/usr/share/doc/ganeti/html/design-htools-2.3.html is in ganeti-doc 2.16.0~rc2-1build1.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 | <!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" lang="en">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
<title>Synchronising htools to Ganeti 2.3 — Ganeti 2.16.0~rc2 documentation</title>
<link rel="stylesheet" href="_static/style.css" type="text/css" />
<link rel="stylesheet" href="_static/pygments.css" type="text/css" />
<script type="text/javascript">
var DOCUMENTATION_OPTIONS = {
URL_ROOT: './',
VERSION: '2.16.0~rc2',
COLLAPSE_INDEX: false,
FILE_SUFFIX: '.html',
HAS_SOURCE: true,
SOURCELINK_SUFFIX: '.txt'
};
</script>
<script type="text/javascript" src="_static/jquery.js"></script>
<script type="text/javascript" src="_static/underscore.js"></script>
<script type="text/javascript" src="_static/doctools.js"></script>
<link rel="search" title="Search" href="search.html" />
<link rel="next" title="Ganeti 2.4 design" href="design-2.4.html" />
<link rel="prev" title="Ganeti 2.3 design" href="design-2.3.html" />
</head>
<body>
<div class="related" role="navigation" aria-label="related navigation">
<h3>Navigation</h3>
<ul>
<li class="right" style="margin-right: 10px">
<a href="design-2.4.html" title="Ganeti 2.4 design"
accesskey="N">next</a></li>
<li class="right" >
<a href="design-2.3.html" title="Ganeti 2.3 design"
accesskey="P">previous</a> |</li>
<li class="nav-item nav-item-0"><a href="index.html">Ganeti 2.16.0~rc2 documentation</a> »</li>
</ul>
</div>
<div class="document">
<div class="documentwrapper">
<div class="bodywrapper">
<div class="body" role="main">
<div class="section" id="synchronising-htools-to-ganeti-2-3">
<h1>Synchronising htools to Ganeti 2.3<a class="headerlink" href="#synchronising-htools-to-ganeti-2-3" title="Permalink to this headline">¶</a></h1>
<table class="docutils field-list" frame="void" rules="none">
<col class="field-name" />
<col class="field-body" />
<tbody valign="top">
<tr class="field-odd field"><th class="field-name">Created:</th><td class="field-body">2011-Mar-18</td>
</tr>
<tr class="field-even field"><th class="field-name">Status:</th><td class="field-body">Implemented</td>
</tr>
<tr class="field-odd field"><th class="field-name">Ganeti-Version:</th><td class="field-body">2.5.0</td>
</tr>
</tbody>
</table>
<p>Ganeti 2.3 introduces a number of new features that change the cluster
internals significantly enough that the htools suite needs to be
updated accordingly in order to function correctly.</p>
<div class="section" id="shared-storage-support">
<h2>Shared storage support<a class="headerlink" href="#shared-storage-support" title="Permalink to this headline">¶</a></h2>
<p>Currently, the htools algorithms presume a model where all of an
instance’s resources is served from within the cluster, more
specifically from the nodes comprising the cluster. While is this
usual for memory and CPU, deployments which use shared storage will
invalidate this assumption for storage.</p>
<p>To account for this, we need to move some assumptions from being
implicit (and hardcoded) to being explicitly exported from Ganeti.</p>
<div class="section" id="new-instance-parameters">
<h3>New instance parameters<a class="headerlink" href="#new-instance-parameters" title="Permalink to this headline">¶</a></h3>
<p>It is presumed that Ganeti will export for all instances a new
<code class="docutils literal"><span class="pre">storage_type</span></code> parameter, that will denote either internal storage
(e.g. <em>plain</em> or <em>drbd</em>), or external storage.</p>
<p>Furthermore, a new <code class="docutils literal"><span class="pre">storage_pool</span></code> parameter will classify, for both
internal and external storage, the pool out of which the storage is
allocated. For internal storage, this will be either <code class="docutils literal"><span class="pre">lvm</span></code> (the pool
that provides space to both <code class="docutils literal"><span class="pre">plain</span></code> and <code class="docutils literal"><span class="pre">drbd</span></code> instances) or
<code class="docutils literal"><span class="pre">file</span></code> (for file-storage-based instances). For external storage,
this will be the respective NAS/SAN/cloud storage that backs up the
instance. Note that for htools, external storage pools are opaque; we
only care that they have an identifier, so that we can distinguish
between two different pools.</p>
<p>If these two parameters are not present, the instances will be
presumed to be <code class="docutils literal"><span class="pre">internal/lvm</span></code>.</p>
</div>
<div class="section" id="new-node-parameters">
<h3>New node parameters<a class="headerlink" href="#new-node-parameters" title="Permalink to this headline">¶</a></h3>
<p>For each node, it is expected that Ganeti will export what storage
types it supports and pools it has access to. So a classic 2.2 cluster
will have all nodes supporting <code class="docutils literal"><span class="pre">internal/lvm</span></code> and/or
<code class="docutils literal"><span class="pre">internal/file</span></code>, whereas a new shared storage only 2.3 cluster could
have <code class="docutils literal"><span class="pre">external/my-nas</span></code> storage.</p>
<p>Whatever the mechanism that Ganeti will use internally to configure
the associations between nodes and storage pools, we consider that
we’ll have available two node attributes inside htools: the list of internal
and external storage pools.</p>
</div>
<div class="section" id="external-storage-and-instances">
<h3>External storage and instances<a class="headerlink" href="#external-storage-and-instances" title="Permalink to this headline">¶</a></h3>
<p>Currently, for an instance we allow one cheap move type: failover to
the current secondary, if it is a healthy node, and four other
“expensive” (as in, including data copies) moves that involve changing
either the secondary or the primary node or both.</p>
<p>In presence of an external storage type, the following things will
change:</p>
<ul class="simple">
<li>the disk-based moves will be disallowed; this is already a feature
in the algorithm, controlled by a boolean switch, so adapting
external storage here will be trivial</li>
<li>instead of the current one secondary node, the secondaries will
become a list of potential secondaries, based on access to the
instance’s storage pool</li>
</ul>
<p>Except for this, the basic move algorithm remains unchanged.</p>
</div>
<div class="section" id="external-storage-and-nodes">
<h3>External storage and nodes<a class="headerlink" href="#external-storage-and-nodes" title="Permalink to this headline">¶</a></h3>
<p>Two separate areas will have to change for nodes and external storage.</p>
<p>First, then allocating instances (either as part of a move or a new
allocation), if the instance is using external storage, then the
internal disk metrics should be ignored (for both the primary and
secondary cases).</p>
<p>Second, the per-node metrics used in the cluster scoring must take
into account that nodes might not have internal storage at all, and
handle this as a well-balanced case (score 0).</p>
</div>
<div class="section" id="n-1-status">
<h3>N+1 status<a class="headerlink" href="#n-1-status" title="Permalink to this headline">¶</a></h3>
<p>Currently, computing the N+1 status of a node is simple:</p>
<ul class="simple">
<li>group the current secondary instances by their primary node, and
compute the sum of each instance group memory</li>
<li>choose the maximum sum, and check if it’s smaller than the current
available memory on this node</li>
</ul>
<p>In effect, computing the N+1 status is a per-node matter. However,
with shared storage, we don’t have secondary nodes, just potential
secondaries. Thus computing the N+1 status will be a cluster-level
matter, and much more expensive.</p>
<p>A simple version of the N+1 checks would be that for each instance
having said node as primary, we have enough memory in the cluster for
relocation. This means we would actually need to run allocation
checks, and update the cluster status from within allocation on one
node, while being careful that we don’t recursively check N+1 status
during this relocation, which is too expensive.</p>
<p>However, the shared storage model has some properties that changes the
rules of the computation. Speaking broadly (and ignoring hard
restrictions like tag based exclusion and CPU limits), the exact
location of an instance in the cluster doesn’t matter as long as
memory is available. This results in two changes:</p>
<ul class="simple">
<li>simply tracking the amount of free memory buckets is enough,
cluster-wide</li>
<li>moving an instance from one node to another would not change the N+1
status of any node, and only allocation needs to deal with N+1
checks</li>
</ul>
<p>Unfortunately, this very cheap solution fails in case of any other
exclusion or prevention factors.</p>
<p>TODO: find a solution for N+1 checks.</p>
</div>
</div>
<div class="section" id="node-groups-support">
<h2>Node groups support<a class="headerlink" href="#node-groups-support" title="Permalink to this headline">¶</a></h2>
<p>The addition of node groups has a small impact on the actual
algorithms, which will simply operate at node group level instead of
cluster level, but it requires the addition of new algorithms for
inter-node group operations.</p>
<p>The following two definitions will be used in the following
paragraphs:</p>
<dl class="docutils">
<dt>local group</dt>
<dd>The local group refers to a node’s own node group, or when speaking
about an instance, the node group of its primary node</dd>
<dt>regular cluster</dt>
<dd>A cluster composed of a single node group, or pre-2.3 cluster</dd>
<dt>super cluster</dt>
<dd>This term refers to a cluster which comprises multiple node groups,
as opposed to a 2.2 and earlier cluster with a single node group</dd>
</dl>
<p>In all the below operations, it’s assumed that Ganeti can gather the
entire super cluster state cheaply.</p>
<div class="section" id="balancing-changes">
<h3>Balancing changes<a class="headerlink" href="#balancing-changes" title="Permalink to this headline">¶</a></h3>
<p>Balancing will move from cluster-level balancing to group
balancing. In order to achieve a reasonable improvement in a super
cluster, without needing to keep state of what groups have been
already balanced previously, the balancing algorithm will run as
follows:</p>
<ol class="arabic simple">
<li>the cluster data is gathered</li>
<li>if this is a regular cluster, as opposed to a super cluster,
balancing will proceed normally as previously</li>
<li>otherwise, compute the cluster scores for all groups</li>
<li>choose the group with the worst score and see if we can improve it;
if not choose the next-worst group, so on</li>
<li>once a group has been identified, run the balancing for it</li>
</ol>
<p>Of course, explicit selection of a group will be allowed.</p>
<div class="section" id="super-cluster-operations">
<h4>Super cluster operations<a class="headerlink" href="#super-cluster-operations" title="Permalink to this headline">¶</a></h4>
<p>Beside the regular group balancing, in a super cluster we have more
operations.</p>
<div class="section" id="redistribution">
<h5>Redistribution<a class="headerlink" href="#redistribution" title="Permalink to this headline">¶</a></h5>
<p>In a regular cluster, once we run out of resources (offline nodes
which can’t be fully evacuated, N+1 failures, etc.) there is nothing
we can do unless nodes are added or instances are removed.</p>
<p>In a super cluster however, there might be resources available in
another group, so there is the possibility of relocating instances
between groups to re-establish N+1 success within each group.</p>
<p>One difficulty in the presence of both super clusters and shared
storage is that the move paths of instances are quite complicated;
basically an instance can move inside its local group, and to any
other groups which have access to the same storage type and storage
pool pair. In effect, the super cluster is composed of multiple
‘partitions’, each containing one or more groups, but a node is
simultaneously present in multiple partitions, one for each storage
type and storage pool it supports. As such, the interactions between
the individual partitions are too complex for non-trivial clusters to
assume we can compute a perfect solution: we might need to move some
instances using shared storage pool ‘A’ in order to clear some more
memory to accept an instance using local storage, which will further
clear more VCPUs in a third partition, etc. As such, we’ll limit
ourselves at simple relocation steps within a single partition.</p>
<p>Algorithm:</p>
<ol class="arabic">
<li><p class="first">read super cluster data, and exit if cluster doesn’t allow
inter-group moves</p>
</li>
<li><p class="first">filter out any groups that are “alone” in their partition
(i.e. no other group sharing at least one storage method)</p>
</li>
<li><p class="first">determine list of healthy versus unhealthy groups:</p>
<blockquote>
<div><ol class="arabic simple">
<li>a group which contains offline nodes still hosting instances is
definitely not healthy</li>
<li>a group which has nodes failing N+1 is ‘weakly’ unhealthy</li>
</ol>
</div></blockquote>
</li>
<li><p class="first">if either list is empty, exit (no work to do, or no way to fix problems)</p>
</li>
<li><p class="first">for each unhealthy group:</p>
<blockquote>
<div><ol class="arabic simple">
<li>compute the instances that are causing the problems: all
instances living on offline nodes, all instances living as
secondary on N+1 failing nodes, all instances living as primaries
on N+1 failing nodes (in this order)</li>
<li>remove instances, one by one, until the source group is healthy
again</li>
<li>try to run a standard allocation procedure for each instance on
all potential groups in its partition</li>
<li>if all instances were relocated successfully, it means we have a
solution for repairing the original group</li>
</ol>
</div></blockquote>
</li>
</ol>
</div>
<div class="section" id="compression">
<h5>Compression<a class="headerlink" href="#compression" title="Permalink to this headline">¶</a></h5>
<p>In a super cluster which has had many instance reclamations, it is
possible that while none of the groups is empty, overall there is
enough empty capacity that an entire group could be removed.</p>
<p>The algorithm for “compressing” the super cluster is as follows:</p>
<ol class="arabic">
<li><p class="first">read super cluster data</p>
</li>
<li><p class="first">compute total <em>(memory, disk, cpu)</em>, and free <em>(memory, disk, cpu)</em>
for the super-cluster</p>
</li>
<li><p class="first">computer per-group used and free <em>(memory, disk, cpu)</em></p>
</li>
<li><p class="first">select candidate groups for evacuation:</p>
<blockquote>
<div><ol class="arabic simple">
<li>they must be connected to other groups via a common storage type
and pool</li>
<li>they must have fewer used resources than the global free
resources (minus their own free resources)</li>
</ol>
</div></blockquote>
</li>
<li><p class="first">for each of these groups, try to relocate all its instances to
connected peer groups</p>
</li>
<li><p class="first">report the list of groups that could be evacuated, or if instructed
so, perform the evacuation of the group with the largest free
resources (i.e. in order to reclaim the most capacity)</p>
</li>
</ol>
</div>
<div class="section" id="load-balancing">
<h5>Load balancing<a class="headerlink" href="#load-balancing" title="Permalink to this headline">¶</a></h5>
<p>Assuming a super cluster using shared storage, where instance failover
is cheap, it should be possible to do a load-based balancing across
groups.</p>
<p>As opposed to the normal balancing, where we want to balance on all
node attributes, here we should look only at the load attributes; in
other words, compare the available (total) node capacity with the
(total) load generated by instances in a given group, and computing
such scores for all groups, trying to see if we have any outliers.</p>
<p>Once a reliable load-weighting method for groups exists, we can apply
a modified version of the cluster scoring method to score not
imbalances across nodes, but imbalances across groups which result in
a super cluster load-related score.</p>
</div>
</div>
</div>
<div class="section" id="allocation-changes">
<h3>Allocation changes<a class="headerlink" href="#allocation-changes" title="Permalink to this headline">¶</a></h3>
<p>It is important to keep the allocation method across groups internal
(in the Ganeti/Iallocator combination), instead of delegating it to an
external party (e.g. a RAPI client). For this, the IAllocator protocol
should be extended to provide proper group support.</p>
<p>For htools, the new algorithm will work as follows:</p>
<ol class="arabic simple">
<li>read/receive cluster data from Ganeti</li>
<li>filter out any groups that do not supports the requested storage
method</li>
<li>for remaining groups, try allocation and compute scores after
allocation</li>
<li>sort valid allocation solutions accordingly and return the entire
list to Ganeti</li>
</ol>
<p>The rationale for returning the entire group list, and not only the
best choice, is that we anyway have the list, and Ganeti might have
other criteria (e.g. the best group might be busy/locked down, etc.)
so even if from the point of view of resources it is the best choice,
it might not be the overall best one.</p>
</div>
<div class="section" id="node-evacuation-changes">
<h3>Node evacuation changes<a class="headerlink" href="#node-evacuation-changes" title="Permalink to this headline">¶</a></h3>
<p>While the basic concept in the <code class="docutils literal"><span class="pre">multi-evac</span></code> iallocator
mode remains unchanged (it’s a simple local group issue), when failing
to evacuate and running in a super cluster, we could have resources
available elsewhere in the cluster for evacuation.</p>
<p>The algorithm for computing this will be the same as the one for super
cluster compression and redistribution, except that the list of
instances is fixed to the ones living on the nodes to-be-evacuated.</p>
<p>If the inter-group relocation is successful, the result to Ganeti will
not be a local group evacuation target, but instead (for each
instance) a pair <em>(remote group, nodes)</em>. Ganeti itself will have to
decide (based on user input) whether to continue with inter-group
evacuation or not.</p>
<p>In case that Ganeti doesn’t provide complete cluster data, just the
local group, the inter-group relocation won’t be attempted.</p>
</div>
</div>
</div>
</div>
</div>
</div>
<div class="sphinxsidebar" role="navigation" aria-label="main navigation">
<div class="sphinxsidebarwrapper">
<h3><a href="index.html">Table Of Contents</a></h3>
<ul>
<li><a class="reference internal" href="#">Synchronising htools to Ganeti 2.3</a><ul>
<li><a class="reference internal" href="#shared-storage-support">Shared storage support</a><ul>
<li><a class="reference internal" href="#new-instance-parameters">New instance parameters</a></li>
<li><a class="reference internal" href="#new-node-parameters">New node parameters</a></li>
<li><a class="reference internal" href="#external-storage-and-instances">External storage and instances</a></li>
<li><a class="reference internal" href="#external-storage-and-nodes">External storage and nodes</a></li>
<li><a class="reference internal" href="#n-1-status">N+1 status</a></li>
</ul>
</li>
<li><a class="reference internal" href="#node-groups-support">Node groups support</a><ul>
<li><a class="reference internal" href="#balancing-changes">Balancing changes</a><ul>
<li><a class="reference internal" href="#super-cluster-operations">Super cluster operations</a><ul>
<li><a class="reference internal" href="#redistribution">Redistribution</a></li>
<li><a class="reference internal" href="#compression">Compression</a></li>
<li><a class="reference internal" href="#load-balancing">Load balancing</a></li>
</ul>
</li>
</ul>
</li>
<li><a class="reference internal" href="#allocation-changes">Allocation changes</a></li>
<li><a class="reference internal" href="#node-evacuation-changes">Node evacuation changes</a></li>
</ul>
</li>
</ul>
</li>
</ul>
<h4>Previous topic</h4>
<p class="topless"><a href="design-2.3.html"
title="previous chapter">Ganeti 2.3 design</a></p>
<h4>Next topic</h4>
<p class="topless"><a href="design-2.4.html"
title="next chapter">Ganeti 2.4 design</a></p>
<div role="note" aria-label="source link">
<h3>This Page</h3>
<ul class="this-page-menu">
<li><a href="_sources/design-htools-2.3.rst.txt"
rel="nofollow">Show Source</a></li>
</ul>
</div>
<div id="searchbox" style="display: none" role="search">
<h3>Quick search</h3>
<form class="search" action="search.html" method="get">
<div><input type="text" name="q" /></div>
<div><input type="submit" value="Go" /></div>
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
<script type="text/javascript">$('#searchbox').show(0);</script>
</div>
</div>
<div class="clearer"></div>
</div>
<div class="related" role="navigation" aria-label="related navigation">
<h3>Navigation</h3>
<ul>
<li class="right" style="margin-right: 10px">
<a href="design-2.4.html" title="Ganeti 2.4 design"
>next</a></li>
<li class="right" >
<a href="design-2.3.html" title="Ganeti 2.3 design"
>previous</a> |</li>
<li class="nav-item nav-item-0"><a href="index.html">Ganeti 2.16.0~rc2 documentation</a> »</li>
</ul>
</div>
<div class="footer" role="contentinfo">
© Copyright 2018, 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015 Google Inc..
Created using <a href="http://sphinx-doc.org/">Sphinx</a> 1.6.7.
</div>
</body>
</html>
|