---
layout: docpage

title: "Documentation"

is_homepage: false
is_sphinx_doc: true

doc-parent: "Operating Cassandra"

doc-title: "Hints"
doc-header-links: '
  <link rel="top" title="Apache Cassandra Documentation v4.0-alpha5" href="../index.html"/>
      <link rel="up" title="Operating Cassandra" href="index.html"/>
      <link rel="next" title="Compaction" href="compaction/index.html"/>
      <link rel="prev" title="Read repair" href="read_repair.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: ".txt"
    };
</script>
'

---
<div class="container-fluid">
  <div class="row">
    <div class="col-md-3">
      <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="../new/index.html">New Features in Apache Cassandra 4.0</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="../cql/index.html">The Cassandra Query Language (CQL)</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="../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 current"><a class="current reference internal" href="#">Hints</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#hinted-handoff">Hinted Handoff</a></li>
<li class="toctree-l3"><a class="reference internal" href="#configuring-hints">Configuring Hints</a></li>
<li class="toctree-l3"><a class="reference internal" href="#configuring-hints-at-runtime-with-nodetool">Configuring Hints at Runtime with <code class="docutils literal notranslate"><span class="pre">nodetool</span></code></a></li>
<li class="toctree-l3"><a class="reference internal" href="#monitoring-hint-delivery">Monitoring Hint Delivery</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="compaction/index.html">Compaction</a></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">Contributing to Cassandra</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="../plugins/index.html">Third-Party Plugins</a></li>
<li class="toctree-l1"><a class="reference internal" href="../bugs.html">Reporting Bugs</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="content-container">
          
  <div class="section" id="hints">
<span id="id1"></span><h1>Hints<a class="headerlink" href="#hints" title="Permalink to this headline">¶</a></h1>
<p>Hinting is a data repair technique applied during write operations. When
replica nodes are unavailable to accept a mutation, either due to failure or
more commonly routine maintenance, coordinators attempting to write to those
replicas store temporary hints on their local filesystem for later application
to the unavailable replica. Hints are an important way to help reduce the
duration of data inconsistency. Coordinators replay hints quickly after
unavailable replica nodes return to the ring. Hints are best effort, however,
and do not guarantee eventual consistency like <a class="reference internal" href="repair.html#repair"><span class="std std-ref">anti-entropy repair</span></a> does.</p>
<p>Hints are useful because of how Apache Cassandra replicates data to provide
fault tolerance, high availability and durability. Cassandra <a class="reference internal" href="../architecture/dynamo.html#consistent-hashing-token-ring"><span class="std std-ref">partitions
data across the cluster</span></a> using consistent
hashing, and then replicates keys to multiple nodes along the hash ring. To
guarantee availability, all replicas of a key can accept mutations without
consensus, but this means it is possible for some replicas to accept a mutation
while others do not. When this happens an inconsistency is introduced.</p>
<p>Hints are one of the three ways, in addition to read-repair and
full/incremental anti-entropy repair, that Cassandra implements the eventual
consistency guarantee that all updates are eventually received by all replicas.
Hints, like read-repair, are best effort and not an alternative to performing
full repair, but they do help reduce the duration of inconsistency between
replicas in practice.</p>
<div class="section" id="hinted-handoff">
<h2>Hinted Handoff<a class="headerlink" href="#hinted-handoff" title="Permalink to this headline">¶</a></h2>
<p>Hinted handoff is the process by which Cassandra applies hints to unavailable
nodes.</p>
<p>For example, consider a mutation is to be made at <code class="docutils literal notranslate"><span class="pre">Consistency</span> <span class="pre">Level</span></code>
<code class="docutils literal notranslate"><span class="pre">LOCAL_QUORUM</span></code> against a keyspace with <code class="docutils literal notranslate"><span class="pre">Replication</span> <span class="pre">Factor</span></code> of <code class="docutils literal notranslate"><span class="pre">3</span></code>.
Normally the client sends the mutation to a single coordinator, who then sends
the mutation to all three replicas, and when two of the three replicas
acknowledge the mutation the coordinator responds successfully to the client.
If a replica node is unavailable, however, the coordinator stores a hint
locally to the filesystem for later application. New hints will be retained for
up to <code class="docutils literal notranslate"><span class="pre">max_hint_window_in_ms</span></code> of downtime (defaults to <code class="docutils literal notranslate"><span class="pre">3</span> <span class="pre">hours</span></code>).  If the
unavailable replica does return to the cluster before the window expires, the
coordinator applies any pending hinted mutations against the replica to ensure
that eventual consistency is maintained.</p>
<div class="figure" id="id2">
<img alt="Hinted Handoff Example" src="../_images/hints.svg" /><p class="caption"><span class="caption-text">Hinted Handoff in Action</span></p>
</div>
<ul class="simple">
<li>(<code class="docutils literal notranslate"><span class="pre">t0</span></code>): The write is sent by the client, and the coordinator sends it
to the three replicas. Unfortunately <code class="docutils literal notranslate"><span class="pre">replica_2</span></code> is restarting and cannot
receive the mutation.</li>
<li>(<code class="docutils literal notranslate"><span class="pre">t1</span></code>): The client receives a quorum acknowledgement from the coordinator.
At this point the client believe the write to be durable and visible to reads
(which it is).</li>
<li>(<code class="docutils literal notranslate"><span class="pre">t2</span></code>): After the write timeout (default <code class="docutils literal notranslate"><span class="pre">2s</span></code>), the coordinator decides
that <code class="docutils literal notranslate"><span class="pre">replica_2</span></code> is unavailable and stores a hint to its local disk.</li>
<li>(<code class="docutils literal notranslate"><span class="pre">t3</span></code>): Later, when <code class="docutils literal notranslate"><span class="pre">replica_2</span></code> starts back up it sends a gossip message
to all nodes, including the coordinator.</li>
<li>(<code class="docutils literal notranslate"><span class="pre">t4</span></code>): The coordinator replays hints including the missed mutation
against <code class="docutils literal notranslate"><span class="pre">replica_2</span></code>.</li>
</ul>
<p>If the node does not return in time, the destination replica will be
permanently out of sync until either read-repair or full/incremental
anti-entropy repair propagates the mutation.</p>
<div class="section" id="application-of-hints">
<h3>Application of Hints<a class="headerlink" href="#application-of-hints" title="Permalink to this headline">¶</a></h3>
<p>Hints are streamed in bulk, a segment at a time, to the target replica node and
the target node replays them locally. After the target node has replayed a
segment it deletes the segment and receives the next segment. This continues
until all hints are drained.</p>
</div>
<div class="section" id="storage-of-hints-on-disk">
<h3>Storage of Hints on Disk<a class="headerlink" href="#storage-of-hints-on-disk" title="Permalink to this headline">¶</a></h3>
<p>Hints are stored in flat files in the coordinator node’s
<code class="docutils literal notranslate"><span class="pre">$CASSANDRA_HOME/data/hints</span></code> directory. A hint includes a hint id, the target
replica node on which the mutation is meant to be stored, the serialized
mutation (stored as a blob) that couldn’t be delivered to the replica node, the
mutation timestamp, and the Cassandra version used to serialize the mutation.
By default hints are compressed using <code class="docutils literal notranslate"><span class="pre">LZ4Compressor</span></code>. Multiple hints are
appended to the same hints file.</p>
<p>Since hints contain the original unmodified mutation timestamp, hint application
is idempotent and cannot overwrite a future mutation.</p>
</div>
<div class="section" id="hints-for-timed-out-write-requests">
<h3>Hints for Timed Out Write Requests<a class="headerlink" href="#hints-for-timed-out-write-requests" title="Permalink to this headline">¶</a></h3>
<p>Hints are also stored for write requests that time out. The
<code class="docutils literal notranslate"><span class="pre">write_request_timeout_in_ms</span></code> setting in <code class="docutils literal notranslate"><span class="pre">cassandra.yaml</span></code> configures the
timeout for write requests.</p>
<div class="highlight-none notranslate"><div class="highlight"><pre><span></span>write_request_timeout_in_ms: 2000
</pre></div>
</div>
<p>The coordinator waits for the configured amount of time for write requests to
complete, at which point it will time out and generate a hint for the timed out
request. The lowest acceptable value for <code class="docutils literal notranslate"><span class="pre">write_request_timeout_in_ms</span></code> is 10 ms.</p>
</div>
</div>
<div class="section" id="configuring-hints">
<h2>Configuring Hints<a class="headerlink" href="#configuring-hints" title="Permalink to this headline">¶</a></h2>
<p>Hints are enabled by default as they are critical for data consistency. The
<code class="docutils literal notranslate"><span class="pre">cassandra.yaml</span></code> configuration file provides several settings for configuring
hints:</p>
<p>Table 1. Settings for Hints</p>
<table border="1" class="docutils">
<colgroup>
<col width="37%" />
<col width="36%" />
<col width="26%" />
</colgroup>
<tbody valign="top">
<tr class="row-odd"><td>Setting</td>
<td>Description</td>
<td>Default Value</td>
</tr>
<tr class="row-even"><td><code class="docutils literal notranslate"><span class="pre">hinted_handoff_enabled</span></code></td>
<td>Enables/Disables hinted handoffs</td>
<td><code class="docutils literal notranslate"><span class="pre">true</span></code></td>
</tr>
<tr class="row-odd"><td><code class="docutils literal notranslate"><span class="pre">hinted_handoff_disabled_datacenters</span></code></td>
<td><p class="first">A list of data centers that do not perform
hinted handoffs even when handoff is
otherwise enabled.
Example:</p>
<blockquote class="last">
<div><div class="highlight-yaml notranslate"><div class="highlight"><pre><span></span><span class="nt">hinted_handoff_disabled_datacenters</span><span class="p">:</span>
  <span class="p p-Indicator">-</span> <span class="l l-Scalar l-Scalar-Plain">DC1</span>
  <span class="p p-Indicator">-</span> <span class="l l-Scalar l-Scalar-Plain">DC2</span>
</pre></div>
</div>
</div></blockquote>
</td>
<td><code class="docutils literal notranslate"><span class="pre">unset</span></code></td>
</tr>
<tr class="row-even"><td><code class="docutils literal notranslate"><span class="pre">max_hint_window_in_ms</span></code></td>
<td>Defines the maximum amount of time (ms)
a node shall have hints generated after it
has failed.</td>
<td><code class="docutils literal notranslate"><span class="pre">10800000</span></code> # 3 hours</td>
</tr>
<tr class="row-odd"><td><code class="docutils literal notranslate"><span class="pre">hinted_handoff_throttle_in_kb</span></code></td>
<td>Maximum throttle in KBs per second, per
delivery thread. This will be reduced
proportionally to the number of nodes in
the cluster.
(If there are two nodes in the cluster,
each delivery thread will use the maximum
rate; if there are 3, each will throttle
to half of the maximum,since it is expected
for two nodes to be delivering hints
simultaneously.)</td>
<td><code class="docutils literal notranslate"><span class="pre">1024</span></code></td>
</tr>
<tr class="row-even"><td><code class="docutils literal notranslate"><span class="pre">max_hints_delivery_threads</span></code></td>
<td>Number of threads with which to deliver
hints; Consider increasing this number when
you have multi-dc deployments, since
cross-dc handoff tends to be slower</td>
<td><code class="docutils literal notranslate"><span class="pre">2</span></code></td>
</tr>
<tr class="row-odd"><td><code class="docutils literal notranslate"><span class="pre">hints_directory</span></code></td>
<td>Directory where Cassandra stores hints.</td>
<td><code class="docutils literal notranslate"><span class="pre">$CASSANDRA_HOME/data/hints</span></code></td>
</tr>
<tr class="row-even"><td><code class="docutils literal notranslate"><span class="pre">hints_flush_period_in_ms</span></code></td>
<td>How often hints should be flushed from the
internal buffers to disk. Will <em>not</em>
trigger fsync.</td>
<td><code class="docutils literal notranslate"><span class="pre">10000</span></code></td>
</tr>
<tr class="row-odd"><td><code class="docutils literal notranslate"><span class="pre">max_hints_file_size_in_mb</span></code></td>
<td>Maximum size for a single hints file, in
megabytes.</td>
<td><code class="docutils literal notranslate"><span class="pre">128</span></code></td>
</tr>
<tr class="row-even"><td><code class="docutils literal notranslate"><span class="pre">hints_compression</span></code></td>
<td>Compression to apply to the hint files.
If omitted, hints files will be written
uncompressed. LZ4, Snappy, and Deflate
compressors are supported.</td>
<td><code class="docutils literal notranslate"><span class="pre">LZ4Compressor</span></code></td>
</tr>
</tbody>
</table>
</div>
<div class="section" id="configuring-hints-at-runtime-with-nodetool">
<h2>Configuring Hints at Runtime with <code class="docutils literal notranslate"><span class="pre">nodetool</span></code><a class="headerlink" href="#configuring-hints-at-runtime-with-nodetool" title="Permalink to this headline">¶</a></h2>
<p><code class="docutils literal notranslate"><span class="pre">nodetool</span></code> provides several commands for configuring hints or getting hints
related information. The nodetool commands override the corresponding
settings if any in <code class="docutils literal notranslate"><span class="pre">cassandra.yaml</span></code> for the node running the command.</p>
<p>Table 2. Nodetool Commands for Hints</p>
<table border="1" class="docutils">
<colgroup>
<col width="43%" />
<col width="57%" />
</colgroup>
<tbody valign="top">
<tr class="row-odd"><td>Command</td>
<td>Description</td>
</tr>
<tr class="row-even"><td><code class="docutils literal notranslate"><span class="pre">nodetool</span> <span class="pre">disablehandoff</span></code></td>
<td>Disables storing and delivering hints</td>
</tr>
<tr class="row-odd"><td><code class="docutils literal notranslate"><span class="pre">nodetool</span> <span class="pre">disablehintsfordc</span></code></td>
<td>Disables storing and delivering hints to a
data center</td>
</tr>
<tr class="row-even"><td><code class="docutils literal notranslate"><span class="pre">nodetool</span> <span class="pre">enablehandoff</span></code></td>
<td>Re-enables future hints storing and
delivery on the current node</td>
</tr>
<tr class="row-odd"><td><code class="docutils literal notranslate"><span class="pre">nodetool</span> <span class="pre">enablehintsfordc</span></code></td>
<td>Enables hints for a data center that was
previously disabled</td>
</tr>
<tr class="row-even"><td><code class="docutils literal notranslate"><span class="pre">nodetool</span> <span class="pre">getmaxhintwindow</span></code></td>
<td>Prints the max hint window in ms. New in
Cassandra 4.0.</td>
</tr>
<tr class="row-odd"><td><code class="docutils literal notranslate"><span class="pre">nodetool</span> <span class="pre">handoffwindow</span></code></td>
<td>Prints current hinted handoff window</td>
</tr>
<tr class="row-even"><td><code class="docutils literal notranslate"><span class="pre">nodetool</span> <span class="pre">pausehandoff</span></code></td>
<td>Pauses hints delivery process</td>
</tr>
<tr class="row-odd"><td><code class="docutils literal notranslate"><span class="pre">nodetool</span> <span class="pre">resumehandoff</span></code></td>
<td>Resumes hints delivery process</td>
</tr>
<tr class="row-even"><td><code class="docutils literal notranslate"><span class="pre">nodetool</span>
<span class="pre">sethintedhandoffthrottlekb</span></code></td>
<td>Sets hinted handoff throttle in kb
per second, per delivery thread</td>
</tr>
<tr class="row-odd"><td><code class="docutils literal notranslate"><span class="pre">nodetool</span> <span class="pre">setmaxhintwindow</span></code></td>
<td>Sets the specified max hint window in ms</td>
</tr>
<tr class="row-even"><td><code class="docutils literal notranslate"><span class="pre">nodetool</span> <span class="pre">statushandoff</span></code></td>
<td>Status of storing future hints on the
current node</td>
</tr>
<tr class="row-odd"><td><code class="docutils literal notranslate"><span class="pre">nodetool</span> <span class="pre">truncatehints</span></code></td>
<td>Truncates all hints on the local node, or
truncates hints for the endpoint(s)
specified.</td>
</tr>
</tbody>
</table>
<div class="section" id="make-hints-play-faster-at-runtime">
<h3>Make Hints Play Faster at Runtime<a class="headerlink" href="#make-hints-play-faster-at-runtime" title="Permalink to this headline">¶</a></h3>
<p>The default of <code class="docutils literal notranslate"><span class="pre">1024</span> <span class="pre">kbps</span></code> handoff throttle is conservative for most modern
networks, and it is entirely possible that in a simple node restart you may
accumulate many gigabytes hints that may take hours to play back. For example if
you are ingesting <code class="docutils literal notranslate"><span class="pre">100</span> <span class="pre">Mbps</span></code> of data per node, a single 10 minute long
restart will create <code class="docutils literal notranslate"><span class="pre">10</span> <span class="pre">minutes</span> <span class="pre">*</span> <span class="pre">(100</span> <span class="pre">megabit</span> <span class="pre">/</span> <span class="pre">second)</span> <span class="pre">~=</span> <span class="pre">7</span> <span class="pre">GiB</span></code> of data
which at <code class="docutils literal notranslate"><span class="pre">(1024</span> <span class="pre">KiB</span> <span class="pre">/</span> <span class="pre">second)</span></code> would take <code class="docutils literal notranslate"><span class="pre">7.5</span> <span class="pre">GiB</span> <span class="pre">/</span> <span class="pre">(1024</span> <span class="pre">KiB</span> <span class="pre">/</span> <span class="pre">second)</span> <span class="pre">=</span>
<span class="pre">2.03</span> <span class="pre">hours</span></code> to play back. The exact math depends on the load balancing strategy
(round robin is better than token aware), number of tokens per node (more
tokens is better than fewer), and naturally the cluster’s write rate, but
regardless you may find yourself wanting to increase this throttle at runtime.</p>
<p>If you find yourself in such a situation, you may consider raising
the <code class="docutils literal notranslate"><span class="pre">hinted_handoff_throttle</span></code> dynamically via the
<code class="docutils literal notranslate"><span class="pre">nodetool</span> <span class="pre">sethintedhandoffthrottlekb</span></code> command.</p>
</div>
<div class="section" id="allow-a-node-to-be-down-longer-at-runtime">
<h3>Allow a Node to be Down Longer at Runtime<a class="headerlink" href="#allow-a-node-to-be-down-longer-at-runtime" title="Permalink to this headline">¶</a></h3>
<p>Sometimes a node may be down for more than the normal <code class="docutils literal notranslate"><span class="pre">max_hint_window_in_ms</span></code>,
(default of three hours), but the hardware and data itself will still be
accessible.  In such a case you may consider raising the
<code class="docutils literal notranslate"><span class="pre">max_hint_window_in_ms</span></code> dynamically via the <code class="docutils literal notranslate"><span class="pre">nodetool</span> <span class="pre">setmaxhintwindow</span></code>
command added in Cassandra 4.0 (<a class="reference external" href="https://issues.apache.org/jira/browse/CASSANDRA-11720">CASSANDRA-11720</a>).
This will instruct Cassandra to continue holding hints for the down
endpoint for a longer amount of time.</p>
<p>This command should be applied on all nodes in the cluster that may be holding
hints. If needed, the setting can be applied permanently by setting the
<code class="docutils literal notranslate"><span class="pre">max_hint_window_in_ms</span></code> setting in <code class="docutils literal notranslate"><span class="pre">cassandra.yaml</span></code> followed by a rolling
restart.</p>
</div>
</div>
<div class="section" id="monitoring-hint-delivery">
<h2>Monitoring Hint Delivery<a class="headerlink" href="#monitoring-hint-delivery" title="Permalink to this headline">¶</a></h2>
<p>Cassandra 4.0 adds histograms available to understand how long it takes to deliver
hints which is useful for operators to better identify problems (<a class="reference external" href="https://issues.apache.org/jira/browse/CASSANDRA-13234">CASSANDRA-13234</a>).</p>
<p>There are also metrics available for tracking <a class="reference internal" href="metrics.html#handoff-metrics"><span class="std std-ref">Hinted Handoff</span></a>
and <a class="reference internal" href="metrics.html#hintsservice-metrics"><span class="std std-ref">Hints Service</span></a> metrics.</p>
</div>
</div>


          <div class="doc-prev-next-links" role="navigation" aria-label="footer navigation">
            
            <a href="compaction/index.html" class="btn btn-default pull-right " role="button" title="Compaction" accesskey="n">Next <span class="glyphicon glyphicon-circle-arrow-right" aria-hidden="true"></span></a>
            
            
            <a href="read_repair.html" class="btn btn-default" role="button" title="Read repair" accesskey="p"><span class="glyphicon glyphicon-circle-arrow-left" aria-hidden="true"></span> Previous</a>
            
          </div>
          
        </div>
      </div>
    </div>
  </div>
</div>