/usr/share/doc/ganeti-doc/html/design-impexp2.html is in ganeti-doc 2.9.3-1.
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 | <!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">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
<title>Design for import/export version 2 — Ganeti 2.9.3 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: '2.9.3',
COLLAPSE_INDEX: false,
FILE_SUFFIX: '.html',
HAS_SOURCE: true
};
</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="top" title="Ganeti 2.9.3 documentation" href="index.html" />
<link rel="up" title="Design document drafts" href="design-draft.html" />
<link rel="next" title="Resource model changes" href="design-resource-model.html" />
<link rel="prev" title="Design for replacing Ganeti’s HTTP server" href="design-http-server.html" />
</head>
<body>
<div class="related">
<h3>Navigation</h3>
<ul>
<li class="right" style="margin-right: 10px">
<a href="design-resource-model.html" title="Resource model changes"
accesskey="N">next</a></li>
<li class="right" >
<a href="design-http-server.html" title="Design for replacing Ganeti’s HTTP server"
accesskey="P">previous</a> |</li>
<li><a href="index.html">Ganeti 2.9.3 documentation</a> »</li>
<li><a href="design-draft.html" accesskey="U">Design document drafts</a> »</li>
</ul>
</div>
<div class="document">
<div class="documentwrapper">
<div class="bodywrapper">
<div class="body">
<div class="section" id="design-for-import-export-version-2">
<h1><a class="toc-backref" href="#id2">Design for import/export version 2</a><a class="headerlink" href="#design-for-import-export-version-2" title="Permalink to this headline">¶</a></h1>
<div class="contents topic" id="contents">
<p class="topic-title first">Contents</p>
<ul class="simple">
<li><a class="reference internal" href="#design-for-import-export-version-2" id="id2">Design for import/export version 2</a><ul>
<li><a class="reference internal" href="#current-state-and-shortcomings" id="id3">Current state and shortcomings</a></li>
<li><a class="reference internal" href="#proposed-changes" id="id4">Proposed changes</a><ul>
<li><a class="reference internal" href="#software-requirements" id="id5">Software requirements</a></li>
<li><a class="reference internal" href="#transport" id="id6">Transport</a></li>
<li><a class="reference internal" href="#ssl-certificates" id="id7">SSL certificates</a></li>
<li><a class="reference internal" href="#source" id="id8">Source</a><ul>
<li><a class="reference internal" href="#option-1-cgi" id="id9">Option 1: CGI</a></li>
<li><a class="reference internal" href="#option-2-fastcgi" id="id10">Option 2: FastCGI</a></li>
</ul>
</li>
<li><a class="reference internal" href="#destination" id="id11">Destination</a></li>
</ul>
</li>
<li><a class="reference internal" href="#the-big-picture-for-instance-moves" id="id12">The big picture for instance moves</a></li>
<li><a class="reference internal" href="#http-resources-on-source" id="id13">HTTP resources on source</a></li>
<li><a class="reference internal" href="#code-samples" id="id14">Code samples</a><ul>
<li><a class="reference internal" href="#pycurl-to-file" id="id15">pycURL to file</a></li>
</ul>
</li>
<li><a class="reference internal" href="#backwards-compatibility" id="id16">Backwards compatibility</a><ul>
<li><a class="reference internal" href="#version-1" id="id17">Version 1</a></li>
<li><a class="reference internal" href="#exp-size-fd" id="id18"><tt class="docutils literal"><span class="pre">EXP_SIZE_FD</span></tt></a></li>
<li><a class="reference internal" href="#compression" id="id19">Compression</a></li>
</ul>
</li>
<li><a class="reference internal" href="#notes" id="id20">Notes</a></li>
<li><a class="reference internal" href="#performance-considerations" id="id21">Performance considerations</a></li>
<li><a class="reference internal" href="#other-considered-solutions" id="id22">Other considered solutions</a></li>
</ul>
</li>
</ul>
</div>
<div class="section" id="current-state-and-shortcomings">
<h2><a class="toc-backref" href="#id3">Current state and shortcomings</a><a class="headerlink" href="#current-state-and-shortcomings" title="Permalink to this headline">¶</a></h2>
<p>Ganeti 2.2 introduced <a class="reference internal" href="design-2.2.html"><em>inter-cluster instance moves</em></a>
and replaced the import/export mechanism with the same technology. It’s
since shown that the chosen implementation was too complicated and and
can be difficult to debug.</p>
<p>The old implementation is henceforth called “version 1”. It used
<tt class="docutils literal"><span class="pre">socat</span></tt> in combination with a rather complex tree of <tt class="docutils literal"><span class="pre">bash</span></tt> and
Python utilities to move instances between clusters and import/export
them inside the cluster. Due to protocol limitations, the master daemon
starts a daemon on the involved nodes and then keeps polling a status
file for updates. A non-trivial number of timeouts ensures that jobs
don’t freeze.</p>
<p>In version 1, the destination node would start a daemon listening on a
random TCP port. Upon receiving the destination information, the source
node would temporarily stop the instance, create snapshots, and start
exporting the data by connecting to the destination. The random TCP port
is chosen by the operating system by binding the socket to port 0.
While this is a somewhat elegant solution, it causes problems in setups
with restricted connectivity (e.g. iptables).</p>
<p>Another issue encountered was with dual-stack IPv6 setups. <tt class="docutils literal"><span class="pre">socat</span></tt> can
only listen on one protocol, IPv4 or IPv6, at a time. The connecting
node can not simply resolve the DNS name, but it must be told the exact
IP address.</p>
<p>Instance OS definitions can provide custom import/export scripts. They
were working well in the early days when a filesystem was usually
created directly on the block device. Around Ganeti 2.0 there was a
transition to using partitions on the block devices. Import/export
scripts could no longer use simple <tt class="docutils literal"><span class="pre">dump</span></tt> and <tt class="docutils literal"><span class="pre">restore</span></tt> commands,
but usually ended up doing raw data dumps.</p>
</div>
<div class="section" id="proposed-changes">
<h2><a class="toc-backref" href="#id4">Proposed changes</a><a class="headerlink" href="#proposed-changes" title="Permalink to this headline">¶</a></h2>
<p>Unlike in version 1, in version 2 the destination node will connect to
the source. The active side is swapped. This design assumes the
following design documents have been implemented:</p>
<ul class="simple">
<li><a class="reference internal" href="design-x509-ca.html"><em>Design for a X509 Certificate Authority</em></a></li>
<li><a class="reference internal" href="design-http-server.html"><em>Design for replacing Ganeti’s HTTP server</em></a></li>
</ul>
<p>The following design is mostly targetted at inter-cluster instance
moves. Intra-cluster import and export use the same technology, but do
so in a less complicated way (e.g. reusing the node daemon certificate
in version 1).</p>
<p>Support for instance OS import/export scripts, which have been in Ganeti
since the beginning, will be dropped with this design. Should the need
arise, they can be re-added later.</p>
<div class="section" id="software-requirements">
<h3><a class="toc-backref" href="#id5">Software requirements</a><a class="headerlink" href="#software-requirements" title="Permalink to this headline">¶</a></h3>
<ul class="simple">
<li>HTTP client: cURL/pycURL (already used for inter-node RPC and RAPI
client)</li>
<li>Authentication: X509 certificates (server and client)</li>
</ul>
</div>
<div class="section" id="transport">
<h3><a class="toc-backref" href="#id6">Transport</a><a class="headerlink" href="#transport" title="Permalink to this headline">¶</a></h3>
<p>Instead of a home-grown, mostly raw protocol the widely used HTTP
protocol will be used. Ganeti already uses HTTP for its <a class="reference internal" href="rapi.html"><em>Remote API</em></a> and inter-node communication. Encryption and authentication will
be implemented using SSL and X509 certificates.</p>
</div>
<div class="section" id="ssl-certificates">
<h3><a class="toc-backref" href="#id7">SSL certificates</a><a class="headerlink" href="#ssl-certificates" title="Permalink to this headline">¶</a></h3>
<p>The source machine will identify connecting clients by their SSL
certificate. Unknown certificates will be refused.</p>
<p>Version 1 created a new self-signed certificate per instance
import/export, allowing the certificate to be used as a Certificate
Authority (CA). This worked by means of starting a new <tt class="docutils literal"><span class="pre">socat</span></tt>
instance per instance import/export.</p>
<p>Under the version 2 model, a continously running HTTP server will be
used. This disallows the use of self-signed certificates for
authentication as the CA needs to be the same for all issued
certificates.</p>
<p>See the <a class="reference internal" href="design-x509-ca.html"><em>separate design document for more details on how the
certificate authority will be implemented</em></a>.</p>
<p>Local imports/exports will, like version 1, use the node daemon’s
certificate/key. Doing so allows the verification of local connections.
The client’s certificate can be exported to the CGI/FastCGI handler
using lighttpd’s <tt class="docutils literal"><span class="pre">ssl.verifyclient.exportcert</span></tt> setting. If a
cluster-local import/export is being done, the handler verifies if the
used certificate matches with the local node daemon key.</p>
</div>
<div class="section" id="source">
<h3><a class="toc-backref" href="#id8">Source</a><a class="headerlink" href="#source" title="Permalink to this headline">¶</a></h3>
<p>The source can be the same physical machine as the destination, another
node in the same cluster, or a node in another cluster. A
physical-to-virtual migration mechanism could be implemented as an
alternative source.</p>
<p>In the case of a traditional import, the source is usually a file on the
source machine. For exports and remote imports, the source is an
instance’s raw disk data. In all cases the transported data is opaque to
Ganeti.</p>
<p>All nodes of a cluster will run an instance of Lighttpd. The
configuration is automatically generated when starting Ganeti. The HTTP
server is configured to listen on IPv4 and IPv6 simultaneously.
Imports/exports will use a dedicated TCP port, similar to the Remote
API.</p>
<p>See the separate <a class="reference internal" href="design-http-server.html#http-srv-shortcomings"><em>HTTP server design document</em></a> for why Ganeti’s existing, built-in HTTP server
is not a good choice.</p>
<p>The source cluster is provided with a X509 Certificate Signing Request
(CSR) for a key private to the destination cluster.</p>
<p>After shutting down the instance, creating snapshots and restarting the
instance the master will sign the destination’s X509 certificate using
the <a class="reference internal" href="design-x509-ca.html"><em>X509 CA</em></a> once per instance disk. Instead of
using another identifier, the certificate’s serial number (<a class="reference internal" href="design-x509-ca.html#x509-ca-serial"><em>never
reused</em></a>) and fingerprint are used to identify incoming
requests. Once ready, the master will call an RPC method on the source
node and provide it with the input information (e.g. file paths or block
devices) and the certificate identities.</p>
<p>The RPC method will write the identities to a place accessible by the
HTTP request handler, generate unique transfer IDs and return them to
the master. The transfer ID could be a filename containing the
certificate’s serial number, fingerprint and some disk information. The
file containing the per-transfer information is signed using the node
daemon key and the signature written to a separate file.</p>
<p>Once everything is in place, the master sends the certificates, the data
and notification URLs (which include the transfer IDs) and the public
part of the source’s CA to the job submitter. Like in version 1,
everything will be signed using the cluster domain secret.</p>
<p>Upon receiving a request, the handler verifies the identity and
continues to stream the instance data. The serial number and fingerprint
contained in the transfer ID should be matched with the certificate
used. If a cluster-local import/export was requested, the remote’s
certificate is verified with the local node daemon key. The signature of
the information file from which the handler takes the path of the block
device (and more) is verified using the local node daemon certificate.
There are two options for handling requests, <a class="reference internal" href="#lighttpd-cgi-opt"><em>CGI</em></a> and <a class="reference internal" href="#lighttpd-fastcgi-opt"><em>FastCGI</em></a>.</p>
<p>To wait for all requests to finish, the master calls another RPC method.
The destination should notify the source once it’s done with downloading
the data. Since this notification may never arrive (e.g. network
issues), an additional timeout needs to be used.</p>
<p>There is no good way to avoid polling as the HTTP requests will be
handled asynchronously in another process. Once, and if, implemented
<a class="reference internal" href="design-http-server.html#rpc-feedback"><em>RPC feedback</em></a> could be used to combine the two RPC
methods.</p>
<p>Upon completion of the transfer requests, the instance is removed if
requested.</p>
<div class="section" id="option-1-cgi">
<span id="lighttpd-cgi-opt"></span><h4><a class="toc-backref" href="#id9">Option 1: CGI</a><a class="headerlink" href="#option-1-cgi" title="Permalink to this headline">¶</a></h4>
<p>While easier to implement, this option requires the HTTP server to
either run as “root” or a so-called SUID binary to elevate the started
process to run as “root”.</p>
<p>The export data can be sent directly to the HTTP server without any
further processing.</p>
</div>
<div class="section" id="option-2-fastcgi">
<span id="lighttpd-fastcgi-opt"></span><h4><a class="toc-backref" href="#id10">Option 2: FastCGI</a><a class="headerlink" href="#option-2-fastcgi" title="Permalink to this headline">¶</a></h4>
<p>Unlike plain CGI, FastCGI scripts are run separately from the webserver.
The webserver talks to them via a Unix socket. Webserver and scripts can
run as separate users. Unlike for CGI, there are almost no bootstrap
costs attached to each request.</p>
<p>The FastCGI protocol requires data to be sent in length-prefixed
packets, something which wouldn’t be very efficient to do in Python for
large amounts of data (instance imports/exports can be hundreds of
gigabytes). For this reason the proposal is to use a wrapper program
written in C (e.g. <a class="reference external" href="http://nginx.localdomain.pl/wiki/FcgiWrap">fcgiwrap</a>) and to write the handler
like an old-style CGI program with standard input/output. If data should
be copied from a file, <tt class="docutils literal"><span class="pre">cat</span></tt>, <tt class="docutils literal"><span class="pre">dd</span></tt> or <tt class="docutils literal"><span class="pre">socat</span></tt> can be used (see
note about <a class="reference internal" href="#python-sendfile"><em>sendfile(2)/splice(2) with Python</em></a>).</p>
<p>The bootstrap cost associated with starting a Python interpreter for
a disk export is expected to be negligible.</p>
<p>The <a class="reference external" href="http://cgit.stbuehler.de/gitosis/spawn-fcgi/about/">spawn-fcgi</a>
program will be used to start the CGI wrapper as “root”.</p>
<p>FastCGI is, in the author’s opinion, the better choice as it allows user
separation. As a first implementation step the export handler can be run
as a standard CGI program. User separation can be implemented as a
second step.</p>
</div>
</div>
<div class="section" id="destination">
<h3><a class="toc-backref" href="#id11">Destination</a><a class="headerlink" href="#destination" title="Permalink to this headline">¶</a></h3>
<p>The destination can be the same physical machine as the source, another
node in the same cluster, or a node in another cluster. While not
considered in this design document, instances could be exported from the
cluster by implementing an external client for exports.</p>
<p>For traditional exports the destination is usually a file on the
destination machine. For imports and remote exports, the destination is
an instance’s disks. All transported data is opaque to Ganeti.</p>
<p>Before an import can be started, an RSA key and corresponding
Certificate Signing Request (CSR) must be generated using the new opcode
<tt class="docutils literal"><span class="pre">OpInstanceImportPrepare</span></tt>. The returned information is signed using
the cluster domain secret. The RSA key backing the CSR must not leave
the destination cluster. After being passed through a third party, the
source cluster will generate signed certificates from the CSR.</p>
<p>Once the request for creating the instance arrives at the master daemon,
it’ll create the instance and call an RPC method on the instance’s
primary node to download all data. The RPC method does not return until
the transfer is complete or failed (see <a class="reference internal" href="#exp-size-fd"><em>EXP_SIZE_FD</em></a>
and <a class="reference internal" href="design-http-server.html#rpc-feedback"><em>RPC feedback</em></a>).</p>
<p>The node will use pycURL to connect to the source machine and identify
itself with the signed certificate received. pycURL will be configured
to write directly to a file descriptor pointing to either a regular file
or block device. The file descriptor needs to point to the correct
offset for resuming downloads.</p>
<p>Using cURL’s multi interface, more than one transfer can be made at the
same time. While parallel transfers are used by the version 1
import/export, it can be decided at a later time whether to use them in
version 2 too. More investigation is necessary to determine whether
<tt class="docutils literal"><span class="pre">CURLOPT_MAXCONNECTS</span></tt> is enough to limit the number of connections or
whether more logic is necessary.</p>
<p>If a transfer fails before it’s finished (e.g. timeout or network
issues) it should be retried using an exponential backoff delay. The
opcode submitter can specify for how long the transfer should be
retried.</p>
<p>At the end of a transfer, succssful or not, the source cluster must be
notified. A the same time the RSA key needs to be destroyed.</p>
<p>Support for HTTP proxies can be implemented by setting
<tt class="docutils literal"><span class="pre">CURLOPT_PROXY</span></tt>. Proxies could be used for moving instances in/out of
restricted network environments or across protocol borders (e.g. IPv4
networks unable to talk to IPv6 networks).</p>
</div>
</div>
<div class="section" id="the-big-picture-for-instance-moves">
<h2><a class="toc-backref" href="#id12">The big picture for instance moves</a><a class="headerlink" href="#the-big-picture-for-instance-moves" title="Permalink to this headline">¶</a></h2>
<ol class="arabic simple">
<li><tt class="docutils literal"><span class="pre">OpInstanceImportPrepare</span></tt> (destination cluster)</li>
</ol>
<blockquote>
<div>Create RSA key and CSR (certificate signing request), return signed
with cluster domain secret.</div></blockquote>
<ol class="arabic simple">
<li><tt class="docutils literal"><span class="pre">OpBackupPrepare</span></tt> (source cluster)</li>
</ol>
<blockquote>
<div>Becomes a no-op in version 2, but see <a class="reference internal" href="#backwards-compat"><em>Backwards compatibility</em></a>.</div></blockquote>
<ol class="arabic simple">
<li><tt class="docutils literal"><span class="pre">OpBackupExport</span></tt> (source cluster)</li>
</ol>
<blockquote>
<div><ul class="simple">
<li>Receives destination cluster’s CSR, verifies signature using
cluster domain secret.</li>
<li>Creates certificates using CSR and <a class="reference internal" href="design-x509-ca.html"><em>cluster CA</em></a>, one for each disk</li>
<li>Stop instance, create snapshots, start instance</li>
<li>Prepare HTTP resources on node</li>
<li>Send certificates, URLs and CA certificate to job submitter using
feedback mechanism</li>
<li>Wait for all transfers to finish or fail (with timeout)</li>
<li>Remove snapshots</li>
</ul>
</div></blockquote>
<ol class="arabic simple">
<li><tt class="docutils literal"><span class="pre">OpInstanceCreate</span></tt> (destination cluster)</li>
</ol>
<blockquote>
<div><ul>
<li><p class="first">Receives certificates signed by destination cluster, verifies
certificates and URLs using cluster domain secret</p>
<p>Note that the parameters should be implemented in a generic way
allowing future extensions, e.g. to download disk images from a
public, remote server. The cluster domain secret allows Ganeti to
check data received from a third party, but since this won’t work
with such extensions, other checks will have to be designed.</p>
</li>
<li><p class="first">Create block devices</p>
</li>
<li><p class="first">Download every disk from source, verified using remote’s CA and
authenticated using signed certificates</p>
</li>
<li><p class="first">Destroy RSA key and certificates</p>
</li>
<li><p class="first">Start instance</p>
</li>
</ul>
</div></blockquote>
</div>
<div class="section" id="http-resources-on-source">
<span id="impexp2-http-resources"></span><h2><a class="toc-backref" href="#id13">HTTP resources on source</a><a class="headerlink" href="#http-resources-on-source" title="Permalink to this headline">¶</a></h2>
<p>The HTTP resources listed below will be made available by the source
machine. The transfer ID is generated while preparing the export and is
unique per disk and instance. No caching should be used and the
<tt class="docutils literal"><span class="pre">Pragma</span></tt> (HTTP/1.0) and <tt class="docutils literal"><span class="pre">Cache-Control</span></tt> (HTTP/1.1) headers set
accordingly by the server.</p>
<dl class="docutils">
<dt><tt class="docutils literal"><span class="pre">GET</span> <span class="pre">/transfers/[transfer_id]/contents</span></tt></dt>
<dd><p class="first">Dump disk contents. Important request headers:</p>
<dl class="last docutils">
<dt><tt class="docutils literal"><span class="pre">Accept</span></tt> (<span class="target" id="index-0"></span><a class="rfc reference external" href="http://tools.ietf.org/html/rfc2616.html"><strong>RFC 2616</strong></a>, section 14.1)</dt>
<dd><p class="first">Specify preferred media types. Only one type is supported in the
initial implementation:</p>
<dl class="docutils">
<dt><tt class="docutils literal"><span class="pre">application/octet-stream</span></tt></dt>
<dd>Request raw disk content.</dd>
</dl>
<p>If support for more media types were to be implemented in the
future, the “q” parameter used for “indicating a relative quality
factor” needs to be used. In the meantime parameters need to be
expected, but can be ignored.</p>
<p>If support for OS scripts were to be re-added in the future, the
MIME type <tt class="docutils literal"><span class="pre">application/x-ganeti-instance-export</span></tt> is hereby
reserved for disk dumps using an export script.</p>
<p class="last">If the source can not satisfy the request the response status code
will be 406 (Not Acceptable). Successful requests will specify the
used media type using the <tt class="docutils literal"><span class="pre">Content-Type</span></tt> header. Unless only
exactly one media type is requested, the client must handle the
different response types.</p>
</dd>
<dt><tt class="docutils literal"><span class="pre">Accept-Encoding</span></tt> (<span class="target" id="index-1"></span><a class="rfc reference external" href="http://tools.ietf.org/html/rfc2616.html"><strong>RFC 2616</strong></a>, section 14.3)</dt>
<dd><p class="first">Specify desired content coding. Supported are <tt class="docutils literal"><span class="pre">identity</span></tt> for
uncompressed data, <tt class="docutils literal"><span class="pre">gzip</span></tt> for compressed data and <tt class="docutils literal"><span class="pre">*</span></tt> for any.
The response will include a <tt class="docutils literal"><span class="pre">Content-Encoding</span></tt> header with the
actual coding used. If the client specifies an unknown coding, the
response status code will be 406 (Not Acceptable).</p>
<p class="last">If the client specifically needs compressed data (see
<a class="reference internal" href="#impexp2-compression"><em>Compression</em></a>) but only gets <tt class="docutils literal"><span class="pre">identity</span></tt>, it can
either compress locally or abort the request.</p>
</dd>
<dt><tt class="docutils literal"><span class="pre">Range</span></tt> (<span class="target" id="index-2"></span><a class="rfc reference external" href="http://tools.ietf.org/html/rfc2616.html"><strong>RFC 2616</strong></a>, section 14.35)</dt>
<dd><p class="first">Raw disk dumps can be resumed using this header (e.g. after a
network issue).</p>
<p>If this header was given in the request and the source supports
resuming, the status code of the response will be 206 (Partial
Content) and it’ll include the <tt class="docutils literal"><span class="pre">Content-Range</span></tt> header as per
<span class="target" id="index-3"></span><a class="rfc reference external" href="http://tools.ietf.org/html/rfc2616.html"><strong>RFC 2616</strong></a>. If it does not support resuming or the request was not
specifying a range, the status code will be 200 (OK).</p>
<p>Only a single byte range is supported. cURL does not support
<tt class="docutils literal"><span class="pre">multipart/byteranges</span></tt> responses by itself. Even if they could be
somehow implemented, doing so would be of doubtful benefit for
import/export.</p>
<p>For raw data dumps handling ranges is pretty straightforward by just
dumping the requested range.</p>
<p class="last">cURL will fail with the error code <tt class="docutils literal"><span class="pre">CURLE_RANGE_ERROR</span></tt> if a
request included a range but the server can’t handle it. The request
must be retried without a range.</p>
</dd>
</dl>
</dd>
<dt><tt class="docutils literal"><span class="pre">POST</span> <span class="pre">/transfers/[transfer_id]/done</span></tt></dt>
<dd>Use this resource to notify the source when transfer is finished (even
if not successful). The status code will be 204 (No Content).</dd>
</dl>
</div>
<div class="section" id="code-samples">
<h2><a class="toc-backref" href="#id14">Code samples</a><a class="headerlink" href="#code-samples" title="Permalink to this headline">¶</a></h2>
<div class="section" id="pycurl-to-file">
<h3><a class="toc-backref" href="#id15">pycURL to file</a><a class="headerlink" href="#pycurl-to-file" title="Permalink to this headline">¶</a></h3>
<p>The following code sample shows how to write downloaded data directly to
a file without pumping it through Python:</p>
<div class="highlight-python"><div class="highlight"><pre><span class="n">curl</span> <span class="o">=</span> <span class="n">pycurl</span><span class="o">.</span><span class="n">Curl</span><span class="p">()</span>
<span class="n">curl</span><span class="o">.</span><span class="n">setopt</span><span class="p">(</span><span class="n">pycurl</span><span class="o">.</span><span class="n">URL</span><span class="p">,</span> <span class="s">"http://www.google.com/"</span><span class="p">)</span>
<span class="n">curl</span><span class="o">.</span><span class="n">setopt</span><span class="p">(</span><span class="n">pycurl</span><span class="o">.</span><span class="n">WRITEDATA</span><span class="p">,</span> <span class="nb">open</span><span class="p">(</span><span class="s">"googlecom.html"</span><span class="p">,</span> <span class="s">"w"</span><span class="p">))</span>
<span class="n">curl</span><span class="o">.</span><span class="n">perform</span><span class="p">()</span>
</pre></div>
</div>
<p>This works equally well if the file descriptor is a pipe to another
process.</p>
</div>
</div>
<div class="section" id="backwards-compatibility">
<span id="backwards-compat"></span><h2><a class="toc-backref" href="#id16">Backwards compatibility</a><a class="headerlink" href="#backwards-compatibility" title="Permalink to this headline">¶</a></h2>
<div class="section" id="version-1">
<span id="backwards-compat-v1"></span><h3><a class="toc-backref" href="#id17">Version 1</a><a class="headerlink" href="#version-1" title="Permalink to this headline">¶</a></h3>
<p>The old inter-cluster import/export implementation described in the
<a class="reference internal" href="design-2.2.html"><em>Ganeti 2.2 design document</em></a> will be supported for at
least one minor (2.x) release. Intra-cluster imports/exports will use
the new version right away.</p>
</div>
<div class="section" id="exp-size-fd">
<span id="id1"></span><h3><a class="toc-backref" href="#id18"><tt class="docutils literal"><span class="pre">EXP_SIZE_FD</span></tt></a><a class="headerlink" href="#exp-size-fd" title="Permalink to this headline">¶</a></h3>
<p>Together with the improved import/export infrastructure Ganeti 2.2
allowed instance export scripts to report the expected data size. This
was then used to provide the user with an estimated remaining time.
Version 2 no longer supports OS import/export scripts and therefore
<tt class="docutils literal"><span class="pre">EXP_SIZE_FD</span></tt> is no longer needed.</p>
</div>
<div class="section" id="compression">
<span id="impexp2-compression"></span><h3><a class="toc-backref" href="#id19">Compression</a><a class="headerlink" href="#compression" title="Permalink to this headline">¶</a></h3>
<p>Version 1 used explicit compression using <tt class="docutils literal"><span class="pre">gzip</span></tt> for transporting
data, but the dumped files didn’t use any compression. Version 2 will
allow the destination to specify which encoding should be used. This way
the transported data is already compressed and can be directly used by
the client (see <a class="reference internal" href="#impexp2-http-resources"><em>HTTP resources on source</em></a>). The cURL option
<tt class="docutils literal"><span class="pre">CURLOPT_ENCODING</span></tt> can be used to set the <tt class="docutils literal"><span class="pre">Accept-Encoding</span></tt> header.
cURL will not decompress received data when
<tt class="docutils literal"><span class="pre">CURLOPT_HTTP_CONTENT_DECODING</span></tt> is set to zero (if another HTTP client
library were used which doesn’t support disabling transparent
compression, a custom content-coding type could be defined, e.g.
<tt class="docutils literal"><span class="pre">x-ganeti-gzip</span></tt>).</p>
</div>
</div>
<div class="section" id="notes">
<h2><a class="toc-backref" href="#id20">Notes</a><a class="headerlink" href="#notes" title="Permalink to this headline">¶</a></h2>
<p>The HTTP/1.1 protocol (<span class="target" id="index-4"></span><a class="rfc reference external" href="http://tools.ietf.org/html/rfc2616.html"><strong>RFC 2616</strong></a>) defines trailing headers for chunked
transfers in section 3.6.1. This could be used to transfer a checksum at
the end of an import/export. cURL supports trailing headers since
version 7.14.1. Lighttpd doesn’t seem to support them for FastCGI, but
they appear to be usable in combination with an NPH CGI (No Parsed
Headers).</p>
<p id="lighttp-sendfile">Lighttpd allows FastCGI applications to send the special headers
<tt class="docutils literal"><span class="pre">X-Sendfile</span></tt> and <tt class="docutils literal"><span class="pre">X-Sendfile2</span></tt> (the latter with a range). Using
these headers applications can send response headers and tell the
webserver to serve regular file stored on the file system as a response
body. The webserver will then take care of sending that file.
Unfortunately this mechanism is restricted to regular files and can not
be used for data from programs, neither direct nor via named pipes,
without writing to a file first. The latter is not an option as instance
data can be very large. Theoretically <tt class="docutils literal"><span class="pre">X-Sendfile</span></tt> could be used for
sending the input for a file-based instance import, but that’d require
the webserver to run as “root”.</p>
<p id="python-sendfile">Python does not include interfaces for the <tt class="docutils literal"><span class="pre">sendfile(2)</span></tt> or
<tt class="docutils literal"><span class="pre">splice(2)</span></tt> system calls. The latter can be useful for faster copying
of data between file descriptors. There are some 3rd-party modules (e.g.
<a class="reference external" href="http://pypi.python.org/pypi/py-sendfile/">http://pypi.python.org/pypi/py-sendfile/</a>) and discussions
(<a class="reference external" href="http://bugs.python.org/issue10882">http://bugs.python.org/issue10882</a>) for including support for
<tt class="docutils literal"><span class="pre">sendfile(2)</span></tt>, but the later is certainly not going to happen for the
Python versions supported by Ganeti. Calling the function using the
<tt class="docutils literal"><span class="pre">ctypes</span></tt> module might be possible.</p>
</div>
<div class="section" id="performance-considerations">
<h2><a class="toc-backref" href="#id21">Performance considerations</a><a class="headerlink" href="#performance-considerations" title="Permalink to this headline">¶</a></h2>
<p>The design described above was confirmed to be one of the better choices
in terms of download performance with bigger block sizes. All numbers
were gathered on the same physical machine with a single CPU and 1 GB of
RAM while downloading 2 GB of zeros read from <tt class="docutils literal"><span class="pre">/dev/zero</span></tt>. <tt class="docutils literal"><span class="pre">wget</span></tt>
(version 1.10.2) was used as the client, <tt class="docutils literal"><span class="pre">lighttpd</span></tt> (version 1.4.28)
as the server. The numbers in the first line are in megabytes per
second. The second line in each row is the CPU time spent in userland
respective system (measured for the CGI/FastCGI program using <tt class="docutils literal"><span class="pre">time</span>
<span class="pre">-v</span></tt>).</p>
<div class="highlight-python"><div class="highlight"><pre>----------------------------------------------------------------------
Block size 4 KB 64 KB 128 KB 1 MB 4 MB
======================================================================
Plain CGI script reading 83 174 180 122 120
from ``/dev/zero``
0.6/3.9 0.1/2.4 0.1/2.2 0.0/1.9 0.0/2.1
----------------------------------------------------------------------
FastCGI with ``fcgiwrap``, 86 167 170 177 174
``dd`` reading from
``/dev/zero`` 1.1/5 0.5/2.9 0.5/2.7 0.7/3.1 0.7/2.8
----------------------------------------------------------------------
FastCGI with ``fcgiwrap``, 68 146 150 170 170
Python script copying from
``/dev/zero`` to stdout
1.3/5.1 0.8/3.7 0.7/3.3 0.9/2.9 0.8/3
----------------------------------------------------------------------
FastCGI, Python script using 31 48 47 5 1
``flup`` library (version
1.0.2) reading from
``/dev/zero``
23.5/9.8 14.3/8.5 16.1/8 - -
----------------------------------------------------------------------
</pre></div>
</div>
<p>It should be mentioned that the <tt class="docutils literal"><span class="pre">flup</span></tt> library is not implemented in
the most efficient way, but even with some changes it doesn’t get much
faster. It is fine for small amounts of data, but not for huge
transfers.</p>
</div>
<div class="section" id="other-considered-solutions">
<h2><a class="toc-backref" href="#id22">Other considered solutions</a><a class="headerlink" href="#other-considered-solutions" title="Permalink to this headline">¶</a></h2>
<p>Another possible solution considered was to use <tt class="docutils literal"><span class="pre">socat</span></tt> like version 1
did. Due to the changing model, a large part of the code would’ve
required a rewrite anyway, while still not fixing all shortcomings. For
example, <tt class="docutils literal"><span class="pre">socat</span></tt> could still listen on only one protocol, IPv4 or
IPv6. Running two separate instances might have fixed that, but it’d get
more complicated. Using an existing HTTP server will provide us with a
number of other benefits as well, such as easier user separation between
server and backend.</p>
</div>
</div>
</div>
</div>
</div>
<div class="sphinxsidebar">
<div class="sphinxsidebarwrapper">
<h3><a href="index.html">Table Of Contents</a></h3>
<ul>
<li><a class="reference internal" href="#">Design for import/export version 2</a><ul>
<li><a class="reference internal" href="#current-state-and-shortcomings">Current state and shortcomings</a></li>
<li><a class="reference internal" href="#proposed-changes">Proposed changes</a><ul>
<li><a class="reference internal" href="#software-requirements">Software requirements</a></li>
<li><a class="reference internal" href="#transport">Transport</a></li>
<li><a class="reference internal" href="#ssl-certificates">SSL certificates</a></li>
<li><a class="reference internal" href="#source">Source</a><ul>
<li><a class="reference internal" href="#option-1-cgi">Option 1: CGI</a></li>
<li><a class="reference internal" href="#option-2-fastcgi">Option 2: FastCGI</a></li>
</ul>
</li>
<li><a class="reference internal" href="#destination">Destination</a></li>
</ul>
</li>
<li><a class="reference internal" href="#the-big-picture-for-instance-moves">The big picture for instance moves</a></li>
<li><a class="reference internal" href="#http-resources-on-source">HTTP resources on source</a></li>
<li><a class="reference internal" href="#code-samples">Code samples</a><ul>
<li><a class="reference internal" href="#pycurl-to-file">pycURL to file</a></li>
</ul>
</li>
<li><a class="reference internal" href="#backwards-compatibility">Backwards compatibility</a><ul>
<li><a class="reference internal" href="#version-1">Version 1</a></li>
<li><a class="reference internal" href="#exp-size-fd"><tt class="docutils literal"><span class="pre">EXP_SIZE_FD</span></tt></a></li>
<li><a class="reference internal" href="#compression">Compression</a></li>
</ul>
</li>
<li><a class="reference internal" href="#notes">Notes</a></li>
<li><a class="reference internal" href="#performance-considerations">Performance considerations</a></li>
<li><a class="reference internal" href="#other-considered-solutions">Other considered solutions</a></li>
</ul>
</li>
</ul>
<h4>Previous topic</h4>
<p class="topless"><a href="design-http-server.html"
title="previous chapter">Design for replacing Ganeti’s HTTP server</a></p>
<h4>Next topic</h4>
<p class="topless"><a href="design-resource-model.html"
title="next chapter">Resource model changes</a></p>
<h3>This Page</h3>
<ul class="this-page-menu">
<li><a href="_sources/design-impexp2.txt"
rel="nofollow">Show Source</a></li>
</ul>
<div id="searchbox" style="display: none">
<h3>Quick search</h3>
<form class="search" action="search.html" method="get">
<input type="text" name="q" />
<input type="submit" value="Go" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
<p class="searchtip" style="font-size: 90%">
Enter search terms or a module, class or function name.
</p>
</div>
<script type="text/javascript">$('#searchbox').show(0);</script>
</div>
</div>
<div class="clearer"></div>
</div>
<div class="related">
<h3>Navigation</h3>
<ul>
<li class="right" style="margin-right: 10px">
<a href="design-resource-model.html" title="Resource model changes"
>next</a></li>
<li class="right" >
<a href="design-http-server.html" title="Design for replacing Ganeti’s HTTP server"
>previous</a> |</li>
<li><a href="index.html">Ganeti 2.9.3 documentation</a> »</li>
<li><a href="design-draft.html" >Design document drafts</a> »</li>
</ul>
</div>
<div class="footer">
© Copyright 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013 Google Inc..
Created using <a href="http://sphinx-doc.org/">Sphinx</a> 1.2.1.
</div>
</body>
</html>
|