since it might not be possible to drop any tombstones when doing a single sstable compaction we need to make sure
that one sstable is not constantly getting recompacted - this option states how often we should try for a given
sstable.</dd>
<dt><code class="docutils literal"><span class="pre">log_all</span></code> (default: false)</dt>
<dd>New detailed compaction logging, see <a class="reference internal" href="#detailed-compaction-logging"><span class="std std-ref">below</span></a>.</dd>
<dt><code class="docutils literal"><span class="pre">unchecked_tombstone_compaction</span></code> (default: false)</dt>
<dd>The single sstable compaction has quite strict checks for whether it should be started, this option disables those
checks and for some usecases this might be needed.  Note that this does not change anything for the actual
compaction, tombstones are only dropped if it is safe to do so - it might just rewrite an sstable without being able
to drop any tombstones.</dd>
<dt><code class="docutils literal"><span class="pre">only_purge_repaired_tombstone</span></code> (default: false)</dt>
<dd>Option to enable the extra safety of making sure that tombstones are only dropped if the data has been repaired.</dd>
<dt><code class="docutils literal"><span class="pre">min_threshold</span></code> (default: 4)</dt>
<dd>Lower limit of number of sstables before a compaction is triggered. Not used for <code class="docutils literal"><span class="pre">LeveledCompactionStrategy</span></code>.</dd>
<dt><code class="docutils literal"><span class="pre">max_threshold</span></code> (default: 32)</dt>
<dd>Upper limit of number of sstables before a compaction is triggered. Not used for <code class="docutils literal"><span class="pre">LeveledCompactionStrategy</span></code>.</dd>
</dl>
<p>Further, see the section on each strategy for specific additional options.</p>
</div>
<div class="section" id="compaction-nodetool-commands">
<h2>Compaction nodetool commands<a class="headerlink" href="#compaction-nodetool-commands" title="Permalink to this headline">¶</a></h2>
<p>The <a class="reference internal" href="../tools/nodetool.html#nodetool"><span class="std std-ref">nodetool</span></a> utility provides a number of commands related to compaction:</p>
<dl class="docutils">
<dt><code class="docutils literal"><span class="pre">enableautocompaction</span></code></dt>
<dd>Enable compaction.</dd>
<dt><code class="docutils literal"><span class="pre">disableautocompaction</span></code></dt>
<dd>Disable compaction.</dd>
<dt><code class="docutils literal"><span class="pre">setcompactionthroughput</span></code></dt>
<dd>How fast compaction should run at most - defaults to 16MB/s, but note that it is likely not possible to reach this
throughput.</dd>
<dt><code class="docutils literal"><span class="pre">compactionstats</span></code></dt>
<dd>Statistics about current and pending compactions.</dd>
<dt><code class="docutils literal"><span class="pre">compactionhistory</span></code></dt>
<dd>List details about the last compactions.</dd>
<dt><code class="docutils literal"><span class="pre">setcompactionthreshold</span></code></dt>
<dd>Set the min/max sstable count for when to trigger compaction, defaults to 4/32.</dd>
</dl>
</div>
<div class="section" id="switching-the-compaction-strategy-and-options-using-jmx">
<h2>Switching the compaction strategy and options using JMX<a class="headerlink" href="#switching-the-compaction-strategy-and-options-using-jmx" title="Permalink to this headline">¶</a></h2>
<p>It is possible to switch compaction strategies and its options on just a single node using JMX, this is a great way to
experiment with settings without affecting the whole cluster. The mbean is:</p>
<div class="highlight-none"><div class="highlight"><pre><span></span>org.apache.cassandra.db:type=ColumnFamilies,keyspace=&lt;keyspace_name&gt;,columnfamily=&lt;table_name&gt;
</pre></div>
</div>
<p>and the attribute to change is <code class="docutils literal"><span class="pre">CompactionParameters</span></code> or <code class="docutils literal"><span class="pre">CompactionParametersJson</span></code> if you use jconsole or jmc. The
syntax for the json version is the same as you would use in an <a class="reference internal" href="../cql/ddl.html#alter-table-statement"><span class="std std-ref">ALTER TABLE</span></a> statement -
for example:</p>
<div class="highlight-none"><div class="highlight"><pre><span></span>{ &#39;class&#39;: &#39;LeveledCompactionStrategy&#39;, &#39;sstable_size_in_mb&#39;: 123 }
</pre></div>
</div>
<p>The setting is kept until someone executes an <a class="reference internal" href="../cql/ddl.html#alter-table-statement"><span class="std std-ref">ALTER TABLE</span></a> that touches the compaction
settings or restarts the node.</p>
</div>
<div class="section" id="more-detailed-compaction-logging">
<span id="detailed-compaction-logging"></span><h2>More detailed compaction logging<a class="headerlink" href="#more-detailed-compaction-logging" title="Permalink to this headline">¶</a></h2>
<p>Enable with the compaction option <code class="docutils literal"><span class="pre">log_all</span></code> and a more detailed compaction log file will be produced in your log
directory.</p>
</div>
<div class="section" id="size-tiered-compaction-strategy">
<span id="stcs"></span><h2>Size Tiered Compaction Strategy<a class="headerlink" href="#size-tiered-compaction-strategy" title="Permalink to this headline">¶</a></h2>
<p>The basic idea of <code class="docutils literal"><span class="pre">SizeTieredCompactionStrategy</span></code> (STCS) is to merge sstables of approximately the same size. All
sstables are put in different buckets depending on their size. An sstable is added to the bucket if size of the sstable
is within <code class="docutils literal"><span class="pre">bucket_low</span></code> and <code class="docutils literal"><span class="pre">bucket_high</span></code> of the current average size of the sstables already in the bucket. This
will create several buckets and the most interesting of those buckets will be compacted. The most interesting one is
decided by figuring out which bucket&#8217;s sstables takes the most reads.</p>
<div class="section" id="major-compaction">
<h3>Major compaction<a class="headerlink" href="#major-compaction" title="Permalink to this headline">¶</a></h3>
<p>When running a major compaction with STCS you will end up with two sstables per data directory (one for repaired data
and one for unrepaired data). There is also an option (-s) to do a major compaction that splits the output into several
sstables. The sizes of the sstables are approximately 50%, 25%, 12.5%... of the total size.</p>
</div>
<div class="section" id="stcs-options">
<span id="id2"></span><h3>STCS options<a class="headerlink" href="#stcs-options" title="Permalink to this headline">¶</a></h3>
<dl class="docutils">
<dt><code class="docutils literal"><span class="pre">min_sstable_size</span></code> (default: 50MB)</dt>
<dd>Sstables smaller than this are put in the same bucket.</dd>
<dt><code class="docutils literal"><span class="pre">bucket_low</span></code> (default: 0.5)</dt>
<dd>How much smaller than the average size of a bucket a sstable should be before not being included in the bucket. That
is, if <code class="docutils literal"><span class="pre">bucket_low</span> <span class="pre">*</span> <span class="pre">avg_bucket_size</span> <span class="pre">&lt;</span> <span class="pre">sstable_size</span></code> (and the <code class="docutils literal"><span class="pre">bucket_high</span></code> condition holds, see below), then
the sstable is added to the bucket.</dd>
<dt><code class="docutils literal"><span class="pre">bucket_high</span></code> (default: 1.5)</dt>
<dd>How much bigger than the average size of a bucket a sstable should be before not being included in the bucket. That
is, if <code class="docutils literal"><span class="pre">sstable_size</span> <span class="pre">&lt;</span> <span class="pre">bucket_high</span> <span class="pre">*</span> <span class="pre">avg_bucket_size</span></code> (and the <code class="docutils literal"><span class="pre">bucket_low</span></code> condition holds, see above), then
the sstable is added to the bucket.</dd>
</dl>
</div>
<div class="section" id="defragmentation">
<h3>Defragmentation<a class="headerlink" href="#defragmentation" title="Permalink to this headline">¶</a></h3>
<p>Defragmentation is done when many sstables are touched during a read.  The result of the read is put in to the memtable
so that the next read will not have to touch as many sstables. This can cause writes on a read-only-cluster.</p>
</div>
</div>
<div class="section" id="leveled-compaction-strategy">
<span id="lcs"></span><h2>Leveled Compaction Strategy<a class="headerlink" href="#leveled-compaction-strategy" title="Permalink to this headline">¶</a></h2>
<p>The idea of <code class="docutils literal"><span class="pre">LeveledCompactionStrategy</span></code> (LCS) is that all sstables are put into different levels where we guarantee
that no overlapping sstables are in the same level. By overlapping we mean that the first/last token of a single sstable
are never overlapping with other sstables. This means that for a SELECT we will only have to look for the partition key
in a single sstable per level. Each level is 10x the size of the previous one and each sstable is 160MB by default. L0
is where sstables are streamed/flushed - no overlap guarantees are given here.</p>
<p>When picking compaction candidates we have to make sure that the compaction does not create overlap in the target level.
This is done by always including all overlapping sstables in the next level. For example if we select an sstable in L3,
we need to guarantee that we pick all overlapping sstables in L4 and make sure that no currently ongoing compactions
will create overlap if we start that compaction. We can start many parallel compactions in a level if we guarantee that
we wont create overlap. For L0 -&gt; L1 compactions we almost always need to include all L1 sstables since most L0 sstables
cover the full range. We also can&#8217;t compact all L0 sstables with all L1 sstables in a single compaction since that can
use too much memory.</p>
<p>When deciding which level to compact LCS checks the higher levels first (with LCS, a &#8220;higher&#8221; level is one with a higher
number, L0 being the lowest one) and if the level is behind a compaction will be started in that level.</p>
<div class="section" id="id3">
<h3>Major compaction<a class="headerlink" href="#id3" title="Permalink to this headline">¶</a></h3>
<p>It is possible to do a major compaction with LCS - it will currently start by filling out L1 and then once L1 is full,
it continues with L2 etc. This is sub optimal and will change to create all the sstables in a high level instead,
CASSANDRA-11817.</p>
</div>
<div class="section" id="bootstrapping">
<h3>Bootstrapping<a class="headerlink" href="#bootstrapping" title="Permalink to this headline">¶</a></h3>
<p>During bootstrap sstables are streamed from other nodes. The level of the remote sstable is kept to avoid many
compactions after the bootstrap is done. During bootstrap the new node also takes writes while it is streaming the data
from a remote node - these writes are flushed to L0 like all other writes and to avoid those sstables blocking the
remote sstables from going to the correct level, we only do STCS in L0 until the bootstrap is done.</p>
</div>
<div class="section" id="stcs-in-l0">
<h3>STCS in L0<a class="headerlink" href="#stcs-in-l0" title="Permalink to this headline">¶</a></h3>
<p>If LCS gets very many L0 sstables reads are going to hit all (or most) of the L0 sstables since they are likely to be
overlapping. To more quickly remedy this LCS does STCS compactions in L0 if there are more than 32 sstables there. This
should improve read performance more quickly compared to letting LCS do its L0 -&gt; L1 compactions. If you keep getting
too many sstables in L0 it is likely that LCS is not the best fit for your workload and STCS could work out better.</p>
</div>
<div class="section" id="starved-sstables">
<h3>Starved sstables<a class="headerlink" href="#starved-sstables" title="Permalink to this headline">¶</a></h3>
<p>If a node ends up with a leveling where there are a few very high level sstables that are not getting compacted they
might make it impossible for lower levels to drop tombstones etc. For example, if there are sstables in L6 but there is
only enough data to actually get a L4 on the node the left over sstables in L6 will get starved and not compacted.  This
can happen if a user changes sstable_size_in_mb from 5MB to 160MB for example. To avoid this LCS tries to include
those starved high level sstables in other compactions if there has been 25 compaction rounds where the highest level
has not been involved.</p>
</div>
<div class="section" id="lcs-options">
<span id="id4"></span><h3>LCS options<a class="headerlink" href="#lcs-options" title="Permalink to this headline">¶</a></h3>
<dl class="docutils">
<dt><code class="docutils literal"><span class="pre">sstable_size_in_mb</span></code> (default: 160MB)</dt>
<dd>The target compressed (if using compression) sstable size - the sstables can end up being larger if there are very
large partitions on the node.</dd>
</dl>
<p>LCS also support the <code class="docutils literal"><span class="pre">cassandra.disable_stcs_in_l0</span></code> startup option (<code class="docutils literal"><span class="pre">-Dcassandra.disable_stcs_in_l0=true</span></code>) to avoid
doing STCS in L0.</p>
</div>
</div>
<div class="section" id="time-window-compactionstrategy">
<span id="twcs"></span><h2>Time Window CompactionStrategy<a class="headerlink" href="#time-window-compactionstrategy" title="Permalink to this headline">¶</a></h2>
<p><code class="docutils literal"><span class="pre">TimeWindowCompactionStrategy</span></code> (TWCS) is designed specifically for workloads where it&#8217;s beneficial to have data on
disk grouped by the timestamp of the data, a common goal when the workload is time-series in nature or when all data is
written with a TTL. In an expiring/TTL workload, the contents of an entire SSTable likely expire at approximately the
same time, allowing them to be dropped completely, and space reclaimed much more reliably than when using
<code class="docutils literal"><span class="pre">SizeTieredCompactionStrategy</span></code> or <code class="docutils literal"><span class="pre">LeveledCompactionStrategy</span></code>. The basic concept is that
<code class="docutils literal"><span class="pre">TimeWindowCompactionStrategy</span></code> will create 1 sstable per file for a given window, where a window is simply calculated
as the combination of two primary options:</p>
<dl class="docutils">
<dt><code class="docutils literal"><span class="pre">compaction_window_unit</span></code> (default: DAYS)</dt>
<dd>A Java TimeUnit (MINUTES, HOURS, or DAYS).</dd>
<dt><code class="docutils literal"><span class="pre">compaction_window_size</span></code> (default: 1)</dt>
<dd>The number of units that make up a window.</dd>
</dl>
<p>Taken together, the operator can specify windows of virtually any size, and <cite>TimeWindowCompactionStrategy</cite> will work to
create a single sstable for writes within that window. For efficiency during writing, the newest window will be
compacted using <cite>SizeTieredCompactionStrategy</cite>.</p>
<p>Ideally, operators should select a <code class="docutils literal"><span class="pre">compaction_window_unit</span></code> and <code class="docutils literal"><span class="pre">compaction_window_size</span></code> pair that produces
approximately 20-30 windows - if writing with a 90 day TTL, for example, a 3 Day window would be a reasonable choice
(<code class="docutils literal"><span class="pre">'compaction_window_unit':'DAYS','compaction_window_size':3</span></code>).</p>
<div class="section" id="timewindowcompactionstrategy-operational-concerns">
<h3>TimeWindowCompactionStrategy Operational Concerns<a class="headerlink" href="#timewindowcompactionstrategy-operational-concerns" title="Permalink to this headline">¶</a></h3>
<p>The primary motivation for TWCS is to separate data on disk by timestamp and to allow fully expired SSTables to drop
more efficiently. One potential way this optimal behavior can be subverted is if data is written to SSTables out of
order, with new data and old data in the same SSTable. Out of order data can appear in two ways:</p>
<ul class="simple">
<li>If the user mixes old data and new data in the traditional write path, the data will be comingled in the memtables
and flushed into the same SSTable, where it will remain comingled.</li>
<li>If the user&#8217;s read requests for old data cause read repairs that pull old data into the current memtable, that data
will be comingled and flushed into the same SSTable.</li>
</ul>
<p>While TWCS tries to minimize the impact of comingled data, users should attempt to avoid this behavior.  Specifically,
users should avoid queries that explicitly set the timestamp via CQL <code class="docutils literal"><span class="pre">USING</span> <span class="pre">TIMESTAMP</span></code>. Additionally, users should run
frequent repairs (which streams data in such a way that it does not become comingled), and disable background read
repair by setting the table&#8217;s <code class="docutils literal"><span class="pre">read_repair_chance</span></code> and <code class="docutils literal"><span class="pre">dclocal_read_repair_chance</span></code> to 0.</p>
</div>
<div class="section" id="changing-timewindowcompactionstrategy-options">
<h3>Changing TimeWindowCompactionStrategy Options<a class="headerlink" href="#changing-timewindowcompactionstrategy-options" title="Permalink to this headline">¶</a></h3>
<p>Operators wishing to enable <code class="docutils literal"><span class="pre">TimeWindowCompactionStrategy</span></code> on existing data should consider running a major compaction
first, placing all existing data into a single (old) window. Subsequent newer writes will then create typical SSTables
as expected.</p>
<p>Operators wishing to change <code class="docutils literal"><span class="pre">compaction_window_unit</span></code> or <code class="docutils literal"><span class="pre">compaction_window_size</span></code> can do so, but may trigger
additional compactions as adjacent windows are joined together. If the window size is decrease d (for example, from 24
hours to 12 hours), then the existing SSTables will not be modified - TWCS can not split existing SSTables into multiple
windows.</p>
</div>
</div>
</div>


          <div class="doc-prev-next-links" role="navigation" aria-label="footer navigation">
            
            <a href="bloom_filters.html" class="btn btn-default pull-right " role="button" title="Bloom Filters" accesskey="n">Next <span class="glyphicon glyphicon-circle-arrow-right" aria-hidden="true"></span></a>
            
            
            <a href="hints.html" class="btn btn-default" role="button" title="Hints" accesskey="p"><span class="glyphicon glyphicon-circle-arrow-left" aria-hidden="true"></span> Previous</a>
            
          </div>
          
        </div>
      </div>
    </div>
    <div class="col-md-2">
    </div>
  </div>
</div>