Google Git
Sign in
apache / mesos-site / f23230b4b4834dce22570caef1c6d9f46b227655 / . / publish / documentation / quota / index.html
blob: 01779e9ca9709587726dc8fedae3308ae5b873f6 [file] [log] [blame]
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<title>Apache Mesos - Quota</title>
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<meta property="og:locale" content="en_US"/>
<meta property="og:type" content="website"/>
<meta property="og:title" content="Apache Mesos"/>
<meta property="og:site_name" content="Apache Mesos"/>
<meta property="og:url" content="http://mesos.apache.org/"/>
<meta property="og:image" content="http://mesos.apache.org/assets/img/mesos_logo_fb_preview.png"/>
<meta property="og:description"
content="Apache Mesos abstracts resources away from machines,
enabling fault-tolerant and elastic distributed systems
to easily be built and run effectively."/>
<meta name="twitter:card" content="summary"/>
<meta name="twitter:site" content="@ApacheMesos"/>
<meta name="twitter:title" content="Apache Mesos"/>
<meta name="twitter:image" content="http://mesos.apache.org/assets/img/mesos_logo_fb_preview.png"/>
<meta name="twitter:description"
content="Apache Mesos abstracts resources away from machines,
enabling fault-tolerant and elastic distributed systems
to easily be built and run effectively."/>
<link href="//netdna.bootstrapcdn.com/bootstrap/3.1.1/css/bootstrap.min.css" rel="stylesheet">
<link rel="alternate" type="application/atom+xml" title="Apache Mesos Blog" href="/blog/feed.xml">
<link href="../../assets/css/main.css" media="screen" rel="stylesheet" type="text/css" />
<!-- Google Analytics Magic -->
<script type="text/javascript">
var _gaq = _gaq || [];
_gaq.push(['_setAccount', 'UA-20226872-1']);
_gaq.push(['_setDomainName', 'apache.org']);
_gaq.push(['_trackPageview']);
(function() {
var ga = document.createElement('script'); ga.type = 'text/javascript'; ga.async = true;
ga.src = ('https:' == document.location.protocol ? 'https://ssl' : 'http://www') + '.google-analytics.com/ga.js';
var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(ga, s);
})();
</script>
</head>
<body>
<!-- magical breadcrumbs -->
<div class="topnav">
<div class="container">
<ul class="breadcrumb">
<li>
<div class="dropdown">
<a data-toggle="dropdown" href="#">Apache Software Foundation <span class="caret"></span></a>
<ul class="dropdown-menu" role="menu" aria-labelledby="dLabel">
<li><a href="http://www.apache.org">Apache Homepage</a></li>
<li><a href="http://www.apache.org/licenses/">License</a></li>
<li><a href="http://www.apache.org/foundation/sponsorship.html">Sponsorship</a></li>
<li><a href="http://www.apache.org/foundation/thanks.html">Thanks</a></li>
<li><a href="http://www.apache.org/security/">Security</a></li>
</ul>
</div>
</li>
<li><a href="http://mesos.apache.org">Apache Mesos</a></li>
<li><a href="/documentation
/">Documentation
</a></li>
</ul><!-- /.breadcrumb -->
</div><!-- /.container -->
</div><!-- /.topnav -->
<!-- navbar excitement -->
<div class="navbar navbar-default navbar-static-top" role="navigation">
<div class="container">
<div class="navbar-header">
<button type="button" class="navbar-toggle collapsed" data-toggle="collapse" data-target="#mesos-menu" aria-expanded="false">
<span class="sr-only">Toggle navigation</span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
<a class="navbar-brand" href="/"><img src="/assets/img/mesos_logo.png" alt="Apache Mesos logo"/></a>
</div><!-- /.navbar-header -->
<div class="navbar-collapse collapse" id="mesos-menu">
<ul class="nav navbar-nav navbar-right">
<li><a href="/gettingstarted/">Getting Started</a></li>
<li><a href="/blog/">Blog</a></li>
<li><a href="/documentation/latest/">Documentation</a></li>
<li><a href="/downloads/">Downloads</a></li>
<li><a href="/community/">Community</a></li>
</ul>
</div><!-- /#mesos-menu -->
</div><!-- /.container -->
</div><!-- /.navbar -->
<div class="content">
<div class="container">
<div class="row-fluid">
<div class="col-md-4">
<h4>If you're new to Mesos</h4>
<p>See the <a href="/gettingstarted/">getting started</a> page for more
information about downloading, building, and deploying Mesos.</p>
<h4>If you'd like to get involved or you're looking for support</h4>
<p>See our <a href="/community/">community</a> page for more details.</p>
</div>
<div class="col-md-8">
<h1>Quota</h1>
<p>A problem when running multiple frameworks on Mesos is that the default
<a href="https://www.cs.berkeley.edu/~alig/papers/drf.pdf">wDRF allocator</a> may offer
resources to frameworks even though they are beyond their fair share (e.g., when
no other framework currently accepts these resources). There is no mechanism to
lay away resources for future consumption in an entire cluster. Even though
<a href="/documentation/latest/./reservation/">dynamic reservations</a> allow both operators and frameworks
to dynamically reserve resources on particular Mesos agents, it does not solve
the above problem because any individual agents might fail.</p>
<p>Mesos 0.27 introduced support for <em>quotas</em>, which are a mechanism for
guaranteeing that a <a href="/documentation/latest/./roles/">role</a> will receive a certain minimum amount of
resources. A quota specifies a <em>minimum</em> amount of resources that the role is
guaranteed to receive (unless the total resources in the cluster are less than
the configured quota resources, which often indicates a misconfiguration). Of
course, the role might receive offers for more resources than its configured
quota, as dictated by the default Mesos DRF allocation scheme.</p>
<p>Quotas are somewhat similar to <a href="/documentation/latest/./reservation/">reserved resources</a>, but they
have several differences. Most importantly, quota resources are not tied to a
particular agent&mdash;that is, a quota ensures that a role will receive a certain
amount of resources somewhere in the cluster, whereas reservations can be used
to assign specific resources at a particular agent to a given role. Quotas can
only be configured by operators, via the HTTP endpoint described below; dynamic
reservations can be made by frameworks, provided the framework&rsquo;s principal is
<a href="/documentation/latest/./authorization/">authorized</a> to make reservations.</p>
<p>Note that reserved resources are considered to satisfy a role&rsquo;s quota. For
example, if a role has been assigned a quota of 4 CPUs and also has 2 reserved
resources at a particular agent, the role has a minimum guarantee of 4 CPUs, not
6.</p>
<h2>Terminology</h2>
<p>For the purpose of this document, an &ldquo;Operator&rdquo; is a person, tool, or script
that manages the Mesos cluster.</p>
<p>In computer science, a &ldquo;quota&rdquo; usually refers to one of the following:</p>
<ul>
<li>A minimal guarantee.</li>
<li>A maximal limit.</li>
<li>A pair of both.</li>
</ul>
<p>In Mesos, a quota is a <strong>guaranteed</strong> resource allocation that a role may rely
on; in other words, a minimum share a role is entitled to receive.</p>
<h2>Motivation and Limitations</h2>
<p>Consider the following scenarios in a Mesos cluster to better understand which
use-cases are supported by quota, and which are not.</p>
<h3>Scenario 1: Greedy Framework</h3>
<p>There are two frameworks in a cluster, each running in a separate role with
equal weights: framework fA in role rA and framework fB in role rB. There is a
single resource available in the cluster: 100 CPUs. fA consumes 10 CPUs and is
idle (declines resource offers), while fB is greedy and accepts all offers it gets,
hogging the remaining 90 CPUs. Without quota, though fA&rsquo;s fair share is 50 CPUs
it will not be able to make use of additional the 40 CPUs until some of fB&rsquo;s
tasks terminate.</p>
<h3>Scenario 2: Resources for a new Framework</h3>
<p>A greedy framework fB in role rB is currently the only framework in the cluster
and it uses all available resources&mdash;100 CPUs. If a new framework fA in role rA
joins the cluster, it will not receive its fair share of the cluster resources
(50 CPUs) until some of fB&rsquo;s tasks terminate.</p>
<p>To deal with Scenario 2, quota by itself is not a sufficient solution as it
would be set after fB has started using all resources. Instead Scenario 2
requires either always keeping a pool of resources which are not offered or
introducing preemption for running tasks.</p>
<h2>Operator HTTP Endpoint</h2>
<p>The master <a href="/documentation/latest/./endpoints/master/quota/">/quota</a> HTTP endpoint enables operators
to configure quotas. The endpoint currently offers a REST-like interface and
supports the following operations:</p>
<ul>
<li><a href="#setRequest">Setting</a> a new quota with POST.</li>
<li><a href="#removeRequest">Removing</a> an existing quota with DELETE.</li>
<li><a href="#statusRequest">Querying</a> the currently set quota with GET.</li>
</ul>
<p>Currently it is not possible to update previously configured quotas. This means
in order to update a quota for a given role, the operator has to remove the
existing quota and then set a new one.</p>
<p>The endpoint can optionally use authentication and authorization. See
<a href="/documentation/latest/./authentication/">authentication guide</a> for details.</p>
<p><a name="setRequest"></a></p>
<h3>Set</h3>
<p>The operator can set a new quota by sending an HTTP POST request to the <code>/quota</code>
endpoint.</p>
<p>An example request to the quota endpoint could look like this (using the JSON
file below):</p>
<pre><code> $ curl -d @quota.json -X POST http://&lt;master-ip&gt;:&lt;port&gt;/quota
</code></pre>
<p>For example to set a quota of 12 CPUs and 6144 MB of RAM for <code>role1</code> the operator
can use the following <code>quota.json</code>:</p>
<pre><code> {
"role": "role1",
"guarantee": [
{
"name": "cpus",
"type": "SCALAR",
"scalar": { "value": 12 }
},
{
"name": "mem",
"type": "SCALAR",
"scalar": { "value": 6144 }
}
]
}
</code></pre>
<p>A set request is only valid for roles for which no quota is currently set.
However if the master is configured without an explicit
<a href="/documentation/latest/./roles/">role whitelist</a>, a set request can introduce new roles.</p>
<p>In order to bypass the <a href="#capacityHeuristic">capacity heuristic</a> check the
operator should set an optional <code>force</code> field:</p>
<pre><code> {
"force": true,
"role": "role1",
...
}
</code></pre>
<p>The operator will receive one of the following HTTP response codes:</p>
<ul>
<li><code>200 OK</code>: Success (the set request was successful).</li>
<li><code>400 BadRequest</code>: Invalid arguments (e.g., quota for role exists, invalid
JSON, non-scalar resources included).</li>
<li><code>401 Unauthorized</code>: Unauthenticated request.</li>
<li><code>403 Forbidden</code>: Unauthorized request.</li>
<li><code>409 Conflict</code>: The capacity heuristic check failed due to insufficient
resources.</li>
</ul>
<p><a name="removeRequest"></a></p>
<h3>Remove</h3>
<p>The operator can remove a previously set quota by sending an HTTP DELETE request
to the <code>/quota/&lt;role&gt;</code> endpoint. For example the following command removes
a previously set quota for <code>role1</code>:</p>
<pre><code> $ curl -X DELETE http://&lt;master-ip&gt;:&lt;port&gt;/quota/role1
</code></pre>
<p>The operator will receive one of the following HTTP response codes:</p>
<ul>
<li><code>200 OK</code>: Success (the remove request was successful).</li>
<li><code>400 BadRequest</code>: Invalid arguments (e.g., removing a quota for a role which
does not have any quota set).</li>
<li><code>401 Unauthorized</code>: Unauthenticated request.</li>
<li><code>403 Forbidden</code>: Unauthorized request.</li>
</ul>
<p><a name="statusRequest"></a></p>
<h3>Status</h3>
<p>The operator can query the configured quotas by sending an HTTP GET request
to the <code>/quota</code> endpoint.</p>
<pre><code> $ curl -X GET http://&lt;master-ip&gt;:&lt;port&gt;/quota
</code></pre>
<p>The response message body includes a JSON representation of the current quota
status for role(s) which principal is authorized to query quota status (if
authorization is enabled). For example:</p>
<pre><code> {
"infos": [
{
"role": "role1",
"guarantee": [
{
"name": "cpus",
"role": "*",
"type": "SCALAR",
"scalar": { "value": 12 }
},
{
"name": "mem",
"role": "*",
"type": "SCALAR",
"scalar": { "value": 6144 }
}
]
}
]
}
</code></pre>
<p>The operator will receive one of the following HTTP response codes:</p>
<ul>
<li><code>200 OK</code>: Success.</li>
<li><code>401 Unauthorized</code>: Unauthenticated request.</li>
</ul>
<p><strong>NOTE:</strong> If the principal is not authorized to query quota status for certain
role(s), the result will not include corresponding quota information.</p>
<h2>How Does It Work?</h2>
<p>There are several stages in the lifetime of a quota issued by operator. First
the <a href="#requestProcessing">quota set request is handled by the master</a>, after that
the <a href="#allocatorEnforcement">allocator enforces the quota</a>.
Quotas can be <a href="#removeProcessing">removed</a> by the operator.</p>
<p>It is important to understand that the enforcement of quota depends on
the allocator being used. A custom allocator could choose to handle quota in its
own way or even to ignore quota.
<strong>The following section assumes the default
<a href="https://www.cs.berkeley.edu/~alig/papers/drf.pdf">wDRF</a> allocator is used.</strong></p>
<p>Be aware that setting quota may affect other frameworks in the cluster,
because resources will be laid away and not offered to other frameworks.
Also note, that quota is only applicable for scalar resources (e.g., it is not
possible to set quota for port resources).</p>
<p><a name="requestProcessing"></a></p>
<h2>Quota Request Processing</h2>
<p>When an operator submits a quota set request via the master <code>/quota</code> HTTP
endpoint, the following steps are triggered:</p>
<ol>
<li><a href="/documentation/latest/./authentication/">Authenticate</a> the HTTP request.</li>
<li>Parse and validate the request.
See <a href="#setRequest">description of possible error codes</a>.</li>
<li><a href="/documentation/latest/./authentication/">Authorize</a> the HTTP request if authorization is enabled.</li>
<li>Run the <a href="#capacityHeuristic">capacity heuristic</a> if not disabled by
<a href="#setRequest">the <code>force</code> flag</a>.</li>
<li>Reliably store quota. See <a href="#failover">details on failover recovery</a>.</li>
<li><a href="#rescindOffers">Rescind outstanding offers</a>.</li>
</ol>
<p><a name="removeProcessing"></a>
The quota remove request processing is simpler and triggers the following steps:</p>
<ol>
<li><a href="/documentation/latest/./authentication/">Authenticate</a> the HTTP request.</li>
<li>Validate the request.
See <a href="#removeRequest">description of potential error codes</a>.</li>
<li><a href="/documentation/latest/./authentication/">Authorize</a> the HTTP request if authorization is enabled.</li>
<li>Reliably remove quota.</li>
</ol>
<p><a name="capacityHeuristic"></a></p>
<h3>Capacity Heuristic Check</h3>
<p>Misconfigured quota can render a cluster into a state where no offers are made
to any frameworks. For example imagine an operator setting quota 1000 CPUs
for a role <code>prosuction</code> (note the typo) in a cluster with a total capacity of
100 CPUs. In that case after the quota is accepted by the master, no offers will
be made to any framework in any actual role (including <code>production</code>).</p>
<p>In order to prevent such extreme situations, the Mesos Master employs a capacity
heuristic check that ensures a quota set request can reasonably be satisfied
given the total cluster capacity. This heuristic tests whether the total quota,
including the new request, does not exceed the sum of total
non-statically-reserved cluster resources, i.e. the following inequality holds:</p>
<pre><code> total resources - statically reserved &gt;= total quota + quota request
</code></pre>
<p>Please be advised that even if there are enough resources at the moment of
this check, agents may terminate at any time, rendering the cluster incapable
of satisfying the configured quotas.</p>
<p>A <a href="#setRequest"><code>force</code> flag</a> can be set to bypass this check. For example, this
flag can be useful when the operator would like to configure a quota that
exceeds the current cluster capacity, but they know that additional cluster
resources will be added shortly.</p>
<p><a name="rescindOffers"></a></p>
<h3>Rescinding Outstanding Offers</h3>
<p>When setting a new quota, the master rescinds outstanding offers. This avoids
situations where the quota request cannot be satisfied by the remaining
unoffered resources, but there are enough resources tied up in outstanding
offers to frameworks that have not accepted them yet. Hence, we rescind
outstanding offers with the following rules:</p>
<ul>
<li>Rescind at least as many resources as there are in the quota request.</li>
<li>If at least one offer is to be rescinded from an agent, all offers from this
agent are rescinded. This is done in order to make the potential offer bigger,
which increases the chances that a quota'ed framework will be able to use the
offer.</li>
<li>Rescind offers from at least as many agents as there are frameworks in the
role for which quota is being set. This enables (but does not guarantee, due
to fair sharing) each framework in the role to receive an offer.</li>
</ul>
<p><a name="allocatorEnforcement"></a></p>
<h2>Enforcement by wDRF Allocator</h2>
<p>The wDRF allocator first allocates (or lays away if offers are declined)
resources to framework in roles with quota set. Once all quotas are
satisfied, it proceeds with the standard wDRF for all frameworks.</p>
<p><strong>NOTE:</strong> A quota'ed role may not be allocated any unreserved non-revocable
resources beyond its quota guarantee. If frameworks in the quota'ed role have
not opted for revocable resources, they may stop getting offers once quota for
the role is satisfied. In this case setting quota to any value that is less than
the role&rsquo;s fair share may reduce the amount of resources offered to this role.</p>
<p><strong>NOTE:</strong> Currently quota guarantee also serves as quota limit, i.e. once quota
for the role is satisfied, frameworks in this role will not be offered any
resources except those reserved for the role. This behavior aims to mitigate the
absence of quota limit and will be changed in future releases.</p>
<p>If there are multiple frameworks in a role with quota set, the standard wDRF
algorithm determines framework priority inside this role.</p>
<p>The default wDRF allocator considers only non-revocable resources as applicable
towards quota.</p>
<p><a name="failover"></a></p>
<h2>Failover</h2>
<p>If there is at least one role with quota set, the master failover recovery
changes significantly. The reason for this is that during the recovery there is
a period of time when not all agents have registered with the Master. Therefore
it is impossible to reason about whether all quotas are satisfied.</p>
<p>To address this issue, if upon recovery any previously set quota are detected,
the allocator enters recovery mode, during which the allocator
<em>does not issue offers</em>. The recovery mode&mdash;and therefore offer
suspension&mdash;ends when either:</p>
<ul>
<li>A certain amount of agents reregister (by default 80% of agents known before
the failover), or</li>
<li>a timeout expires (by default 10 minutes).</li>
</ul>
<h2>Current Limitations</h2>
<ul>
<li>The quota set request does not allow specifying the granularity of the
requested resources (e.g., 10 CPUs on a single node).</li>
<li>The quota set request does not allow to specify constraints (e.g., 2*5 cpus
on disjoint nodes for an HA like setup).</li>
<li>Quota is not allowed for the default role &lsquo;*&rsquo; (see
<a href="https://issues.apache.org/jira/browse/MESOS-3938">MESOS-3938</a>).</li>
<li>Currently it is not possible to update previously configured quotas. See
<a href="#setRequest">quota set request</a> for details.</li>
</ul>
</div>
</div>
</div><!-- /.container -->
</div><!-- /.content -->
<hr>
<!-- footer -->
<div class="footer">
<div class="container">
<div class="col-md-4 social-blk">
<span class="social">
<a href="https://twitter.com/ApacheMesos"
class="twitter-follow-button"
data-show-count="false" data-size="large">Follow @ApacheMesos</a>
<script>!function(d,s,id){var js,fjs=d.getElementsByTagName(s)[0],p=/^http:/.test(d.location)?'http':'https';if(!d.getElementById(id)){js=d.createElement(s);js.id=id;js.src=p+'://platform.twitter.com/widgets.js';fjs.parentNode.insertBefore(js,fjs);}}(document, 'script', 'twitter-wjs');</script>
<a href="https://twitter.com/intent/tweet?button_hashtag=mesos"
class="twitter-hashtag-button"
data-size="large"
data-related="ApacheMesos">Tweet #mesos</a>
<script>!function(d,s,id){var js,fjs=d.getElementsByTagName(s)[0],p=/^http:/.test(d.location)?'http':'https';if(!d.getElementById(id)){js=d.createElement(s);js.id=id;js.src=p+'://platform.twitter.com/widgets.js';fjs.parentNode.insertBefore(js,fjs);}}(document, 'script', 'twitter-wjs');</script>
</span>
</div>
<div class="col-md-8 trademark">
<p>&copy; 2012-2017 <a href="http://apache.org">The Apache Software Foundation</a>.
Apache Mesos, the Apache feather logo, and the Apache Mesos project logo are trademarks of The Apache Software Foundation.
<p>
</div>
</div><!-- /.container -->
</div><!-- /.footer -->
<!-- JS -->
<script src="//code.jquery.com/jquery-1.11.0.min.js" type="text/javascript"></script>
<script src="//netdna.bootstrapcdn.com/bootstrap/3.1.1/js/bootstrap.min.js" type="text/javascript"></script>
</body>
</html>