Statistics
| Branch: | Tag: | Revision:

root / doc / design-2.0.rst @ f154a7a3

History | View | Annotate | Download (75.4 kB)

1 5c0c1eeb Iustin Pop
=================
2 5c0c1eeb Iustin Pop
Ganeti 2.0 design
3 5c0c1eeb Iustin Pop
=================
4 5c0c1eeb Iustin Pop
5 5c0c1eeb Iustin Pop
This document describes the major changes in Ganeti 2.0 compared to
6 5c0c1eeb Iustin Pop
the 1.2 version.
7 5c0c1eeb Iustin Pop
8 5c0c1eeb Iustin Pop
The 2.0 version will constitute a rewrite of the 'core' architecture,
9 5c0c1eeb Iustin Pop
paving the way for additional features in future 2.x versions.
10 5c0c1eeb Iustin Pop
11 e0eb13de Iustin Pop
.. contents:: :depth: 3
12 5c0c1eeb Iustin Pop
13 5c0c1eeb Iustin Pop
Objective
14 5c0c1eeb Iustin Pop
=========
15 5c0c1eeb Iustin Pop
16 5c0c1eeb Iustin Pop
Ganeti 1.2 has many scalability issues and restrictions due to its
17 5c0c1eeb Iustin Pop
roots as software for managing small and 'static' clusters.
18 5c0c1eeb Iustin Pop
19 5c0c1eeb Iustin Pop
Version 2.0 will attempt to remedy first the scalability issues and
20 5c0c1eeb Iustin Pop
then the restrictions.
21 5c0c1eeb Iustin Pop
22 5c0c1eeb Iustin Pop
Background
23 5c0c1eeb Iustin Pop
==========
24 5c0c1eeb Iustin Pop
25 6c2d0b44 Iustin Pop
While Ganeti 1.2 is usable, it severely limits the flexibility of the
26 5c0c1eeb Iustin Pop
cluster administration and imposes a very rigid model. It has the
27 5c0c1eeb Iustin Pop
following main scalability issues:
28 5c0c1eeb Iustin Pop
29 5c0c1eeb Iustin Pop
- only one operation at a time on the cluster [#]_
30 5c0c1eeb Iustin Pop
- poor handling of node failures in the cluster
31 5c0c1eeb Iustin Pop
- mixing hypervisors in a cluster not allowed
32 5c0c1eeb Iustin Pop
33 7faf5110 Michael Hanselmann
It also has a number of artificial restrictions, due to historical
34 7faf5110 Michael Hanselmann
design:
35 5c0c1eeb Iustin Pop
36 5c0c1eeb Iustin Pop
- fixed number of disks (two) per instance
37 6c2d0b44 Iustin Pop
- fixed number of NICs
38 5c0c1eeb Iustin Pop
39 5c0c1eeb Iustin Pop
.. [#] Replace disks will release the lock, but this is an exception
40 5c0c1eeb Iustin Pop
       and not a recommended way to operate
41 5c0c1eeb Iustin Pop
42 5c0c1eeb Iustin Pop
The 2.0 version is intended to address some of these problems, and
43 6c2d0b44 Iustin Pop
create a more flexible code base for future developments.
44 6c2d0b44 Iustin Pop
45 6c2d0b44 Iustin Pop
Among these problems, the single-operation at a time restriction is
46 6c2d0b44 Iustin Pop
biggest issue with the current version of Ganeti. It is such a big
47 6c2d0b44 Iustin Pop
impediment in operating bigger clusters that many times one is tempted
48 6c2d0b44 Iustin Pop
to remove the lock just to do a simple operation like start instance
49 6c2d0b44 Iustin Pop
while an OS installation is running.
50 5c0c1eeb Iustin Pop
51 5c0c1eeb Iustin Pop
Scalability problems
52 5c0c1eeb Iustin Pop
--------------------
53 5c0c1eeb Iustin Pop
54 5c0c1eeb Iustin Pop
Ganeti 1.2 has a single global lock, which is used for all cluster
55 5c0c1eeb Iustin Pop
operations.  This has been painful at various times, for example:
56 5c0c1eeb Iustin Pop
57 5c0c1eeb Iustin Pop
- It is impossible for two people to efficiently interact with a cluster
58 5c0c1eeb Iustin Pop
  (for example for debugging) at the same time.
59 7faf5110 Michael Hanselmann
- When batch jobs are running it's impossible to do other work (for
60 7faf5110 Michael Hanselmann
  example failovers/fixes) on a cluster.
61 5c0c1eeb Iustin Pop
62 5c0c1eeb Iustin Pop
This poses scalability problems: as clusters grow in node and instance
63 5c0c1eeb Iustin Pop
size it's a lot more likely that operations which one could conceive
64 5c0c1eeb Iustin Pop
should run in parallel (for example because they happen on different
65 5c0c1eeb Iustin Pop
nodes) are actually stalling each other while waiting for the global
66 5c0c1eeb Iustin Pop
lock, without a real reason for that to happen.
67 5c0c1eeb Iustin Pop
68 5c0c1eeb Iustin Pop
One of the main causes of this global lock (beside the higher
69 5c0c1eeb Iustin Pop
difficulty of ensuring data consistency in a more granular lock model)
70 6c2d0b44 Iustin Pop
is the fact that currently there is no long-lived process in Ganeti
71 6c2d0b44 Iustin Pop
that can coordinate multiple operations. Each command tries to acquire
72 6c2d0b44 Iustin Pop
the so called *cmd* lock and when it succeeds, it takes complete
73 6c2d0b44 Iustin Pop
ownership of the cluster configuration and state.
74 5c0c1eeb Iustin Pop
75 5c0c1eeb Iustin Pop
Other scalability problems are due the design of the DRBD device
76 5c0c1eeb Iustin Pop
model, which assumed at its creation a low (one to four) number of
77 5c0c1eeb Iustin Pop
instances per node, which is no longer true with today's hardware.
78 5c0c1eeb Iustin Pop
79 5c0c1eeb Iustin Pop
Artificial restrictions
80 5c0c1eeb Iustin Pop
-----------------------
81 5c0c1eeb Iustin Pop
82 5c0c1eeb Iustin Pop
Ganeti 1.2 (and previous versions) have a fixed two-disks, one-NIC per
83 5c0c1eeb Iustin Pop
instance model. This is a purely artificial restrictions, but it
84 5c0c1eeb Iustin Pop
touches multiple areas (configuration, import/export, command line)
85 5c0c1eeb Iustin Pop
that it's more fitted to a major release than a minor one.
86 5c0c1eeb Iustin Pop
87 6c2d0b44 Iustin Pop
Architecture issues
88 6c2d0b44 Iustin Pop
-------------------
89 6c2d0b44 Iustin Pop
90 6c2d0b44 Iustin Pop
The fact that each command is a separate process that reads the
91 6c2d0b44 Iustin Pop
cluster state, executes the command, and saves the new state is also
92 6c2d0b44 Iustin Pop
an issue on big clusters where the configuration data for the cluster
93 6c2d0b44 Iustin Pop
begins to be non-trivial in size.
94 6c2d0b44 Iustin Pop
95 5c0c1eeb Iustin Pop
Overview
96 5c0c1eeb Iustin Pop
========
97 5c0c1eeb Iustin Pop
98 5c0c1eeb Iustin Pop
In order to solve the scalability problems, a rewrite of the core
99 5c0c1eeb Iustin Pop
design of Ganeti is required. While the cluster operations themselves
100 5c0c1eeb Iustin Pop
won't change (e.g. start instance will do the same things, the way
101 5c0c1eeb Iustin Pop
these operations are scheduled internally will change radically.
102 5c0c1eeb Iustin Pop
103 f86e82ef Iustin Pop
The new design will change the cluster architecture to:
104 f86e82ef Iustin Pop
105 f86e82ef Iustin Pop
.. image:: arch-2.0.png
106 f86e82ef Iustin Pop
107 f86e82ef Iustin Pop
This differs from the 1.2 architecture by the addition of the master
108 f86e82ef Iustin Pop
daemon, which will be the only entity to talk to the node daemons.
109 f86e82ef Iustin Pop
110 f86e82ef Iustin Pop
111 5c0c1eeb Iustin Pop
Detailed design
112 5c0c1eeb Iustin Pop
===============
113 5c0c1eeb Iustin Pop
114 5c0c1eeb Iustin Pop
The changes for 2.0 can be split into roughly three areas:
115 5c0c1eeb Iustin Pop
116 5c0c1eeb Iustin Pop
- core changes that affect the design of the software
117 5c0c1eeb Iustin Pop
- features (or restriction removals) but which do not have a wide
118 5c0c1eeb Iustin Pop
  impact on the design
119 5c0c1eeb Iustin Pop
- user-level and API-level changes which translate into differences for
120 5c0c1eeb Iustin Pop
  the operation of the cluster
121 5c0c1eeb Iustin Pop
122 5c0c1eeb Iustin Pop
Core changes
123 5c0c1eeb Iustin Pop
------------
124 5c0c1eeb Iustin Pop
125 5c0c1eeb Iustin Pop
The main changes will be switching from a per-process model to a
126 5c0c1eeb Iustin Pop
daemon based model, where the individual gnt-* commands will be
127 6c2d0b44 Iustin Pop
clients that talk to this daemon (see `Master daemon`_). This will
128 6c2d0b44 Iustin Pop
allow us to get rid of the global cluster lock for most operations,
129 6c2d0b44 Iustin Pop
having instead a per-object lock (see `Granular locking`_). Also, the
130 6c2d0b44 Iustin Pop
daemon will be able to queue jobs, and this will allow the individual
131 6c2d0b44 Iustin Pop
clients to submit jobs without waiting for them to finish, and also
132 6c2d0b44 Iustin Pop
see the result of old requests (see `Job Queue`_).
133 5c0c1eeb Iustin Pop
134 5c0c1eeb Iustin Pop
Beside these major changes, another 'core' change but that will not be
135 5c0c1eeb Iustin Pop
as visible to the users will be changing the model of object attribute
136 6c2d0b44 Iustin Pop
storage, and separate that into name spaces (such that an Xen PVM
137 5c0c1eeb Iustin Pop
instance will not have the Xen HVM parameters). This will allow future
138 6c2d0b44 Iustin Pop
flexibility in defining additional parameters. For more details see
139 6c2d0b44 Iustin Pop
`Object parameters`_.
140 5c0c1eeb Iustin Pop
141 5c0c1eeb Iustin Pop
The various changes brought in by the master daemon model and the
142 5c0c1eeb Iustin Pop
read-write RAPI will require changes to the cluster security; we move
143 6c2d0b44 Iustin Pop
away from Twisted and use HTTP(s) for intra- and extra-cluster
144 5c0c1eeb Iustin Pop
communications. For more details, see the security document in the
145 5c0c1eeb Iustin Pop
doc/ directory.
146 5c0c1eeb Iustin Pop
147 5c0c1eeb Iustin Pop
Master daemon
148 5c0c1eeb Iustin Pop
~~~~~~~~~~~~~
149 5c0c1eeb Iustin Pop
150 5c0c1eeb Iustin Pop
In Ganeti 2.0, we will have the following *entities*:
151 5c0c1eeb Iustin Pop
152 5c0c1eeb Iustin Pop
- the master daemon (on the master node)
153 5c0c1eeb Iustin Pop
- the node daemon (on all nodes)
154 5c0c1eeb Iustin Pop
- the command line tools (on the master node)
155 5c0c1eeb Iustin Pop
- the RAPI daemon (on the master node)
156 5c0c1eeb Iustin Pop
157 6c2d0b44 Iustin Pop
The master-daemon related interaction paths are:
158 5c0c1eeb Iustin Pop
159 7faf5110 Michael Hanselmann
- (CLI tools/RAPI daemon) and the master daemon, via the so called
160 7faf5110 Michael Hanselmann
  *LUXI* API
161 5c0c1eeb Iustin Pop
- the master daemon and the node daemons, via the node RPC
162 5c0c1eeb Iustin Pop
163 6c2d0b44 Iustin Pop
There are also some additional interaction paths for exceptional cases:
164 6c2d0b44 Iustin Pop
165 6c2d0b44 Iustin Pop
- CLI tools might access via SSH the nodes (for ``gnt-cluster copyfile``
166 6c2d0b44 Iustin Pop
  and ``gnt-cluster command``)
167 6c2d0b44 Iustin Pop
- master failover is a special case when a non-master node will SSH
168 6c2d0b44 Iustin Pop
  and do node-RPC calls to the current master
169 6c2d0b44 Iustin Pop
170 5c0c1eeb Iustin Pop
The protocol between the master daemon and the node daemons will be
171 6c2d0b44 Iustin Pop
changed from (Ganeti 1.2) Twisted PB (perspective broker) to HTTP(S),
172 6c2d0b44 Iustin Pop
using a simple PUT/GET of JSON-encoded messages. This is done due to
173 6c2d0b44 Iustin Pop
difficulties in working with the Twisted framework and its protocols
174 6c2d0b44 Iustin Pop
in a multithreaded environment, which we can overcome by using a
175 6c2d0b44 Iustin Pop
simpler stack (see the caveats section).
176 6c2d0b44 Iustin Pop
177 6c2d0b44 Iustin Pop
The protocol between the CLI/RAPI and the master daemon will be a
178 6c2d0b44 Iustin Pop
custom one (called *LUXI*): on a UNIX socket on the master node, with
179 6c2d0b44 Iustin Pop
rights restricted by filesystem permissions, the CLI/RAPI will talk to
180 6c2d0b44 Iustin Pop
the master daemon using JSON-encoded messages.
181 5c0c1eeb Iustin Pop
182 5c0c1eeb Iustin Pop
The operations supported over this internal protocol will be encoded
183 5c0c1eeb Iustin Pop
via a python library that will expose a simple API for its
184 5c0c1eeb Iustin Pop
users. Internally, the protocol will simply encode all objects in JSON
185 5c0c1eeb Iustin Pop
format and decode them on the receiver side.
186 5c0c1eeb Iustin Pop
187 6c2d0b44 Iustin Pop
For more details about the RAPI daemon see `Remote API changes`_, and
188 6c2d0b44 Iustin Pop
for the node daemon see `Node daemon changes`_.
189 6c2d0b44 Iustin Pop
190 5c0c1eeb Iustin Pop
The LUXI protocol
191 5c0c1eeb Iustin Pop
+++++++++++++++++
192 5c0c1eeb Iustin Pop
193 6c2d0b44 Iustin Pop
As described above, the protocol for making requests or queries to the
194 6c2d0b44 Iustin Pop
master daemon will be a UNIX-socket based simple RPC of JSON-encoded
195 6c2d0b44 Iustin Pop
messages.
196 6c2d0b44 Iustin Pop
197 6c2d0b44 Iustin Pop
The choice of UNIX was in order to get rid of the need of
198 6c2d0b44 Iustin Pop
authentication and authorisation inside Ganeti; for 2.0, the
199 6c2d0b44 Iustin Pop
permissions on the Unix socket itself will determine the access
200 6c2d0b44 Iustin Pop
rights.
201 6c2d0b44 Iustin Pop
202 6c2d0b44 Iustin Pop
We will have two main classes of operations over this API:
203 5c0c1eeb Iustin Pop
204 5c0c1eeb Iustin Pop
- cluster query functions
205 5c0c1eeb Iustin Pop
- job related functions
206 5c0c1eeb Iustin Pop
207 5c0c1eeb Iustin Pop
The cluster query functions are usually short-duration, and are the
208 6c2d0b44 Iustin Pop
equivalent of the ``OP_QUERY_*`` opcodes in Ganeti 1.2 (and they are
209 5c0c1eeb Iustin Pop
internally implemented still with these opcodes). The clients are
210 5c0c1eeb Iustin Pop
guaranteed to receive the response in a reasonable time via a timeout.
211 5c0c1eeb Iustin Pop
212 5c0c1eeb Iustin Pop
The job-related functions will be:
213 5c0c1eeb Iustin Pop
214 5c0c1eeb Iustin Pop
- submit job
215 5c0c1eeb Iustin Pop
- query job (which could also be categorized in the query-functions)
216 5c0c1eeb Iustin Pop
- archive job (see the job queue design doc)
217 5c0c1eeb Iustin Pop
- wait for job change, which allows a client to wait without polling
218 5c0c1eeb Iustin Pop
219 6c2d0b44 Iustin Pop
For more details of the actual operation list, see the `Job Queue`_.
220 5c0c1eeb Iustin Pop
221 6c2d0b44 Iustin Pop
Both requests and responses will consist of a JSON-encoded message
222 6c2d0b44 Iustin Pop
followed by the ``ETX`` character (ASCII decimal 3), which is not a
223 6c2d0b44 Iustin Pop
valid character in JSON messages and thus can serve as a message
224 6c2d0b44 Iustin Pop
delimiter. The contents of the messages will be a dictionary with two
225 6c2d0b44 Iustin Pop
fields:
226 6c2d0b44 Iustin Pop
227 6c2d0b44 Iustin Pop
:method:
228 6c2d0b44 Iustin Pop
  the name of the method called
229 6c2d0b44 Iustin Pop
:args:
230 6c2d0b44 Iustin Pop
  the arguments to the method, as a list (no keyword arguments allowed)
231 6c2d0b44 Iustin Pop
232 6c2d0b44 Iustin Pop
Responses will follow the same format, with the two fields being:
233 6c2d0b44 Iustin Pop
234 6c2d0b44 Iustin Pop
:success:
235 6c2d0b44 Iustin Pop
  a boolean denoting the success of the operation
236 6c2d0b44 Iustin Pop
:result:
237 6c2d0b44 Iustin Pop
  the actual result, or error message in case of failure
238 6c2d0b44 Iustin Pop
239 6c2d0b44 Iustin Pop
There are two special value for the result field:
240 6c2d0b44 Iustin Pop
241 6c2d0b44 Iustin Pop
- in the case that the operation failed, and this field is a list of
242 7faf5110 Michael Hanselmann
  length two, the client library will try to interpret is as an
243 7faf5110 Michael Hanselmann
  exception, the first element being the exception type and the second
244 7faf5110 Michael Hanselmann
  one the actual exception arguments; this will allow a simple method of
245 7faf5110 Michael Hanselmann
  passing Ganeti-related exception across the interface
246 6c2d0b44 Iustin Pop
- for the *WaitForChange* call (that waits on the server for a job to
247 6c2d0b44 Iustin Pop
  change status), if the result is equal to ``nochange`` instead of the
248 6c2d0b44 Iustin Pop
  usual result for this call (a list of changes), then the library will
249 6c2d0b44 Iustin Pop
  internally retry the call; this is done in order to differentiate
250 6c2d0b44 Iustin Pop
  internally between master daemon hung and job simply not changed
251 6c2d0b44 Iustin Pop
252 6c2d0b44 Iustin Pop
Users of the API that don't use the provided python library should
253 6c2d0b44 Iustin Pop
take care of the above two cases.
254 6c2d0b44 Iustin Pop
255 6c2d0b44 Iustin Pop
256 6c2d0b44 Iustin Pop
Master daemon implementation
257 6c2d0b44 Iustin Pop
++++++++++++++++++++++++++++
258 5c0c1eeb Iustin Pop
259 5c0c1eeb Iustin Pop
The daemon will be based around a main I/O thread that will wait for
260 5c0c1eeb Iustin Pop
new requests from the clients, and that does the setup/shutdown of the
261 5c0c1eeb Iustin Pop
other thread (pools).
262 5c0c1eeb Iustin Pop
263 5c0c1eeb Iustin Pop
There will two other classes of threads in the daemon:
264 5c0c1eeb Iustin Pop
265 5c0c1eeb Iustin Pop
- job processing threads, part of a thread pool, and which are
266 5c0c1eeb Iustin Pop
  long-lived, started at daemon startup and terminated only at shutdown
267 5c0c1eeb Iustin Pop
  time
268 5c0c1eeb Iustin Pop
- client I/O threads, which are the ones that talk the local protocol
269 6c2d0b44 Iustin Pop
  (LUXI) to the clients, and are short-lived
270 5c0c1eeb Iustin Pop
271 5c0c1eeb Iustin Pop
Master startup/failover
272 5c0c1eeb Iustin Pop
+++++++++++++++++++++++
273 5c0c1eeb Iustin Pop
274 5c0c1eeb Iustin Pop
In Ganeti 1.x there is no protection against failing over the master
275 5c0c1eeb Iustin Pop
to a node with stale configuration. In effect, the responsibility of
276 5c0c1eeb Iustin Pop
correct failovers falls on the admin. This is true both for the new
277 5c0c1eeb Iustin Pop
master and for when an old, offline master startup.
278 5c0c1eeb Iustin Pop
279 5c0c1eeb Iustin Pop
Since in 2.x we are extending the cluster state to cover the job queue
280 5c0c1eeb Iustin Pop
and have a daemon that will execute by itself the job queue, we want
281 5c0c1eeb Iustin Pop
to have more resilience for the master role.
282 5c0c1eeb Iustin Pop
283 5c0c1eeb Iustin Pop
The following algorithm will happen whenever a node is ready to
284 5c0c1eeb Iustin Pop
transition to the master role, either at startup time or at node
285 5c0c1eeb Iustin Pop
failover:
286 5c0c1eeb Iustin Pop
287 5c0c1eeb Iustin Pop
#. read the configuration file and parse the node list
288 5c0c1eeb Iustin Pop
   contained within
289 5c0c1eeb Iustin Pop
290 5c0c1eeb Iustin Pop
#. query all the nodes and make sure we obtain an agreement via
291 5c0c1eeb Iustin Pop
   a quorum of at least half plus one nodes for the following:
292 5c0c1eeb Iustin Pop
293 5c0c1eeb Iustin Pop
    - we have the latest configuration and job list (as
294 5c0c1eeb Iustin Pop
      determined by the serial number on the configuration and
295 5c0c1eeb Iustin Pop
      highest job ID on the job queue)
296 5c0c1eeb Iustin Pop
297 5c0c1eeb Iustin Pop
    - there is not even a single node having a newer
298 5c0c1eeb Iustin Pop
      configuration file
299 5c0c1eeb Iustin Pop
300 5c0c1eeb Iustin Pop
    - if we are not failing over (but just starting), the
301 5c0c1eeb Iustin Pop
      quorum agrees that we are the designated master
302 5c0c1eeb Iustin Pop
303 6c2d0b44 Iustin Pop
    - if any of the above is false, we prevent the current operation
304 6c2d0b44 Iustin Pop
      (i.e. we don't become the master)
305 6c2d0b44 Iustin Pop
306 5c0c1eeb Iustin Pop
#. at this point, the node transitions to the master role
307 5c0c1eeb Iustin Pop
308 5c0c1eeb Iustin Pop
#. for all the in-progress jobs, mark them as failed, with
309 5c0c1eeb Iustin Pop
   reason unknown or something similar (master failed, etc.)
310 5c0c1eeb Iustin Pop
311 6c2d0b44 Iustin Pop
Since due to exceptional conditions we could have a situation in which
312 6c2d0b44 Iustin Pop
no node can become the master due to inconsistent data, we will have
313 6c2d0b44 Iustin Pop
an override switch for the master daemon startup that will assume the
314 6c2d0b44 Iustin Pop
current node has the right data and will replicate all the
315 6c2d0b44 Iustin Pop
configuration files to the other nodes.
316 6c2d0b44 Iustin Pop
317 6c2d0b44 Iustin Pop
**Note**: the above algorithm is by no means an election algorithm; it
318 6c2d0b44 Iustin Pop
is a *confirmation* of the master role currently held by a node.
319 5c0c1eeb Iustin Pop
320 5c0c1eeb Iustin Pop
Logging
321 5c0c1eeb Iustin Pop
+++++++
322 5c0c1eeb Iustin Pop
323 6c2d0b44 Iustin Pop
The logging system will be switched completely to the standard python
324 6c2d0b44 Iustin Pop
logging module; currently it's logging-based, but exposes a different
325 6c2d0b44 Iustin Pop
API, which is just overhead. As such, the code will be switched over
326 6c2d0b44 Iustin Pop
to standard logging calls, and only the setup will be custom.
327 5c0c1eeb Iustin Pop
328 5c0c1eeb Iustin Pop
With this change, we will remove the separate debug/info/error logs,
329 5c0c1eeb Iustin Pop
and instead have always one logfile per daemon model:
330 5c0c1eeb Iustin Pop
331 5c0c1eeb Iustin Pop
- master-daemon.log for the master daemon
332 5c0c1eeb Iustin Pop
- node-daemon.log for the node daemon (this is the same as in 1.2)
333 5c0c1eeb Iustin Pop
- rapi-daemon.log for the RAPI daemon logs
334 5c0c1eeb Iustin Pop
- rapi-access.log, an additional log file for the RAPI that will be
335 6c2d0b44 Iustin Pop
  in the standard HTTP log format for possible parsing by other tools
336 6c2d0b44 Iustin Pop
337 e2078d28 Iustin Pop
Since the :term:`watcher` will only submit jobs to the master for
338 e2078d28 Iustin Pop
startup of the instances, its log file will contain less information
339 e2078d28 Iustin Pop
than before, mainly that it will start the instance, but not the
340 e2078d28 Iustin Pop
results.
341 6c2d0b44 Iustin Pop
342 6c2d0b44 Iustin Pop
Node daemon changes
343 6c2d0b44 Iustin Pop
+++++++++++++++++++
344 6c2d0b44 Iustin Pop
345 6c2d0b44 Iustin Pop
The only change to the node daemon is that, since we need better
346 6c2d0b44 Iustin Pop
concurrency, we don't process the inter-node RPC calls in the node
347 6c2d0b44 Iustin Pop
daemon itself, but we fork and process each request in a separate
348 6c2d0b44 Iustin Pop
child.
349 5c0c1eeb Iustin Pop
350 6c2d0b44 Iustin Pop
Since we don't have many calls, and we only fork (not exec), the
351 6c2d0b44 Iustin Pop
overhead should be minimal.
352 5c0c1eeb Iustin Pop
353 5c0c1eeb Iustin Pop
Caveats
354 5c0c1eeb Iustin Pop
+++++++
355 5c0c1eeb Iustin Pop
356 5c0c1eeb Iustin Pop
A discussed alternative is to keep the current individual processes
357 5c0c1eeb Iustin Pop
touching the cluster configuration model. The reasons we have not
358 5c0c1eeb Iustin Pop
chosen this approach is:
359 5c0c1eeb Iustin Pop
360 5c0c1eeb Iustin Pop
- the speed of reading and unserializing the cluster state
361 5c0c1eeb Iustin Pop
  today is not small enough that we can ignore it; the addition of
362 5c0c1eeb Iustin Pop
  the job queue will make the startup cost even higher. While this
363 5c0c1eeb Iustin Pop
  runtime cost is low, it can be on the order of a few seconds on
364 5c0c1eeb Iustin Pop
  bigger clusters, which for very quick commands is comparable to
365 5c0c1eeb Iustin Pop
  the actual duration of the computation itself
366 5c0c1eeb Iustin Pop
367 5c0c1eeb Iustin Pop
- individual commands would make it harder to implement a
368 5c0c1eeb Iustin Pop
  fire-and-forget job request, along the lines "start this
369 5c0c1eeb Iustin Pop
  instance but do not wait for it to finish"; it would require a
370 5c0c1eeb Iustin Pop
  model of backgrounding the operation and other things that are
371 5c0c1eeb Iustin Pop
  much better served by a daemon-based model
372 5c0c1eeb Iustin Pop
373 5c0c1eeb Iustin Pop
Another area of discussion is moving away from Twisted in this new
374 6c2d0b44 Iustin Pop
implementation. While Twisted has its advantages, there are also many
375 6c2d0b44 Iustin Pop
disadvantages to using it:
376 5c0c1eeb Iustin Pop
377 5c0c1eeb Iustin Pop
- first and foremost, it's not a library, but a framework; thus, if
378 6c2d0b44 Iustin Pop
  you use twisted, all the code needs to be 'twiste-ized' and written
379 6c2d0b44 Iustin Pop
  in an asynchronous manner, using deferreds; while this method works,
380 6c2d0b44 Iustin Pop
  it's not a common way to code and it requires that the entire process
381 6c2d0b44 Iustin Pop
  workflow is based around a single *reactor* (Twisted name for a main
382 6c2d0b44 Iustin Pop
  loop)
383 6c2d0b44 Iustin Pop
- the more advanced granular locking that we want to implement would
384 6c2d0b44 Iustin Pop
  require, if written in the async-manner, deep integration with the
385 6c2d0b44 Iustin Pop
  Twisted stack, to such an extend that business-logic is inseparable
386 7faf5110 Michael Hanselmann
  from the protocol coding; we felt that this is an unreasonable
387 7faf5110 Michael Hanselmann
  request, and that a good protocol library should allow complete
388 7faf5110 Michael Hanselmann
  separation of low-level protocol calls and business logic; by
389 7faf5110 Michael Hanselmann
  comparison, the threaded approach combined with HTTPs protocol
390 7faf5110 Michael Hanselmann
  required (for the first iteration) absolutely no changes from the 1.2
391 7faf5110 Michael Hanselmann
  code, and later changes for optimizing the inter-node RPC calls
392 7faf5110 Michael Hanselmann
  required just syntactic changes (e.g.  ``rpc.call_...`` to
393 7faf5110 Michael Hanselmann
  ``self.rpc.call_...``)
394 6c2d0b44 Iustin Pop
395 6c2d0b44 Iustin Pop
Another issue is with the Twisted API stability - during the Ganeti
396 6c2d0b44 Iustin Pop
1.x lifetime, we had to to implement many times workarounds to changes
397 6c2d0b44 Iustin Pop
in the Twisted version, so that for example 1.2 is able to use both
398 6c2d0b44 Iustin Pop
Twisted 2.x and 8.x.
399 6c2d0b44 Iustin Pop
400 6c2d0b44 Iustin Pop
In the end, since we already had an HTTP server library for the RAPI,
401 6c2d0b44 Iustin Pop
we just reused that for inter-node communication.
402 5c0c1eeb Iustin Pop
403 5c0c1eeb Iustin Pop
404 5c0c1eeb Iustin Pop
Granular locking
405 5c0c1eeb Iustin Pop
~~~~~~~~~~~~~~~~
406 5c0c1eeb Iustin Pop
407 7faf5110 Michael Hanselmann
We want to make sure that multiple operations can run in parallel on a
408 7faf5110 Michael Hanselmann
Ganeti Cluster. In order for this to happen we need to make sure
409 7faf5110 Michael Hanselmann
concurrently run operations don't step on each other toes and break the
410 7faf5110 Michael Hanselmann
cluster.
411 5c0c1eeb Iustin Pop
412 5c0c1eeb Iustin Pop
This design addresses how we are going to deal with locking so that:
413 5c0c1eeb Iustin Pop
414 6c2d0b44 Iustin Pop
- we preserve data coherency
415 6c2d0b44 Iustin Pop
- we prevent deadlocks
416 6c2d0b44 Iustin Pop
- we prevent job starvation
417 5c0c1eeb Iustin Pop
418 7faf5110 Michael Hanselmann
Reaching the maximum possible parallelism is a Non-Goal. We have
419 7faf5110 Michael Hanselmann
identified a set of operations that are currently bottlenecks and need
420 7faf5110 Michael Hanselmann
to be parallelised and have worked on those. In the future it will be
421 7faf5110 Michael Hanselmann
possible to address other needs, thus making the cluster more and more
422 7faf5110 Michael Hanselmann
parallel one step at a time.
423 5c0c1eeb Iustin Pop
424 6c2d0b44 Iustin Pop
This section only talks about parallelising Ganeti level operations, aka
425 7faf5110 Michael Hanselmann
Logical Units, and the locking needed for that. Any other
426 7faf5110 Michael Hanselmann
synchronization lock needed internally by the code is outside its scope.
427 5c0c1eeb Iustin Pop
428 6c2d0b44 Iustin Pop
Library details
429 6c2d0b44 Iustin Pop
+++++++++++++++
430 5c0c1eeb Iustin Pop
431 5c0c1eeb Iustin Pop
The proposed library has these features:
432 5c0c1eeb Iustin Pop
433 7faf5110 Michael Hanselmann
- internally managing all the locks, making the implementation
434 7faf5110 Michael Hanselmann
  transparent from their usage
435 7faf5110 Michael Hanselmann
- automatically grabbing multiple locks in the right order (avoid
436 7faf5110 Michael Hanselmann
  deadlock)
437 6c2d0b44 Iustin Pop
- ability to transparently handle conversion to more granularity
438 6c2d0b44 Iustin Pop
- support asynchronous operation (future goal)
439 6c2d0b44 Iustin Pop
440 6c2d0b44 Iustin Pop
Locking will be valid only on the master node and will not be a
441 6c2d0b44 Iustin Pop
distributed operation. Therefore, in case of master failure, the
442 6c2d0b44 Iustin Pop
operations currently running will be aborted and the locks will be
443 6c2d0b44 Iustin Pop
lost; it remains to the administrator to cleanup (if needed) the
444 6c2d0b44 Iustin Pop
operation result (e.g. make sure an instance is either installed
445 6c2d0b44 Iustin Pop
correctly or removed).
446 6c2d0b44 Iustin Pop
447 6c2d0b44 Iustin Pop
A corollary of this is that a master-failover operation with both
448 6c2d0b44 Iustin Pop
masters alive needs to happen while no operations are running, and
449 6c2d0b44 Iustin Pop
therefore no locks are held.
450 6c2d0b44 Iustin Pop
451 6c2d0b44 Iustin Pop
All the locks will be represented by objects (like
452 6c2d0b44 Iustin Pop
``lockings.SharedLock``), and the individual locks for each object
453 6c2d0b44 Iustin Pop
will be created at initialisation time, from the config file.
454 6c2d0b44 Iustin Pop
455 7faf5110 Michael Hanselmann
The API will have a way to grab one or more than one locks at the same
456 7faf5110 Michael Hanselmann
time.  Any attempt to grab a lock while already holding one in the wrong
457 7faf5110 Michael Hanselmann
order will be checked for, and fail.
458 6c2d0b44 Iustin Pop
459 5c0c1eeb Iustin Pop
460 5c0c1eeb Iustin Pop
The Locks
461 5c0c1eeb Iustin Pop
+++++++++
462 5c0c1eeb Iustin Pop
463 5c0c1eeb Iustin Pop
At the first stage we have decided to provide the following locks:
464 5c0c1eeb Iustin Pop
465 5c0c1eeb Iustin Pop
- One "config file" lock
466 5c0c1eeb Iustin Pop
- One lock per node in the cluster
467 5c0c1eeb Iustin Pop
- One lock per instance in the cluster
468 5c0c1eeb Iustin Pop
469 7faf5110 Michael Hanselmann
All the instance locks will need to be taken before the node locks, and
470 7faf5110 Michael Hanselmann
the node locks before the config lock. Locks will need to be acquired at
471 7faf5110 Michael Hanselmann
the same time for multiple instances and nodes, and internal ordering
472 7faf5110 Michael Hanselmann
will be dealt within the locking library, which, for simplicity, will
473 7faf5110 Michael Hanselmann
just use alphabetical order.
474 5c0c1eeb Iustin Pop
475 6c2d0b44 Iustin Pop
Each lock has the following three possible statuses:
476 6c2d0b44 Iustin Pop
477 6c2d0b44 Iustin Pop
- unlocked (anyone can grab the lock)
478 6c2d0b44 Iustin Pop
- shared (anyone can grab/have the lock but only in shared mode)
479 6c2d0b44 Iustin Pop
- exclusive (no one else can grab/have the lock)
480 6c2d0b44 Iustin Pop
481 5c0c1eeb Iustin Pop
Handling conversion to more granularity
482 5c0c1eeb Iustin Pop
+++++++++++++++++++++++++++++++++++++++
483 5c0c1eeb Iustin Pop
484 7faf5110 Michael Hanselmann
In order to convert to a more granular approach transparently each time
485 7faf5110 Michael Hanselmann
we split a lock into more we'll create a "metalock", which will depend
486 7faf5110 Michael Hanselmann
on those sub-locks and live for the time necessary for all the code to
487 7faf5110 Michael Hanselmann
convert (or forever, in some conditions). When a metalock exists all
488 7faf5110 Michael Hanselmann
converted code must acquire it in shared mode, so it can run
489 7faf5110 Michael Hanselmann
concurrently, but still be exclusive with old code, which acquires it
490 7faf5110 Michael Hanselmann
exclusively.
491 5c0c1eeb Iustin Pop
492 7faf5110 Michael Hanselmann
In the beginning the only such lock will be what replaces the current
493 7faf5110 Michael Hanselmann
"command" lock, and will acquire all the locks in the system, before
494 7faf5110 Michael Hanselmann
proceeding. This lock will be called the "Big Ganeti Lock" because
495 7faf5110 Michael Hanselmann
holding that one will avoid any other concurrent Ganeti operations.
496 5c0c1eeb Iustin Pop
497 7faf5110 Michael Hanselmann
We might also want to devise more metalocks (eg. all nodes, all
498 7faf5110 Michael Hanselmann
nodes+config) in order to make it easier for some parts of the code to
499 7faf5110 Michael Hanselmann
acquire what it needs without specifying it explicitly.
500 5c0c1eeb Iustin Pop
501 7faf5110 Michael Hanselmann
In the future things like the node locks could become metalocks, should
502 7faf5110 Michael Hanselmann
we decide to split them into an even more fine grained approach, but
503 7faf5110 Michael Hanselmann
this will probably be only after the first 2.0 version has been
504 7faf5110 Michael Hanselmann
released.
505 5c0c1eeb Iustin Pop
506 5c0c1eeb Iustin Pop
Adding/Removing locks
507 5c0c1eeb Iustin Pop
+++++++++++++++++++++
508 5c0c1eeb Iustin Pop
509 7faf5110 Michael Hanselmann
When a new instance or a new node is created an associated lock must be
510 7faf5110 Michael Hanselmann
added to the list. The relevant code will need to inform the locking
511 7faf5110 Michael Hanselmann
library of such a change.
512 5c0c1eeb Iustin Pop
513 7faf5110 Michael Hanselmann
This needs to be compatible with every other lock in the system,
514 7faf5110 Michael Hanselmann
especially metalocks that guarantee to grab sets of resources without
515 7faf5110 Michael Hanselmann
specifying them explicitly. The implementation of this will be handled
516 7faf5110 Michael Hanselmann
in the locking library itself.
517 5c0c1eeb Iustin Pop
518 6c2d0b44 Iustin Pop
When instances or nodes disappear from the cluster the relevant locks
519 6c2d0b44 Iustin Pop
must be removed. This is easier than adding new elements, as the code
520 6c2d0b44 Iustin Pop
which removes them must own them exclusively already, and thus deals
521 6c2d0b44 Iustin Pop
with metalocks exactly as normal code acquiring those locks. Any
522 6c2d0b44 Iustin Pop
operation queuing on a removed lock will fail after its removal.
523 5c0c1eeb Iustin Pop
524 5c0c1eeb Iustin Pop
Asynchronous operations
525 5c0c1eeb Iustin Pop
+++++++++++++++++++++++
526 5c0c1eeb Iustin Pop
527 5c0c1eeb Iustin Pop
For the first version the locking library will only export synchronous
528 7faf5110 Michael Hanselmann
operations, which will block till the needed lock are held, and only
529 7faf5110 Michael Hanselmann
fail if the request is impossible or somehow erroneous.
530 5c0c1eeb Iustin Pop
531 5c0c1eeb Iustin Pop
In the future we may want to implement different types of asynchronous
532 5c0c1eeb Iustin Pop
operations such as:
533 5c0c1eeb Iustin Pop
534 6c2d0b44 Iustin Pop
- try to acquire this lock set and fail if not possible
535 7faf5110 Michael Hanselmann
- try to acquire one of these lock sets and return the first one you
536 7faf5110 Michael Hanselmann
  were able to get (or after a timeout) (select/poll like)
537 5c0c1eeb Iustin Pop
538 7faf5110 Michael Hanselmann
These operations can be used to prioritize operations based on available
539 7faf5110 Michael Hanselmann
locks, rather than making them just blindly queue for acquiring them.
540 7faf5110 Michael Hanselmann
The inherent risk, though, is that any code using the first operation,
541 7faf5110 Michael Hanselmann
or setting a timeout for the second one, is susceptible to starvation
542 7faf5110 Michael Hanselmann
and thus may never be able to get the required locks and complete
543 7faf5110 Michael Hanselmann
certain tasks. Considering this providing/using these operations should
544 7faf5110 Michael Hanselmann
not be among our first priorities.
545 5c0c1eeb Iustin Pop
546 5c0c1eeb Iustin Pop
Locking granularity
547 5c0c1eeb Iustin Pop
+++++++++++++++++++
548 5c0c1eeb Iustin Pop
549 5c0c1eeb Iustin Pop
For the first version of this code we'll convert each Logical Unit to
550 7faf5110 Michael Hanselmann
acquire/release the locks it needs, so locking will be at the Logical
551 7faf5110 Michael Hanselmann
Unit level.  In the future we may want to split logical units in
552 7faf5110 Michael Hanselmann
independent "tasklets" with their own locking requirements. A different
553 7faf5110 Michael Hanselmann
design doc (or mini design doc) will cover the move from Logical Units
554 7faf5110 Michael Hanselmann
to tasklets.
555 5c0c1eeb Iustin Pop
556 6c2d0b44 Iustin Pop
Code examples
557 6c2d0b44 Iustin Pop
+++++++++++++
558 5c0c1eeb Iustin Pop
559 7faf5110 Michael Hanselmann
In general when acquiring locks we should use a code path equivalent
560 7faf5110 Michael Hanselmann
to::
561 5c0c1eeb Iustin Pop
562 5c0c1eeb Iustin Pop
  lock.acquire()
563 5c0c1eeb Iustin Pop
  try:
564 5c0c1eeb Iustin Pop
    ...
565 5c0c1eeb Iustin Pop
    # other code
566 5c0c1eeb Iustin Pop
  finally:
567 5c0c1eeb Iustin Pop
    lock.release()
568 5c0c1eeb Iustin Pop
569 6c2d0b44 Iustin Pop
This makes sure we release all locks, and avoid possible deadlocks. Of
570 6c2d0b44 Iustin Pop
course extra care must be used not to leave, if possible locked
571 6c2d0b44 Iustin Pop
structures in an unusable state. Note that with Python 2.5 a simpler
572 6c2d0b44 Iustin Pop
syntax will be possible, but we want to keep compatibility with Python
573 6c2d0b44 Iustin Pop
2.4 so the new constructs should not be used.
574 5c0c1eeb Iustin Pop
575 7faf5110 Michael Hanselmann
In order to avoid this extra indentation and code changes everywhere in
576 7faf5110 Michael Hanselmann
the Logical Units code, we decided to allow LUs to declare locks, and
577 7faf5110 Michael Hanselmann
then execute their code with their locks acquired. In the new world LUs
578 7faf5110 Michael Hanselmann
are called like this::
579 5c0c1eeb Iustin Pop
580 5c0c1eeb Iustin Pop
  # user passed names are expanded to the internal lock/resource name,
581 5c0c1eeb Iustin Pop
  # then known needed locks are declared
582 5c0c1eeb Iustin Pop
  lu.ExpandNames()
583 5c0c1eeb Iustin Pop
  ... some locking/adding of locks may happen ...
584 5c0c1eeb Iustin Pop
  # late declaration of locks for one level: this is useful because sometimes
585 5c0c1eeb Iustin Pop
  # we can't know which resource we need before locking the previous level
586 5c0c1eeb Iustin Pop
  lu.DeclareLocks() # for each level (cluster, instance, node)
587 5c0c1eeb Iustin Pop
  ... more locking/adding of locks can happen ...
588 5c0c1eeb Iustin Pop
  # these functions are called with the proper locks held
589 5c0c1eeb Iustin Pop
  lu.CheckPrereq()
590 5c0c1eeb Iustin Pop
  lu.Exec()
591 5c0c1eeb Iustin Pop
  ... locks declared for removal are removed, all acquired locks released ...
592 5c0c1eeb Iustin Pop
593 7faf5110 Michael Hanselmann
The Processor and the LogicalUnit class will contain exact documentation
594 7faf5110 Michael Hanselmann
on how locks are supposed to be declared.
595 5c0c1eeb Iustin Pop
596 5c0c1eeb Iustin Pop
Caveats
597 5c0c1eeb Iustin Pop
+++++++
598 5c0c1eeb Iustin Pop
599 5c0c1eeb Iustin Pop
This library will provide an easy upgrade path to bring all the code to
600 5c0c1eeb Iustin Pop
granular locking without breaking everything, and it will also guarantee
601 7faf5110 Michael Hanselmann
against a lot of common errors. Code switching from the old "lock
602 7faf5110 Michael Hanselmann
everything" lock to the new system, though, needs to be carefully
603 7faf5110 Michael Hanselmann
scrutinised to be sure it is really acquiring all the necessary locks,
604 7faf5110 Michael Hanselmann
and none has been overlooked or forgotten.
605 5c0c1eeb Iustin Pop
606 7faf5110 Michael Hanselmann
The code can contain other locks outside of this library, to synchronise
607 7faf5110 Michael Hanselmann
other threaded code (eg for the job queue) but in general these should
608 7faf5110 Michael Hanselmann
be leaf locks or carefully structured non-leaf ones, to avoid deadlock
609 7faf5110 Michael Hanselmann
race conditions.
610 5c0c1eeb Iustin Pop
611 5c0c1eeb Iustin Pop
612 5c0c1eeb Iustin Pop
Job Queue
613 5c0c1eeb Iustin Pop
~~~~~~~~~
614 5c0c1eeb Iustin Pop
615 5c0c1eeb Iustin Pop
Granular locking is not enough to speed up operations, we also need a
616 5c0c1eeb Iustin Pop
queue to store these and to be able to process as many as possible in
617 5c0c1eeb Iustin Pop
parallel.
618 5c0c1eeb Iustin Pop
619 6c2d0b44 Iustin Pop
A Ganeti job will consist of multiple ``OpCodes`` which are the basic
620 5c0c1eeb Iustin Pop
element of operation in Ganeti 1.2 (and will remain as such). Most
621 5c0c1eeb Iustin Pop
command-level commands are equivalent to one OpCode, or in some cases
622 5c0c1eeb Iustin Pop
to a sequence of opcodes, all of the same type (e.g. evacuating a node
623 5c0c1eeb Iustin Pop
will generate N opcodes of type replace disks).
624 5c0c1eeb Iustin Pop
625 5c0c1eeb Iustin Pop
626 5c0c1eeb Iustin Pop
Job execution—“Life of a Ganeti job”
627 5c0c1eeb Iustin Pop
++++++++++++++++++++++++++++++++++++
628 5c0c1eeb Iustin Pop
629 7faf5110 Michael Hanselmann
#. Job gets submitted by the client. A new job identifier is generated
630 7faf5110 Michael Hanselmann
   and assigned to the job. The job is then automatically replicated
631 7faf5110 Michael Hanselmann
   [#replic]_ to all nodes in the cluster. The identifier is returned to
632 7faf5110 Michael Hanselmann
   the client.
633 7faf5110 Michael Hanselmann
#. A pool of worker threads waits for new jobs. If all are busy, the job
634 7faf5110 Michael Hanselmann
   has to wait and the first worker finishing its work will grab it.
635 7faf5110 Michael Hanselmann
   Otherwise any of the waiting threads will pick up the new job.
636 7faf5110 Michael Hanselmann
#. Client waits for job status updates by calling a waiting RPC
637 7faf5110 Michael Hanselmann
   function. Log message may be shown to the user. Until the job is
638 7faf5110 Michael Hanselmann
   started, it can also be canceled.
639 7faf5110 Michael Hanselmann
#. As soon as the job is finished, its final result and status can be
640 7faf5110 Michael Hanselmann
   retrieved from the server.
641 5c0c1eeb Iustin Pop
#. If the client archives the job, it gets moved to a history directory.
642 5c0c1eeb Iustin Pop
   There will be a method to archive all jobs older than a a given age.
643 5c0c1eeb Iustin Pop
644 7faf5110 Michael Hanselmann
.. [#replic] We need replication in order to maintain the consistency
645 7faf5110 Michael Hanselmann
   across all nodes in the system; the master node only differs in the
646 7faf5110 Michael Hanselmann
   fact that now it is running the master daemon, but it if fails and we
647 7faf5110 Michael Hanselmann
   do a master failover, the jobs are still visible on the new master
648 7faf5110 Michael Hanselmann
   (though marked as failed).
649 5c0c1eeb Iustin Pop
650 5c0c1eeb Iustin Pop
Failures to replicate a job to other nodes will be only flagged as
651 5c0c1eeb Iustin Pop
errors in the master daemon log if more than half of the nodes failed,
652 5c0c1eeb Iustin Pop
otherwise we ignore the failure, and rely on the fact that the next
653 5c0c1eeb Iustin Pop
update (for still running jobs) will retry the update. For finished
654 5c0c1eeb Iustin Pop
jobs, it is less of a problem.
655 5c0c1eeb Iustin Pop
656 5c0c1eeb Iustin Pop
Future improvements will look into checking the consistency of the job
657 5c0c1eeb Iustin Pop
list and jobs themselves at master daemon startup.
658 5c0c1eeb Iustin Pop
659 5c0c1eeb Iustin Pop
660 5c0c1eeb Iustin Pop
Job storage
661 5c0c1eeb Iustin Pop
+++++++++++
662 5c0c1eeb Iustin Pop
663 5c0c1eeb Iustin Pop
Jobs are stored in the filesystem as individual files, serialized
664 5c0c1eeb Iustin Pop
using JSON (standard serialization mechanism in Ganeti).
665 5c0c1eeb Iustin Pop
666 5c0c1eeb Iustin Pop
The choice of storing each job in its own file was made because:
667 5c0c1eeb Iustin Pop
668 5c0c1eeb Iustin Pop
- a file can be atomically replaced
669 5c0c1eeb Iustin Pop
- a file can easily be replicated to other nodes
670 7faf5110 Michael Hanselmann
- checking consistency across nodes can be implemented very easily,
671 7faf5110 Michael Hanselmann
  since all job files should be (at a given moment in time) identical
672 5c0c1eeb Iustin Pop
673 5c0c1eeb Iustin Pop
The other possible choices that were discussed and discounted were:
674 5c0c1eeb Iustin Pop
675 7faf5110 Michael Hanselmann
- single big file with all job data: not feasible due to difficult
676 7faf5110 Michael Hanselmann
  updates
677 5c0c1eeb Iustin Pop
- in-process databases: hard to replicate the entire database to the
678 7faf5110 Michael Hanselmann
  other nodes, and replicating individual operations does not mean wee
679 7faf5110 Michael Hanselmann
  keep consistency
680 5c0c1eeb Iustin Pop
681 5c0c1eeb Iustin Pop
682 5c0c1eeb Iustin Pop
Queue structure
683 5c0c1eeb Iustin Pop
+++++++++++++++
684 5c0c1eeb Iustin Pop
685 7faf5110 Michael Hanselmann
All file operations have to be done atomically by writing to a temporary
686 7faf5110 Michael Hanselmann
file and subsequent renaming. Except for log messages, every change in a
687 7faf5110 Michael Hanselmann
job is stored and replicated to other nodes.
688 5c0c1eeb Iustin Pop
689 5c0c1eeb Iustin Pop
::
690 5c0c1eeb Iustin Pop
691 5c0c1eeb Iustin Pop
  /var/lib/ganeti/queue/
692 5c0c1eeb Iustin Pop
    job-1 (JSON encoded job description and status)
693 5c0c1eeb Iustin Pop
    […]
694 5c0c1eeb Iustin Pop
    job-37
695 5c0c1eeb Iustin Pop
    job-38
696 5c0c1eeb Iustin Pop
    job-39
697 5c0c1eeb Iustin Pop
    lock (Queue managing process opens this file in exclusive mode)
698 5c0c1eeb Iustin Pop
    serial (Last job ID used)
699 5c0c1eeb Iustin Pop
    version (Queue format version)
700 5c0c1eeb Iustin Pop
701 5c0c1eeb Iustin Pop
702 5c0c1eeb Iustin Pop
Locking
703 5c0c1eeb Iustin Pop
+++++++
704 5c0c1eeb Iustin Pop
705 7faf5110 Michael Hanselmann
Locking in the job queue is a complicated topic. It is called from more
706 7faf5110 Michael Hanselmann
than one thread and must be thread-safe. For simplicity, a single lock
707 7faf5110 Michael Hanselmann
is used for the whole job queue.
708 5c0c1eeb Iustin Pop
709 9725b53d Michael Hanselmann
A more detailed description can be found in doc/locking.rst.
710 5c0c1eeb Iustin Pop
711 5c0c1eeb Iustin Pop
712 5c0c1eeb Iustin Pop
Internal RPC
713 5c0c1eeb Iustin Pop
++++++++++++
714 5c0c1eeb Iustin Pop
715 5c0c1eeb Iustin Pop
RPC calls available between Ganeti master and node daemons:
716 5c0c1eeb Iustin Pop
717 5c0c1eeb Iustin Pop
jobqueue_update(file_name, content)
718 5c0c1eeb Iustin Pop
  Writes a file in the job queue directory.
719 5c0c1eeb Iustin Pop
jobqueue_purge()
720 5c0c1eeb Iustin Pop
  Cleans the job queue directory completely, including archived job.
721 5c0c1eeb Iustin Pop
jobqueue_rename(old, new)
722 5c0c1eeb Iustin Pop
  Renames a file in the job queue directory.
723 5c0c1eeb Iustin Pop
724 5c0c1eeb Iustin Pop
725 5c0c1eeb Iustin Pop
Client RPC
726 5c0c1eeb Iustin Pop
++++++++++
727 5c0c1eeb Iustin Pop
728 7faf5110 Michael Hanselmann
RPC between Ganeti clients and the Ganeti master daemon supports the
729 7faf5110 Michael Hanselmann
following operations:
730 5c0c1eeb Iustin Pop
731 5c0c1eeb Iustin Pop
SubmitJob(ops)
732 7faf5110 Michael Hanselmann
  Submits a list of opcodes and returns the job identifier. The
733 7faf5110 Michael Hanselmann
  identifier is guaranteed to be unique during the lifetime of a
734 7faf5110 Michael Hanselmann
  cluster.
735 5c0c1eeb Iustin Pop
WaitForJobChange(job_id, fields, […], timeout)
736 7faf5110 Michael Hanselmann
  This function waits until a job changes or a timeout expires. The
737 7faf5110 Michael Hanselmann
  condition for when a job changed is defined by the fields passed and
738 7faf5110 Michael Hanselmann
  the last log message received.
739 5c0c1eeb Iustin Pop
QueryJobs(job_ids, fields)
740 5c0c1eeb Iustin Pop
  Returns field values for the job identifiers passed.
741 5c0c1eeb Iustin Pop
CancelJob(job_id)
742 7faf5110 Michael Hanselmann
  Cancels the job specified by identifier. This operation may fail if
743 7faf5110 Michael Hanselmann
  the job is already running, canceled or finished.
744 5c0c1eeb Iustin Pop
ArchiveJob(job_id)
745 7faf5110 Michael Hanselmann
  Moves a job into the …/archive/ directory. This operation will fail if
746 7faf5110 Michael Hanselmann
  the job has not been canceled or finished.
747 5c0c1eeb Iustin Pop
748 5c0c1eeb Iustin Pop
749 5c0c1eeb Iustin Pop
Job and opcode status
750 5c0c1eeb Iustin Pop
+++++++++++++++++++++
751 5c0c1eeb Iustin Pop
752 5c0c1eeb Iustin Pop
Each job and each opcode has, at any time, one of the following states:
753 5c0c1eeb Iustin Pop
754 5c0c1eeb Iustin Pop
Queued
755 5c0c1eeb Iustin Pop
  The job/opcode was submitted, but did not yet start.
756 5c0c1eeb Iustin Pop
Waiting
757 5c0c1eeb Iustin Pop
  The job/opcode is waiting for a lock to proceed.
758 5c0c1eeb Iustin Pop
Running
759 5c0c1eeb Iustin Pop
  The job/opcode is running.
760 5c0c1eeb Iustin Pop
Canceled
761 5c0c1eeb Iustin Pop
  The job/opcode was canceled before it started.
762 5c0c1eeb Iustin Pop
Success
763 5c0c1eeb Iustin Pop
  The job/opcode ran and finished successfully.
764 5c0c1eeb Iustin Pop
Error
765 5c0c1eeb Iustin Pop
  The job/opcode was aborted with an error.
766 5c0c1eeb Iustin Pop
767 7faf5110 Michael Hanselmann
If the master is aborted while a job is running, the job will be set to
768 7faf5110 Michael Hanselmann
the Error status once the master started again.
769 5c0c1eeb Iustin Pop
770 5c0c1eeb Iustin Pop
771 5c0c1eeb Iustin Pop
History
772 5c0c1eeb Iustin Pop
+++++++
773 5c0c1eeb Iustin Pop
774 5c0c1eeb Iustin Pop
Archived jobs are kept in a separate directory,
775 6c2d0b44 Iustin Pop
``/var/lib/ganeti/queue/archive/``.  This is done in order to speed up
776 6c2d0b44 Iustin Pop
the queue handling: by default, the jobs in the archive are not
777 6c2d0b44 Iustin Pop
touched by any functions. Only the current (unarchived) jobs are
778 6c2d0b44 Iustin Pop
parsed, loaded, and verified (if implemented) by the master daemon.
779 5c0c1eeb Iustin Pop
780 5c0c1eeb Iustin Pop
781 5c0c1eeb Iustin Pop
Ganeti updates
782 5c0c1eeb Iustin Pop
++++++++++++++
783 5c0c1eeb Iustin Pop
784 5c0c1eeb Iustin Pop
The queue has to be completely empty for Ganeti updates with changes
785 5c0c1eeb Iustin Pop
in the job queue structure. In order to allow this, there will be a
786 5c0c1eeb Iustin Pop
way to prevent new jobs entering the queue.
787 5c0c1eeb Iustin Pop
788 5c0c1eeb Iustin Pop
789 5c0c1eeb Iustin Pop
Object parameters
790 5c0c1eeb Iustin Pop
~~~~~~~~~~~~~~~~~
791 5c0c1eeb Iustin Pop
792 5c0c1eeb Iustin Pop
Across all cluster configuration data, we have multiple classes of
793 5c0c1eeb Iustin Pop
parameters:
794 5c0c1eeb Iustin Pop
795 5c0c1eeb Iustin Pop
A. cluster-wide parameters (e.g. name of the cluster, the master);
796 5c0c1eeb Iustin Pop
   these are the ones that we have today, and are unchanged from the
797 5c0c1eeb Iustin Pop
   current model
798 5c0c1eeb Iustin Pop
799 5c0c1eeb Iustin Pop
#. node parameters
800 5c0c1eeb Iustin Pop
801 5c0c1eeb Iustin Pop
#. instance specific parameters, e.g. the name of disks (LV), that
802 5c0c1eeb Iustin Pop
   cannot be shared with other instances
803 5c0c1eeb Iustin Pop
804 5c0c1eeb Iustin Pop
#. instance parameters, that are or can be the same for many
805 5c0c1eeb Iustin Pop
   instances, but are not hypervisor related; e.g. the number of VCPUs,
806 5c0c1eeb Iustin Pop
   or the size of memory
807 5c0c1eeb Iustin Pop
808 5c0c1eeb Iustin Pop
#. instance parameters that are hypervisor specific (e.g. kernel_path
809 5c0c1eeb Iustin Pop
   or PAE mode)
810 5c0c1eeb Iustin Pop
811 5c0c1eeb Iustin Pop
812 5c0c1eeb Iustin Pop
The following definitions for instance parameters will be used below:
813 5c0c1eeb Iustin Pop
814 5c0c1eeb Iustin Pop
:hypervisor parameter:
815 5c0c1eeb Iustin Pop
  a hypervisor parameter (or hypervisor specific parameter) is defined
816 5c0c1eeb Iustin Pop
  as a parameter that is interpreted by the hypervisor support code in
817 5c0c1eeb Iustin Pop
  Ganeti and usually is specific to a particular hypervisor (like the
818 e2078d28 Iustin Pop
  kernel path for :term:`PVM` which makes no sense for :term:`HVM`).
819 5c0c1eeb Iustin Pop
820 5c0c1eeb Iustin Pop
:backend parameter:
821 5c0c1eeb Iustin Pop
  a backend parameter is defined as an instance parameter that can be
822 5c0c1eeb Iustin Pop
  shared among a list of instances, and is either generic enough not
823 5c0c1eeb Iustin Pop
  to be tied to a given hypervisor or cannot influence at all the
824 5c0c1eeb Iustin Pop
  hypervisor behaviour.
825 5c0c1eeb Iustin Pop
826 5c0c1eeb Iustin Pop
  For example: memory, vcpus, auto_balance
827 5c0c1eeb Iustin Pop
828 7faf5110 Michael Hanselmann
  All these parameters will be encoded into constants.py with the prefix
829 7faf5110 Michael Hanselmann
  "BE\_" and the whole list of parameters will exist in the set
830 7faf5110 Michael Hanselmann
  "BES_PARAMETERS"
831 5c0c1eeb Iustin Pop
832 5c0c1eeb Iustin Pop
:proper parameter:
833 7faf5110 Michael Hanselmann
  a parameter whose value is unique to the instance (e.g. the name of a
834 7faf5110 Michael Hanselmann
  LV, or the MAC of a NIC)
835 5c0c1eeb Iustin Pop
836 5c0c1eeb Iustin Pop
As a general rule, for all kind of parameters, “None” (or in
837 5c0c1eeb Iustin Pop
JSON-speak, “nil”) will no longer be a valid value for a parameter. As
838 5c0c1eeb Iustin Pop
such, only non-default parameters will be saved as part of objects in
839 5c0c1eeb Iustin Pop
the serialization step, reducing the size of the serialized format.
840 5c0c1eeb Iustin Pop
841 5c0c1eeb Iustin Pop
Cluster parameters
842 5c0c1eeb Iustin Pop
++++++++++++++++++
843 5c0c1eeb Iustin Pop
844 5c0c1eeb Iustin Pop
Cluster parameters remain as today, attributes at the top level of the
845 5c0c1eeb Iustin Pop
Cluster object. In addition, two new attributes at this level will
846 5c0c1eeb Iustin Pop
hold defaults for the instances:
847 5c0c1eeb Iustin Pop
848 5c0c1eeb Iustin Pop
- hvparams, a dictionary indexed by hypervisor type, holding default
849 6c2d0b44 Iustin Pop
  values for hypervisor parameters that are not defined/overridden by
850 5c0c1eeb Iustin Pop
  the instances of this hypervisor type
851 5c0c1eeb Iustin Pop
852 5c0c1eeb Iustin Pop
- beparams, a dictionary holding (for 2.0) a single element 'default',
853 5c0c1eeb Iustin Pop
  which holds the default value for backend parameters
854 5c0c1eeb Iustin Pop
855 5c0c1eeb Iustin Pop
Node parameters
856 5c0c1eeb Iustin Pop
+++++++++++++++
857 5c0c1eeb Iustin Pop
858 5c0c1eeb Iustin Pop
Node-related parameters are very few, and we will continue using the
859 5c0c1eeb Iustin Pop
same model for these as previously (attributes on the Node object).
860 5c0c1eeb Iustin Pop
861 e0eb13de Iustin Pop
There are three new node flags, described in a separate section "node
862 e0eb13de Iustin Pop
flags" below.
863 e0eb13de Iustin Pop
864 5c0c1eeb Iustin Pop
Instance parameters
865 5c0c1eeb Iustin Pop
+++++++++++++++++++
866 5c0c1eeb Iustin Pop
867 5c0c1eeb Iustin Pop
As described before, the instance parameters are split in three:
868 5c0c1eeb Iustin Pop
instance proper parameters, unique to each instance, instance
869 5c0c1eeb Iustin Pop
hypervisor parameters and instance backend parameters.
870 5c0c1eeb Iustin Pop
871 5c0c1eeb Iustin Pop
The “hvparams” and “beparams” are kept in two dictionaries at instance
872 5c0c1eeb Iustin Pop
level. Only non-default parameters are stored (but once customized, a
873 5c0c1eeb Iustin Pop
parameter will be kept, even with the same value as the default one,
874 5c0c1eeb Iustin Pop
until reset).
875 5c0c1eeb Iustin Pop
876 5c0c1eeb Iustin Pop
The names for hypervisor parameters in the instance.hvparams subtree
877 5c0c1eeb Iustin Pop
should be choosen as generic as possible, especially if specific
878 5c0c1eeb Iustin Pop
parameters could conceivably be useful for more than one hypervisor,
879 6c2d0b44 Iustin Pop
e.g. ``instance.hvparams.vnc_console_port`` instead of using both
880 6c2d0b44 Iustin Pop
``instance.hvparams.hvm_vnc_console_port`` and
881 6c2d0b44 Iustin Pop
``instance.hvparams.kvm_vnc_console_port``.
882 5c0c1eeb Iustin Pop
883 5c0c1eeb Iustin Pop
There are some special cases related to disks and NICs (for example):
884 6c2d0b44 Iustin Pop
a disk has both Ganeti-related parameters (e.g. the name of the LV)
885 5c0c1eeb Iustin Pop
and hypervisor-related parameters (how the disk is presented to/named
886 5c0c1eeb Iustin Pop
in the instance). The former parameters remain as proper-instance
887 5c0c1eeb Iustin Pop
parameters, while the latter value are migrated to the hvparams
888 5c0c1eeb Iustin Pop
structure. In 2.0, we will have only globally-per-instance such
889 5c0c1eeb Iustin Pop
hypervisor parameters, and not per-disk ones (e.g. all NICs will be
890 5c0c1eeb Iustin Pop
exported as of the same type).
891 5c0c1eeb Iustin Pop
892 5c0c1eeb Iustin Pop
Starting from the 1.2 list of instance parameters, here is how they
893 5c0c1eeb Iustin Pop
will be mapped to the three classes of parameters:
894 5c0c1eeb Iustin Pop
895 5c0c1eeb Iustin Pop
- name (P)
896 5c0c1eeb Iustin Pop
- primary_node (P)
897 5c0c1eeb Iustin Pop
- os (P)
898 5c0c1eeb Iustin Pop
- hypervisor (P)
899 5c0c1eeb Iustin Pop
- status (P)
900 5c0c1eeb Iustin Pop
- memory (BE)
901 5c0c1eeb Iustin Pop
- vcpus (BE)
902 5c0c1eeb Iustin Pop
- nics (P)
903 5c0c1eeb Iustin Pop
- disks (P)
904 5c0c1eeb Iustin Pop
- disk_template (P)
905 5c0c1eeb Iustin Pop
- network_port (P)
906 5c0c1eeb Iustin Pop
- kernel_path (HV)
907 5c0c1eeb Iustin Pop
- initrd_path (HV)
908 5c0c1eeb Iustin Pop
- hvm_boot_order (HV)
909 5c0c1eeb Iustin Pop
- hvm_acpi (HV)
910 5c0c1eeb Iustin Pop
- hvm_pae (HV)
911 5c0c1eeb Iustin Pop
- hvm_cdrom_image_path (HV)
912 5c0c1eeb Iustin Pop
- hvm_nic_type (HV)
913 5c0c1eeb Iustin Pop
- hvm_disk_type (HV)
914 5c0c1eeb Iustin Pop
- vnc_bind_address (HV)
915 5c0c1eeb Iustin Pop
- serial_no (P)
916 5c0c1eeb Iustin Pop
917 5c0c1eeb Iustin Pop
918 5c0c1eeb Iustin Pop
Parameter validation
919 5c0c1eeb Iustin Pop
++++++++++++++++++++
920 5c0c1eeb Iustin Pop
921 5c0c1eeb Iustin Pop
To support the new cluster parameter design, additional features will
922 5c0c1eeb Iustin Pop
be required from the hypervisor support implementations in Ganeti.
923 5c0c1eeb Iustin Pop
924 5c0c1eeb Iustin Pop
The hypervisor support  implementation API will be extended with the
925 5c0c1eeb Iustin Pop
following features:
926 5c0c1eeb Iustin Pop
927 5c0c1eeb Iustin Pop
:PARAMETERS: class-level attribute holding the list of valid parameters
928 5c0c1eeb Iustin Pop
  for this hypervisor
929 5c0c1eeb Iustin Pop
:CheckParamSyntax(hvparams): checks that the given parameters are
930 5c0c1eeb Iustin Pop
  valid (as in the names are valid) for this hypervisor; usually just
931 6c2d0b44 Iustin Pop
  comparing ``hvparams.keys()`` and ``cls.PARAMETERS``; this is a class
932 6c2d0b44 Iustin Pop
  method that can be called from within master code (i.e. cmdlib) and
933 6c2d0b44 Iustin Pop
  should be safe to do so
934 5c0c1eeb Iustin Pop
:ValidateParameters(hvparams): verifies the values of the provided
935 5c0c1eeb Iustin Pop
  parameters against this hypervisor; this is a method that will be
936 5c0c1eeb Iustin Pop
  called on the target node, from backend.py code, and as such can
937 5c0c1eeb Iustin Pop
  make node-specific checks (e.g. kernel_path checking)
938 5c0c1eeb Iustin Pop
939 5c0c1eeb Iustin Pop
Default value application
940 5c0c1eeb Iustin Pop
+++++++++++++++++++++++++
941 5c0c1eeb Iustin Pop
942 5c0c1eeb Iustin Pop
The application of defaults to an instance is done in the Cluster
943 5c0c1eeb Iustin Pop
object, via two new methods as follows:
944 5c0c1eeb Iustin Pop
945 5c0c1eeb Iustin Pop
- ``Cluster.FillHV(instance)``, returns 'filled' hvparams dict, based on
946 5c0c1eeb Iustin Pop
  instance's hvparams and cluster's ``hvparams[instance.hypervisor]``
947 5c0c1eeb Iustin Pop
948 5c0c1eeb Iustin Pop
- ``Cluster.FillBE(instance, be_type="default")``, which returns the
949 5c0c1eeb Iustin Pop
  beparams dict, based on the instance and cluster beparams
950 5c0c1eeb Iustin Pop
951 7faf5110 Michael Hanselmann
The FillHV/BE transformations will be used, for example, in the
952 7faf5110 Michael Hanselmann
RpcRunner when sending an instance for activation/stop, and the sent
953 7faf5110 Michael Hanselmann
instance hvparams/beparams will have the final value (noded code doesn't
954 7faf5110 Michael Hanselmann
know about defaults).
955 5c0c1eeb Iustin Pop
956 5c0c1eeb Iustin Pop
LU code will need to self-call the transformation, if needed.
957 5c0c1eeb Iustin Pop
958 5c0c1eeb Iustin Pop
Opcode changes
959 5c0c1eeb Iustin Pop
++++++++++++++
960 5c0c1eeb Iustin Pop
961 5c0c1eeb Iustin Pop
The parameter changes will have impact on the OpCodes, especially on
962 5c0c1eeb Iustin Pop
the following ones:
963 5c0c1eeb Iustin Pop
964 7faf5110 Michael Hanselmann
- ``OpCreateInstance``, where the new hv and be parameters will be sent
965 7faf5110 Michael Hanselmann
  as dictionaries; note that all hv and be parameters are now optional,
966 7faf5110 Michael Hanselmann
  as the values can be instead taken from the cluster
967 6c2d0b44 Iustin Pop
- ``OpQueryInstances``, where we have to be able to query these new
968 5c0c1eeb Iustin Pop
  parameters; the syntax for names will be ``hvparam/$NAME`` and
969 5c0c1eeb Iustin Pop
  ``beparam/$NAME`` for querying an individual parameter out of one
970 5c0c1eeb Iustin Pop
  dictionary, and ``hvparams``, respectively ``beparams``, for the whole
971 5c0c1eeb Iustin Pop
  dictionaries
972 6c2d0b44 Iustin Pop
- ``OpModifyInstance``, where the the modified parameters are sent as
973 5c0c1eeb Iustin Pop
  dictionaries
974 5c0c1eeb Iustin Pop
975 5c0c1eeb Iustin Pop
Additionally, we will need new OpCodes to modify the cluster-level
976 5c0c1eeb Iustin Pop
defaults for the be/hv sets of parameters.
977 5c0c1eeb Iustin Pop
978 5c0c1eeb Iustin Pop
Caveats
979 5c0c1eeb Iustin Pop
+++++++
980 5c0c1eeb Iustin Pop
981 5c0c1eeb Iustin Pop
One problem that might appear is that our classification is not
982 5c0c1eeb Iustin Pop
complete or not good enough, and we'll need to change this model. As
983 5c0c1eeb Iustin Pop
the last resort, we will need to rollback and keep 1.2 style.
984 5c0c1eeb Iustin Pop
985 5c0c1eeb Iustin Pop
Another problem is that classification of one parameter is unclear
986 5c0c1eeb Iustin Pop
(e.g. ``network_port``, is this BE or HV?); in this case we'll take
987 5c0c1eeb Iustin Pop
the risk of having to move parameters later between classes.
988 5c0c1eeb Iustin Pop
989 5c0c1eeb Iustin Pop
Security
990 5c0c1eeb Iustin Pop
++++++++
991 5c0c1eeb Iustin Pop
992 5c0c1eeb Iustin Pop
The only security issue that we foresee is if some new parameters will
993 5c0c1eeb Iustin Pop
have sensitive value. If so, we will need to have a way to export the
994 5c0c1eeb Iustin Pop
config data while purging the sensitive value.
995 5c0c1eeb Iustin Pop
996 5c0c1eeb Iustin Pop
E.g. for the drbd shared secrets, we could export these with the
997 5c0c1eeb Iustin Pop
values replaced by an empty string.
998 5c0c1eeb Iustin Pop
999 e0eb13de Iustin Pop
Node flags
1000 e0eb13de Iustin Pop
~~~~~~~~~~
1001 e0eb13de Iustin Pop
1002 e0eb13de Iustin Pop
Ganeti 2.0 adds three node flags that change the way nodes are handled
1003 e0eb13de Iustin Pop
within Ganeti and the related infrastructure (iallocator interaction,
1004 e0eb13de Iustin Pop
RAPI data export).
1005 e0eb13de Iustin Pop
1006 e0eb13de Iustin Pop
*master candidate* flag
1007 e0eb13de Iustin Pop
+++++++++++++++++++++++
1008 e0eb13de Iustin Pop
1009 e0eb13de Iustin Pop
Ganeti 2.0 allows more scalability in operation by introducing
1010 e0eb13de Iustin Pop
parallelization. However, a new bottleneck is reached that is the
1011 e0eb13de Iustin Pop
synchronization and replication of cluster configuration to all nodes
1012 e0eb13de Iustin Pop
in the cluster.
1013 e0eb13de Iustin Pop
1014 e0eb13de Iustin Pop
This breaks scalability as the speed of the replication decreases
1015 e0eb13de Iustin Pop
roughly with the size of the nodes in the cluster. The goal of the
1016 e0eb13de Iustin Pop
master candidate flag is to change this O(n) into O(1) with respect to
1017 e0eb13de Iustin Pop
job and configuration data propagation.
1018 e0eb13de Iustin Pop
1019 e0eb13de Iustin Pop
Only nodes having this flag set (let's call this set of nodes the
1020 e0eb13de Iustin Pop
*candidate pool*) will have jobs and configuration data replicated.
1021 e0eb13de Iustin Pop
1022 e0eb13de Iustin Pop
The cluster will have a new parameter (runtime changeable) called
1023 e0eb13de Iustin Pop
``candidate_pool_size`` which represents the number of candidates the
1024 e0eb13de Iustin Pop
cluster tries to maintain (preferably automatically).
1025 e0eb13de Iustin Pop
1026 e0eb13de Iustin Pop
This will impact the cluster operations as follows:
1027 e0eb13de Iustin Pop
1028 e0eb13de Iustin Pop
- jobs and config data will be replicated only to a fixed set of nodes
1029 e0eb13de Iustin Pop
- master fail-over will only be possible to a node in the candidate pool
1030 e0eb13de Iustin Pop
- cluster verify needs changing to account for these two roles
1031 e0eb13de Iustin Pop
- external scripts will no longer have access to the configuration
1032 e0eb13de Iustin Pop
  file (this is not recommended anyway)
1033 e0eb13de Iustin Pop
1034 e0eb13de Iustin Pop
1035 e0eb13de Iustin Pop
The caveats of this change are:
1036 e0eb13de Iustin Pop
1037 e0eb13de Iustin Pop
- if all candidates are lost (completely), cluster configuration is
1038 e0eb13de Iustin Pop
  lost (but it should be backed up external to the cluster anyway)
1039 e0eb13de Iustin Pop
1040 e0eb13de Iustin Pop
- failed nodes which are candidate must be dealt with properly, so
1041 e0eb13de Iustin Pop
  that we don't lose too many candidates at the same time; this will be
1042 e0eb13de Iustin Pop
  reported in cluster verify
1043 e0eb13de Iustin Pop
1044 e0eb13de Iustin Pop
- the 'all equal' concept of ganeti is no longer true
1045 e0eb13de Iustin Pop
1046 e0eb13de Iustin Pop
- the partial distribution of config data means that all nodes will
1047 e0eb13de Iustin Pop
  have to revert to ssconf files for master info (as in 1.2)
1048 e0eb13de Iustin Pop
1049 e0eb13de Iustin Pop
Advantages:
1050 e0eb13de Iustin Pop
1051 e0eb13de Iustin Pop
- speed on a 100+ nodes simulated cluster is greatly enhanced, even
1052 e0eb13de Iustin Pop
  for a simple operation; ``gnt-instance remove`` on a diskless instance
1053 e0eb13de Iustin Pop
  remove goes from ~9seconds to ~2 seconds
1054 e0eb13de Iustin Pop
1055 e0eb13de Iustin Pop
- node failure of non-candidates will be less impacting on the cluster
1056 e0eb13de Iustin Pop
1057 e0eb13de Iustin Pop
The default value for the candidate pool size will be set to 10 but
1058 e0eb13de Iustin Pop
this can be changed at cluster creation and modified any time later.
1059 e0eb13de Iustin Pop
1060 e0eb13de Iustin Pop
Testing on simulated big clusters with sequential and parallel jobs
1061 e0eb13de Iustin Pop
show that this value (10) is a sweet-spot from performance and load
1062 e0eb13de Iustin Pop
point of view.
1063 e0eb13de Iustin Pop
1064 e0eb13de Iustin Pop
*offline* flag
1065 e0eb13de Iustin Pop
++++++++++++++
1066 e0eb13de Iustin Pop
1067 e0eb13de Iustin Pop
In order to support better the situation in which nodes are offline
1068 e0eb13de Iustin Pop
(e.g. for repair) without altering the cluster configuration, Ganeti
1069 e0eb13de Iustin Pop
needs to be told and needs to properly handle this state for nodes.
1070 e0eb13de Iustin Pop
1071 e0eb13de Iustin Pop
This will result in simpler procedures, and less mistakes, when the
1072 e0eb13de Iustin Pop
amount of node failures is high on an absolute scale (either due to
1073 e0eb13de Iustin Pop
high failure rate or simply big clusters).
1074 e0eb13de Iustin Pop
1075 e0eb13de Iustin Pop
Nodes having this attribute set will not be contacted for inter-node
1076 e0eb13de Iustin Pop
RPC calls, will not be master candidates, and will not be able to host
1077 e0eb13de Iustin Pop
instances as primaries.
1078 e0eb13de Iustin Pop
1079 e0eb13de Iustin Pop
Setting this attribute on a node:
1080 e0eb13de Iustin Pop
1081 e0eb13de Iustin Pop
- will not be allowed if the node is the master
1082 e0eb13de Iustin Pop
- will not be allowed if the node has primary instances
1083 e0eb13de Iustin Pop
- will cause the node to be demoted from the master candidate role (if
1084 e0eb13de Iustin Pop
  it was), possibly causing another node to be promoted to that role
1085 e0eb13de Iustin Pop
1086 e0eb13de Iustin Pop
This attribute will impact the cluster operations as follows:
1087 e0eb13de Iustin Pop
1088 e0eb13de Iustin Pop
- querying these nodes for anything will fail instantly in the RPC
1089 e0eb13de Iustin Pop
  library, with a specific RPC error (RpcResult.offline == True)
1090 e0eb13de Iustin Pop
1091 e0eb13de Iustin Pop
- they will be listed in the Other section of cluster verify
1092 e0eb13de Iustin Pop
1093 e0eb13de Iustin Pop
The code is changed in the following ways:
1094 e0eb13de Iustin Pop
1095 e0eb13de Iustin Pop
- RPC calls were be converted to skip such nodes:
1096 e0eb13de Iustin Pop
1097 e0eb13de Iustin Pop
  - RpcRunner-instance-based RPC calls are easy to convert
1098 e0eb13de Iustin Pop
1099 e0eb13de Iustin Pop
  - static/classmethod RPC calls are harder to convert, and were left
1100 e0eb13de Iustin Pop
    alone
1101 e0eb13de Iustin Pop
1102 e0eb13de Iustin Pop
- the RPC results were unified so that this new result state (offline)
1103 e0eb13de Iustin Pop
  can be differentiated
1104 e0eb13de Iustin Pop
1105 e0eb13de Iustin Pop
- master voting still queries in repair nodes, as we need to ensure
1106 e0eb13de Iustin Pop
  consistency in case the (wrong) masters have old data, and nodes have
1107 e0eb13de Iustin Pop
  come back from repairs
1108 e0eb13de Iustin Pop
1109 e0eb13de Iustin Pop
Caveats:
1110 e0eb13de Iustin Pop
1111 e0eb13de Iustin Pop
- some operation semantics are less clear (e.g. what to do on instance
1112 7faf5110 Michael Hanselmann
  start with offline secondary?); for now, these will just fail as if
1113 7faf5110 Michael Hanselmann
  the flag is not set (but faster)
1114 e0eb13de Iustin Pop
- 2-node cluster with one node offline needs manual startup of the
1115 e0eb13de Iustin Pop
  master with a special flag to skip voting (as the master can't get a
1116 e0eb13de Iustin Pop
  quorum there)
1117 e0eb13de Iustin Pop
1118 e0eb13de Iustin Pop
One of the advantages of implementing this flag is that it will allow
1119 e0eb13de Iustin Pop
in the future automation tools to automatically put the node in
1120 e0eb13de Iustin Pop
repairs and recover from this state, and the code (should/will) handle
1121 e0eb13de Iustin Pop
this much better than just timing out. So, future possible
1122 e0eb13de Iustin Pop
improvements (for later versions):
1123 e0eb13de Iustin Pop
1124 e0eb13de Iustin Pop
- watcher will detect nodes which fail RPC calls, will attempt to ssh
1125 e0eb13de Iustin Pop
  to them, if failure will put them offline
1126 e0eb13de Iustin Pop
- watcher will try to ssh and query the offline nodes, if successful
1127 e0eb13de Iustin Pop
  will take them off the repair list
1128 e0eb13de Iustin Pop
1129 e0eb13de Iustin Pop
Alternatives considered: The RPC call model in 2.0 is, by default,
1130 e0eb13de Iustin Pop
much nicer - errors are logged in the background, and job/opcode
1131 e0eb13de Iustin Pop
execution is clearer, so we could simply not introduce this. However,
1132 e0eb13de Iustin Pop
having this state will make both the codepaths clearer (offline
1133 e0eb13de Iustin Pop
vs. temporary failure) and the operational model (it's not a node with
1134 e0eb13de Iustin Pop
errors, but an offline node).
1135 e0eb13de Iustin Pop
1136 e0eb13de Iustin Pop
1137 e0eb13de Iustin Pop
*drained* flag
1138 e0eb13de Iustin Pop
++++++++++++++
1139 e0eb13de Iustin Pop
1140 e0eb13de Iustin Pop
Due to parallel execution of jobs in Ganeti 2.0, we could have the
1141 e0eb13de Iustin Pop
following situation:
1142 e0eb13de Iustin Pop
1143 e0eb13de Iustin Pop
- gnt-node migrate + failover is run
1144 e0eb13de Iustin Pop
- gnt-node evacuate is run, which schedules a long-running 6-opcode
1145 e0eb13de Iustin Pop
  job for the node
1146 e0eb13de Iustin Pop
- partway through, a new job comes in that runs an iallocator script,
1147 e0eb13de Iustin Pop
  which finds the above node as empty and a very good candidate
1148 e0eb13de Iustin Pop
- gnt-node evacuate has finished, but now it has to be run again, to
1149 e0eb13de Iustin Pop
  clean the above instance(s)
1150 e0eb13de Iustin Pop
1151 e0eb13de Iustin Pop
In order to prevent this situation, and to be able to get nodes into
1152 7faf5110 Michael Hanselmann
proper offline status easily, a new *drained* flag was added to the
1153 7faf5110 Michael Hanselmann
nodes.
1154 e0eb13de Iustin Pop
1155 e0eb13de Iustin Pop
This flag (which actually means "is being, or was drained, and is
1156 e0eb13de Iustin Pop
expected to go offline"), will prevent allocations on the node, but
1157 e0eb13de Iustin Pop
otherwise all other operations (start/stop instance, query, etc.) are
1158 e0eb13de Iustin Pop
working without any restrictions.
1159 e0eb13de Iustin Pop
1160 e0eb13de Iustin Pop
Interaction between flags
1161 e0eb13de Iustin Pop
+++++++++++++++++++++++++
1162 e0eb13de Iustin Pop
1163 e0eb13de Iustin Pop
While these flags are implemented as separate flags, they are
1164 e0eb13de Iustin Pop
mutually-exclusive and are acting together with the master node role
1165 e0eb13de Iustin Pop
as a single *node status* value. In other words, a flag is only in one
1166 e0eb13de Iustin Pop
of these roles at a given time. The lack of any of these flags denote
1167 e0eb13de Iustin Pop
a regular node.
1168 e0eb13de Iustin Pop
1169 e0eb13de Iustin Pop
The current node status is visible in the ``gnt-cluster verify``
1170 e0eb13de Iustin Pop
output, and the individual flags can be examined via separate flags in
1171 e0eb13de Iustin Pop
the ``gnt-node list`` output.
1172 e0eb13de Iustin Pop
1173 e0eb13de Iustin Pop
These new flags will be exported in both the iallocator input message
1174 e0eb13de Iustin Pop
and via RAPI, see the respective man pages for the exact names.
1175 e0eb13de Iustin Pop
1176 5c0c1eeb Iustin Pop
Feature changes
1177 5c0c1eeb Iustin Pop
---------------
1178 5c0c1eeb Iustin Pop
1179 5c0c1eeb Iustin Pop
The main feature-level changes will be:
1180 5c0c1eeb Iustin Pop
1181 5c0c1eeb Iustin Pop
- a number of disk related changes
1182 5c0c1eeb Iustin Pop
- removal of fixed two-disk, one-nic per instance limitation
1183 5c0c1eeb Iustin Pop
1184 5c0c1eeb Iustin Pop
Disk handling changes
1185 5c0c1eeb Iustin Pop
~~~~~~~~~~~~~~~~~~~~~
1186 5c0c1eeb Iustin Pop
1187 5c0c1eeb Iustin Pop
The storage options available in Ganeti 1.x were introduced based on
1188 5c0c1eeb Iustin Pop
then-current software (first DRBD 0.7 then later DRBD 8) and the
1189 5c0c1eeb Iustin Pop
estimated usage patters. However, experience has later shown that some
1190 5c0c1eeb Iustin Pop
assumptions made initially are not true and that more flexibility is
1191 5c0c1eeb Iustin Pop
needed.
1192 5c0c1eeb Iustin Pop
1193 7faf5110 Michael Hanselmann
One main assumption made was that disk failures should be treated as
1194 7faf5110 Michael Hanselmann
'rare' events, and that each of them needs to be manually handled in
1195 7faf5110 Michael Hanselmann
order to ensure data safety; however, both these assumptions are false:
1196 5c0c1eeb Iustin Pop
1197 7faf5110 Michael Hanselmann
- disk failures can be a common occurrence, based on usage patterns or
1198 7faf5110 Michael Hanselmann
  cluster size
1199 7faf5110 Michael Hanselmann
- our disk setup is robust enough (referring to DRBD8 + LVM) that we
1200 7faf5110 Michael Hanselmann
  could automate more of the recovery
1201 5c0c1eeb Iustin Pop
1202 7faf5110 Michael Hanselmann
Note that we still don't have fully-automated disk recovery as a goal,
1203 7faf5110 Michael Hanselmann
but our goal is to reduce the manual work needed.
1204 5c0c1eeb Iustin Pop
1205 5c0c1eeb Iustin Pop
As such, we plan the following main changes:
1206 5c0c1eeb Iustin Pop
1207 7faf5110 Michael Hanselmann
- DRBD8 is much more flexible and stable than its previous version
1208 7faf5110 Michael Hanselmann
  (0.7), such that removing the support for the ``remote_raid1``
1209 7faf5110 Michael Hanselmann
  template and focusing only on DRBD8 is easier
1210 5c0c1eeb Iustin Pop
1211 7faf5110 Michael Hanselmann
- dynamic discovery of DRBD devices is not actually needed in a cluster
1212 7faf5110 Michael Hanselmann
  that where the DRBD namespace is controlled by Ganeti; switching to a
1213 7faf5110 Michael Hanselmann
  static assignment (done at either instance creation time or change
1214 7faf5110 Michael Hanselmann
  secondary time) will change the disk activation time from O(n) to
1215 7faf5110 Michael Hanselmann
  O(1), which on big clusters is a significant gain
1216 5c0c1eeb Iustin Pop
1217 7faf5110 Michael Hanselmann
- remove the hard dependency on LVM (currently all available storage
1218 7faf5110 Michael Hanselmann
  types are ultimately backed by LVM volumes) by introducing file-based
1219 7faf5110 Michael Hanselmann
  storage
1220 5c0c1eeb Iustin Pop
1221 5c0c1eeb Iustin Pop
Additionally, a number of smaller enhancements are also planned:
1222 5c0c1eeb Iustin Pop
- support variable number of disks
1223 5c0c1eeb Iustin Pop
- support read-only disks
1224 5c0c1eeb Iustin Pop
1225 5c0c1eeb Iustin Pop
Future enhancements in the 2.x series, which do not require base design
1226 5c0c1eeb Iustin Pop
changes, might include:
1227 5c0c1eeb Iustin Pop
1228 5c0c1eeb Iustin Pop
- enhancement of the LVM allocation method in order to try to keep
1229 5c0c1eeb Iustin Pop
  all of an instance's virtual disks on the same physical
1230 5c0c1eeb Iustin Pop
  disks
1231 5c0c1eeb Iustin Pop
1232 5c0c1eeb Iustin Pop
- add support for DRBD8 authentication at handshake time in
1233 5c0c1eeb Iustin Pop
  order to ensure each device connects to the correct peer
1234 5c0c1eeb Iustin Pop
1235 5c0c1eeb Iustin Pop
- remove the restrictions on failover only to the secondary
1236 5c0c1eeb Iustin Pop
  which creates very strict rules on cluster allocation
1237 5c0c1eeb Iustin Pop
1238 5c0c1eeb Iustin Pop
DRBD minor allocation
1239 5c0c1eeb Iustin Pop
+++++++++++++++++++++
1240 5c0c1eeb Iustin Pop
1241 5c0c1eeb Iustin Pop
Currently, when trying to identify or activate a new DRBD (or MD)
1242 5c0c1eeb Iustin Pop
device, the code scans all in-use devices in order to see if we find
1243 5c0c1eeb Iustin Pop
one that looks similar to our parameters and is already in the desired
1244 5c0c1eeb Iustin Pop
state or not. Since this needs external commands to be run, it is very
1245 5c0c1eeb Iustin Pop
slow when more than a few devices are already present.
1246 5c0c1eeb Iustin Pop
1247 5c0c1eeb Iustin Pop
Therefore, we will change the discovery model from dynamic to
1248 5c0c1eeb Iustin Pop
static. When a new device is logically created (added to the
1249 5c0c1eeb Iustin Pop
configuration) a free minor number is computed from the list of
1250 5c0c1eeb Iustin Pop
devices that should exist on that node and assigned to that
1251 5c0c1eeb Iustin Pop
device.
1252 5c0c1eeb Iustin Pop
1253 5c0c1eeb Iustin Pop
At device activation, if the minor is already in use, we check if
1254 5c0c1eeb Iustin Pop
it has our parameters; if not so, we just destroy the device (if
1255 5c0c1eeb Iustin Pop
possible, otherwise we abort) and start it with our own
1256 5c0c1eeb Iustin Pop
parameters.
1257 5c0c1eeb Iustin Pop
1258 5c0c1eeb Iustin Pop
This means that we in effect take ownership of the minor space for
1259 6c2d0b44 Iustin Pop
that device type; if there's a user-created DRBD minor, it will be
1260 5c0c1eeb Iustin Pop
automatically removed.
1261 5c0c1eeb Iustin Pop
1262 5c0c1eeb Iustin Pop
The change will have the effect of reducing the number of external
1263 5c0c1eeb Iustin Pop
commands run per device from a constant number times the index of the
1264 5c0c1eeb Iustin Pop
first free DRBD minor to just a constant number.
1265 5c0c1eeb Iustin Pop
1266 6c2d0b44 Iustin Pop
Removal of obsolete device types (MD, DRBD7)
1267 5c0c1eeb Iustin Pop
++++++++++++++++++++++++++++++++++++++++++++
1268 5c0c1eeb Iustin Pop
1269 5c0c1eeb Iustin Pop
We need to remove these device types because of two issues. First,
1270 6c2d0b44 Iustin Pop
DRBD7 has bad failure modes in case of dual failures (both network and
1271 5c0c1eeb Iustin Pop
disk - it cannot propagate the error up the device stack and instead
1272 6c2d0b44 Iustin Pop
just panics. Second, due to the asymmetry between primary and
1273 6c2d0b44 Iustin Pop
secondary in MD+DRBD mode, we cannot do live failover (not even if we
1274 6c2d0b44 Iustin Pop
had MD+DRBD8).
1275 5c0c1eeb Iustin Pop
1276 5c0c1eeb Iustin Pop
File-based storage support
1277 5c0c1eeb Iustin Pop
++++++++++++++++++++++++++
1278 5c0c1eeb Iustin Pop
1279 6c2d0b44 Iustin Pop
Using files instead of logical volumes for instance storage would
1280 6c2d0b44 Iustin Pop
allow us to get rid of the hard requirement for volume groups for
1281 6c2d0b44 Iustin Pop
testing clusters and it would also allow usage of SAN storage to do
1282 6c2d0b44 Iustin Pop
live failover taking advantage of this storage solution.
1283 5c0c1eeb Iustin Pop
1284 5c0c1eeb Iustin Pop
Better LVM allocation
1285 5c0c1eeb Iustin Pop
+++++++++++++++++++++
1286 5c0c1eeb Iustin Pop
1287 5c0c1eeb Iustin Pop
Currently, the LV to PV allocation mechanism is a very simple one: at
1288 5c0c1eeb Iustin Pop
each new request for a logical volume, tell LVM to allocate the volume
1289 5c0c1eeb Iustin Pop
in order based on the amount of free space. This is good for
1290 5c0c1eeb Iustin Pop
simplicity and for keeping the usage equally spread over the available
1291 5c0c1eeb Iustin Pop
physical disks, however it introduces a problem that an instance could
1292 5c0c1eeb Iustin Pop
end up with its (currently) two drives on two physical disks, or
1293 5c0c1eeb Iustin Pop
(worse) that the data and metadata for a DRBD device end up on
1294 5c0c1eeb Iustin Pop
different drives.
1295 5c0c1eeb Iustin Pop
1296 5c0c1eeb Iustin Pop
This is bad because it causes unneeded ``replace-disks`` operations in
1297 5c0c1eeb Iustin Pop
case of a physical failure.
1298 5c0c1eeb Iustin Pop
1299 5c0c1eeb Iustin Pop
The solution is to batch allocations for an instance and make the LVM
1300 5c0c1eeb Iustin Pop
handling code try to allocate as close as possible all the storage of
1301 5c0c1eeb Iustin Pop
one instance. We will still allow the logical volumes to spill over to
1302 5c0c1eeb Iustin Pop
additional disks as needed.
1303 5c0c1eeb Iustin Pop
1304 5c0c1eeb Iustin Pop
Note that this clustered allocation can only be attempted at initial
1305 5c0c1eeb Iustin Pop
instance creation, or at change secondary node time. At add disk time,
1306 5c0c1eeb Iustin Pop
or at replacing individual disks, it's not easy enough to compute the
1307 5c0c1eeb Iustin Pop
current disk map so we'll not attempt the clustering.
1308 5c0c1eeb Iustin Pop
1309 5c0c1eeb Iustin Pop
DRBD8 peer authentication at handshake
1310 5c0c1eeb Iustin Pop
++++++++++++++++++++++++++++++++++++++
1311 5c0c1eeb Iustin Pop
1312 5c0c1eeb Iustin Pop
DRBD8 has a new feature that allow authentication of the peer at
1313 5c0c1eeb Iustin Pop
connect time. We can use this to prevent connecting to the wrong peer
1314 5c0c1eeb Iustin Pop
more that securing the connection. Even though we never had issues
1315 5c0c1eeb Iustin Pop
with wrong connections, it would be good to implement this.
1316 5c0c1eeb Iustin Pop
1317 5c0c1eeb Iustin Pop
1318 5c0c1eeb Iustin Pop
LVM self-repair (optional)
1319 5c0c1eeb Iustin Pop
++++++++++++++++++++++++++
1320 5c0c1eeb Iustin Pop
1321 5c0c1eeb Iustin Pop
The complete failure of a physical disk is very tedious to
1322 5c0c1eeb Iustin Pop
troubleshoot, mainly because of the many failure modes and the many
1323 5c0c1eeb Iustin Pop
steps needed. We can safely automate some of the steps, more
1324 5c0c1eeb Iustin Pop
specifically the ``vgreduce --removemissing`` using the following
1325 5c0c1eeb Iustin Pop
method:
1326 5c0c1eeb Iustin Pop
1327 5c0c1eeb Iustin Pop
#. check if all nodes have consistent volume groups
1328 5c0c1eeb Iustin Pop
#. if yes, and previous status was yes, do nothing
1329 5c0c1eeb Iustin Pop
#. if yes, and previous status was no, save status and restart
1330 5c0c1eeb Iustin Pop
#. if no, and previous status was no, do nothing
1331 5c0c1eeb Iustin Pop
#. if no, and previous status was yes:
1332 5c0c1eeb Iustin Pop
    #. if more than one node is inconsistent, do nothing
1333 6c2d0b44 Iustin Pop
    #. if only one node is inconsistent:
1334 5c0c1eeb Iustin Pop
        #. run ``vgreduce --removemissing``
1335 6c2d0b44 Iustin Pop
        #. log this occurrence in the Ganeti log in a form that
1336 5c0c1eeb Iustin Pop
           can be used for monitoring
1337 5c0c1eeb Iustin Pop
        #. [FUTURE] run ``replace-disks`` for all
1338 5c0c1eeb Iustin Pop
           instances affected
1339 5c0c1eeb Iustin Pop
1340 5c0c1eeb Iustin Pop
Failover to any node
1341 5c0c1eeb Iustin Pop
++++++++++++++++++++
1342 5c0c1eeb Iustin Pop
1343 5c0c1eeb Iustin Pop
With a modified disk activation sequence, we can implement the
1344 5c0c1eeb Iustin Pop
*failover to any* functionality, removing many of the layout
1345 5c0c1eeb Iustin Pop
restrictions of a cluster:
1346 5c0c1eeb Iustin Pop
1347 7faf5110 Michael Hanselmann
- the need to reserve memory on the current secondary: this gets reduced
1348 7faf5110 Michael Hanselmann
  to a must to reserve memory anywhere on the cluster
1349 5c0c1eeb Iustin Pop
1350 5c0c1eeb Iustin Pop
- the need to first failover and then replace secondary for an
1351 5c0c1eeb Iustin Pop
  instance: with failover-to-any, we can directly failover to
1352 5c0c1eeb Iustin Pop
  another node, which also does the replace disks at the same
1353 5c0c1eeb Iustin Pop
  step
1354 5c0c1eeb Iustin Pop
1355 5c0c1eeb Iustin Pop
In the following, we denote the current primary by P1, the current
1356 5c0c1eeb Iustin Pop
secondary by S1, and the new primary and secondaries by P2 and S2. P2
1357 5c0c1eeb Iustin Pop
is fixed to the node the user chooses, but the choice of S2 can be
1358 5c0c1eeb Iustin Pop
made between P1 and S1. This choice can be constrained, depending on
1359 5c0c1eeb Iustin Pop
which of P1 and S1 has failed.
1360 5c0c1eeb Iustin Pop
1361 7faf5110 Michael Hanselmann
- if P1 has failed, then S1 must become S2, and live migration is not
1362 7faf5110 Michael Hanselmann
  possible
1363 5c0c1eeb Iustin Pop
- if S1 has failed, then P1 must become S2, and live migration could be
1364 5c0c1eeb Iustin Pop
  possible (in theory, but this is not a design goal for 2.0)
1365 5c0c1eeb Iustin Pop
1366 5c0c1eeb Iustin Pop
The algorithm for performing the failover is straightforward:
1367 5c0c1eeb Iustin Pop
1368 5c0c1eeb Iustin Pop
- verify that S2 (the node the user has chosen to keep as secondary) has
1369 5c0c1eeb Iustin Pop
  valid data (is consistent)
1370 5c0c1eeb Iustin Pop
1371 7faf5110 Michael Hanselmann
- tear down the current DRBD association and setup a DRBD pairing
1372 7faf5110 Michael Hanselmann
  between P2 (P2 is indicated by the user) and S2; since P2 has no data,
1373 7faf5110 Michael Hanselmann
  it will start re-syncing from S2
1374 5c0c1eeb Iustin Pop
1375 7faf5110 Michael Hanselmann
- as soon as P2 is in state SyncTarget (i.e. after the resync has
1376 7faf5110 Michael Hanselmann
  started but before it has finished), we can promote it to primary role
1377 7faf5110 Michael Hanselmann
  (r/w) and start the instance on P2
1378 5c0c1eeb Iustin Pop
1379 5c0c1eeb Iustin Pop
- as soon as the P2?S2 sync has finished, we can remove
1380 5c0c1eeb Iustin Pop
  the old data on the old node that has not been chosen for
1381 5c0c1eeb Iustin Pop
  S2
1382 5c0c1eeb Iustin Pop
1383 5c0c1eeb Iustin Pop
Caveats: during the P2?S2 sync, a (non-transient) network error
1384 5c0c1eeb Iustin Pop
will cause I/O errors on the instance, so (if a longer instance
1385 5c0c1eeb Iustin Pop
downtime is acceptable) we can postpone the restart of the instance
1386 5c0c1eeb Iustin Pop
until the resync is done. However, disk I/O errors on S2 will cause
1387 6c2d0b44 Iustin Pop
data loss, since we don't have a good copy of the data anymore, so in
1388 5c0c1eeb Iustin Pop
this case waiting for the sync to complete is not an option. As such,
1389 5c0c1eeb Iustin Pop
it is recommended that this feature is used only in conjunction with
1390 5c0c1eeb Iustin Pop
proper disk monitoring.
1391 5c0c1eeb Iustin Pop
1392 5c0c1eeb Iustin Pop
1393 5c0c1eeb Iustin Pop
Live migration note: While failover-to-any is possible for all choices
1394 5c0c1eeb Iustin Pop
of S2, migration-to-any is possible only if we keep P1 as S2.
1395 5c0c1eeb Iustin Pop
1396 5c0c1eeb Iustin Pop
Caveats
1397 5c0c1eeb Iustin Pop
+++++++
1398 5c0c1eeb Iustin Pop
1399 5c0c1eeb Iustin Pop
The dynamic device model, while more complex, has an advantage: it
1400 6c2d0b44 Iustin Pop
will not reuse by mistake the DRBD device of another instance, since
1401 6c2d0b44 Iustin Pop
it always looks for either our own or a free one.
1402 5c0c1eeb Iustin Pop
1403 5c0c1eeb Iustin Pop
The static one, in contrast, will assume that given a minor number N,
1404 5c0c1eeb Iustin Pop
it's ours and we can take over. This needs careful implementation such
1405 5c0c1eeb Iustin Pop
that if the minor is in use, either we are able to cleanly shut it
1406 5c0c1eeb Iustin Pop
down, or we abort the startup. Otherwise, it could be that we start
1407 6c2d0b44 Iustin Pop
syncing between two instance's disks, causing data loss.
1408 5c0c1eeb Iustin Pop
1409 5c0c1eeb Iustin Pop
1410 5c0c1eeb Iustin Pop
Variable number of disk/NICs per instance
1411 5c0c1eeb Iustin Pop
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1412 5c0c1eeb Iustin Pop
1413 5c0c1eeb Iustin Pop
Variable number of disks
1414 5c0c1eeb Iustin Pop
++++++++++++++++++++++++
1415 5c0c1eeb Iustin Pop
1416 5c0c1eeb Iustin Pop
In order to support high-security scenarios (for example read-only sda
1417 5c0c1eeb Iustin Pop
and read-write sdb), we need to make a fully flexibly disk
1418 5c0c1eeb Iustin Pop
definition. This has less impact that it might look at first sight:
1419 6c2d0b44 Iustin Pop
only the instance creation has hard coded number of disks, not the disk
1420 5c0c1eeb Iustin Pop
handling code. The block device handling and most of the instance
1421 5c0c1eeb Iustin Pop
handling code is already working with "the instance's disks" as
1422 5c0c1eeb Iustin Pop
opposed to "the two disks of the instance", but some pieces are not
1423 5c0c1eeb Iustin Pop
(e.g. import/export) and the code needs a review to ensure safety.
1424 5c0c1eeb Iustin Pop
1425 5c0c1eeb Iustin Pop
The objective is to be able to specify the number of disks at
1426 5c0c1eeb Iustin Pop
instance creation, and to be able to toggle from read-only to
1427 6c2d0b44 Iustin Pop
read-write a disk afterward.
1428 5c0c1eeb Iustin Pop
1429 5c0c1eeb Iustin Pop
Variable number of NICs
1430 5c0c1eeb Iustin Pop
+++++++++++++++++++++++
1431 5c0c1eeb Iustin Pop
1432 5c0c1eeb Iustin Pop
Similar to the disk change, we need to allow multiple network
1433 5c0c1eeb Iustin Pop
interfaces per instance. This will affect the internal code (some
1434 5c0c1eeb Iustin Pop
function will have to stop assuming that ``instance.nics`` is a list
1435 6c2d0b44 Iustin Pop
of length one), the OS API which currently can export/import only one
1436 5c0c1eeb Iustin Pop
instance, and the command line interface.
1437 5c0c1eeb Iustin Pop
1438 5c0c1eeb Iustin Pop
Interface changes
1439 5c0c1eeb Iustin Pop
-----------------
1440 5c0c1eeb Iustin Pop
1441 5c0c1eeb Iustin Pop
There are two areas of interface changes: API-level changes (the OS
1442 5c0c1eeb Iustin Pop
interface and the RAPI interface) and the command line interface
1443 5c0c1eeb Iustin Pop
changes.
1444 5c0c1eeb Iustin Pop
1445 5c0c1eeb Iustin Pop
OS interface
1446 5c0c1eeb Iustin Pop
~~~~~~~~~~~~
1447 5c0c1eeb Iustin Pop
1448 7faf5110 Michael Hanselmann
The current Ganeti OS interface, version 5, is tailored for Ganeti 1.2.
1449 7faf5110 Michael Hanselmann
The interface is composed by a series of scripts which get called with
1450 7faf5110 Michael Hanselmann
certain parameters to perform OS-dependent operations on the cluster.
1451 7faf5110 Michael Hanselmann
The current scripts are:
1452 5c0c1eeb Iustin Pop
1453 5c0c1eeb Iustin Pop
create
1454 5c0c1eeb Iustin Pop
  called when a new instance is added to the cluster
1455 5c0c1eeb Iustin Pop
export
1456 5c0c1eeb Iustin Pop
  called to export an instance disk to a stream
1457 5c0c1eeb Iustin Pop
import
1458 5c0c1eeb Iustin Pop
  called to import from a stream to a new instance
1459 5c0c1eeb Iustin Pop
rename
1460 5c0c1eeb Iustin Pop
  called to perform the os-specific operations necessary for renaming an
1461 5c0c1eeb Iustin Pop
  instance
1462 5c0c1eeb Iustin Pop
1463 7faf5110 Michael Hanselmann
Currently these scripts suffer from the limitations of Ganeti 1.2: for
1464 7faf5110 Michael Hanselmann
example they accept exactly one block and one swap devices to operate
1465 7faf5110 Michael Hanselmann
on, rather than any amount of generic block devices, they blindly assume
1466 7faf5110 Michael Hanselmann
that an instance will have just one network interface to operate, they
1467 7faf5110 Michael Hanselmann
can not be configured to optimise the instance for a particular
1468 7faf5110 Michael Hanselmann
hypervisor.
1469 5c0c1eeb Iustin Pop
1470 7faf5110 Michael Hanselmann
Since in Ganeti 2.0 we want to support multiple hypervisors, and a
1471 7faf5110 Michael Hanselmann
non-fixed number of network and disks the OS interface need to change to
1472 7faf5110 Michael Hanselmann
transmit the appropriate amount of information about an instance to its
1473 7faf5110 Michael Hanselmann
managing operating system, when operating on it. Moreover since some old
1474 7faf5110 Michael Hanselmann
assumptions usually used in OS scripts are no longer valid we need to
1475 7faf5110 Michael Hanselmann
re-establish a common knowledge on what can be assumed and what cannot
1476 7faf5110 Michael Hanselmann
be regarding Ganeti environment.
1477 5c0c1eeb Iustin Pop
1478 5c0c1eeb Iustin Pop
1479 5c0c1eeb Iustin Pop
When designing the new OS API our priorities are:
1480 5c0c1eeb Iustin Pop
- ease of use
1481 5c0c1eeb Iustin Pop
- future extensibility
1482 6c2d0b44 Iustin Pop
- ease of porting from the old API
1483 5c0c1eeb Iustin Pop
- modularity
1484 5c0c1eeb Iustin Pop
1485 7faf5110 Michael Hanselmann
As such we want to limit the number of scripts that must be written to
1486 7faf5110 Michael Hanselmann
support an OS, and make it easy to share code between them by uniforming
1487 7faf5110 Michael Hanselmann
their input.  We also will leave the current script structure unchanged,
1488 7faf5110 Michael Hanselmann
as far as we can, and make a few of the scripts (import, export and
1489 7faf5110 Michael Hanselmann
rename) optional. Most information will be passed to the script through
1490 7faf5110 Michael Hanselmann
environment variables, for ease of access and at the same time ease of
1491 7faf5110 Michael Hanselmann
using only the information a script needs.
1492 5c0c1eeb Iustin Pop
1493 5c0c1eeb Iustin Pop
1494 5c0c1eeb Iustin Pop
The Scripts
1495 5c0c1eeb Iustin Pop
+++++++++++
1496 5c0c1eeb Iustin Pop
1497 7faf5110 Michael Hanselmann
As in Ganeti 1.2, every OS which wants to be installed in Ganeti needs
1498 7faf5110 Michael Hanselmann
to support the following functionality, through scripts:
1499 5c0c1eeb Iustin Pop
1500 5c0c1eeb Iustin Pop
create:
1501 7faf5110 Michael Hanselmann
  used to create a new instance running that OS. This script should
1502 7faf5110 Michael Hanselmann
  prepare the block devices, and install them so that the new OS can
1503 7faf5110 Michael Hanselmann
  boot under the specified hypervisor.
1504 5c0c1eeb Iustin Pop
export (optional):
1505 7faf5110 Michael Hanselmann
  used to export an installed instance using the given OS to a format
1506 7faf5110 Michael Hanselmann
  which can be used to import it back into a new instance.
1507 5c0c1eeb Iustin Pop
import (optional):
1508 7faf5110 Michael Hanselmann
  used to import an exported instance into a new one. This script is
1509 7faf5110 Michael Hanselmann
  similar to create, but the new instance should have the content of the
1510 7faf5110 Michael Hanselmann
  export, rather than contain a pristine installation.
1511 5c0c1eeb Iustin Pop
rename (optional):
1512 7faf5110 Michael Hanselmann
  used to perform the internal OS-specific operations needed to rename
1513 7faf5110 Michael Hanselmann
  an instance.
1514 5c0c1eeb Iustin Pop
1515 7faf5110 Michael Hanselmann
If any optional script is not implemented Ganeti will refuse to perform
1516 7faf5110 Michael Hanselmann
the given operation on instances using the non-implementing OS. Of
1517 7faf5110 Michael Hanselmann
course the create script is mandatory, and it doesn't make sense to
1518 7faf5110 Michael Hanselmann
support the either the export or the import operation but not both.
1519 5c0c1eeb Iustin Pop
1520 5c0c1eeb Iustin Pop
Incompatibilities with 1.2
1521 5c0c1eeb Iustin Pop
__________________________
1522 5c0c1eeb Iustin Pop
1523 7faf5110 Michael Hanselmann
We expect the following incompatibilities between the OS scripts for 1.2
1524 7faf5110 Michael Hanselmann
and the ones for 2.0:
1525 5c0c1eeb Iustin Pop
1526 7faf5110 Michael Hanselmann
- Input parameters: in 1.2 those were passed on the command line, in 2.0
1527 7faf5110 Michael Hanselmann
  we'll use environment variables, as there will be a lot more
1528 7faf5110 Michael Hanselmann
  information and not all OSes may care about all of it.
1529 7faf5110 Michael Hanselmann
- Number of calls: export scripts will be called once for each device
1530 7faf5110 Michael Hanselmann
  the instance has, and import scripts once for every exported disk.
1531 7faf5110 Michael Hanselmann
  Imported instances will be forced to have a number of disks greater or
1532 7faf5110 Michael Hanselmann
  equal to the one of the export.
1533 7faf5110 Michael Hanselmann
- Some scripts are not compulsory: if such a script is missing the
1534 7faf5110 Michael Hanselmann
  relevant operations will be forbidden for instances of that OS. This
1535 7faf5110 Michael Hanselmann
  makes it easier to distinguish between unsupported operations and
1536 7faf5110 Michael Hanselmann
  no-op ones (if any).
1537 5c0c1eeb Iustin Pop
1538 5c0c1eeb Iustin Pop
1539 5c0c1eeb Iustin Pop
Input
1540 5c0c1eeb Iustin Pop
_____
1541 5c0c1eeb Iustin Pop
1542 7faf5110 Michael Hanselmann
Rather than using command line flags, as they do now, scripts will
1543 7faf5110 Michael Hanselmann
accept inputs from environment variables. We expect the following input
1544 7faf5110 Michael Hanselmann
values:
1545 5c0c1eeb Iustin Pop
1546 5c0c1eeb Iustin Pop
OS_API_VERSION
1547 6c2d0b44 Iustin Pop
  The version of the OS API that the following parameters comply with;
1548 5c0c1eeb Iustin Pop
  this is used so that in the future we could have OSes supporting
1549 5c0c1eeb Iustin Pop
  multiple versions and thus Ganeti send the proper version in this
1550 5c0c1eeb Iustin Pop
  parameter
1551 5c0c1eeb Iustin Pop
INSTANCE_NAME
1552 5c0c1eeb Iustin Pop
  Name of the instance acted on
1553 5c0c1eeb Iustin Pop
HYPERVISOR
1554 7faf5110 Michael Hanselmann
  The hypervisor the instance should run on (e.g. 'xen-pvm', 'xen-hvm',
1555 7faf5110 Michael Hanselmann
  'kvm')
1556 5c0c1eeb Iustin Pop
DISK_COUNT
1557 5c0c1eeb Iustin Pop
  The number of disks this instance will have
1558 5c0c1eeb Iustin Pop
NIC_COUNT
1559 6c2d0b44 Iustin Pop
  The number of NICs this instance will have
1560 5c0c1eeb Iustin Pop
DISK_<N>_PATH
1561 5c0c1eeb Iustin Pop
  Path to the Nth disk.
1562 5c0c1eeb Iustin Pop
DISK_<N>_ACCESS
1563 5c0c1eeb Iustin Pop
  W if read/write, R if read only. OS scripts are not supposed to touch
1564 5c0c1eeb Iustin Pop
  read-only disks, but will be passed them to know.
1565 5c0c1eeb Iustin Pop
DISK_<N>_FRONTEND_TYPE
1566 7faf5110 Michael Hanselmann
  Type of the disk as seen by the instance. Can be 'scsi', 'ide',
1567 7faf5110 Michael Hanselmann
  'virtio'
1568 5c0c1eeb Iustin Pop
DISK_<N>_BACKEND_TYPE
1569 5c0c1eeb Iustin Pop
  Type of the disk as seen from the node. Can be 'block', 'file:loop' or
1570 5c0c1eeb Iustin Pop
  'file:blktap'
1571 5c0c1eeb Iustin Pop
NIC_<N>_MAC
1572 5c0c1eeb Iustin Pop
  Mac address for the Nth network interface
1573 5c0c1eeb Iustin Pop
NIC_<N>_IP
1574 5c0c1eeb Iustin Pop
  Ip address for the Nth network interface, if available
1575 5c0c1eeb Iustin Pop
NIC_<N>_BRIDGE
1576 5c0c1eeb Iustin Pop
  Node bridge the Nth network interface will be connected to
1577 5c0c1eeb Iustin Pop
NIC_<N>_FRONTEND_TYPE
1578 6c2d0b44 Iustin Pop
  Type of the Nth NIC as seen by the instance. For example 'virtio',
1579 6c2d0b44 Iustin Pop
  'rtl8139', etc.
1580 5c0c1eeb Iustin Pop
DEBUG_LEVEL
1581 7faf5110 Michael Hanselmann
  Whether more out should be produced, for debugging purposes. Currently
1582 7faf5110 Michael Hanselmann
  the only valid values are 0 and 1.
1583 5c0c1eeb Iustin Pop
1584 6c2d0b44 Iustin Pop
These are only the basic variables we are thinking of now, but more
1585 6c2d0b44 Iustin Pop
may come during the implementation and they will be documented in the
1586 fd07c6b3 Iustin Pop
:manpage:`ganeti-os-api` man page. All these variables will be
1587 fd07c6b3 Iustin Pop
available to all scripts.
1588 5c0c1eeb Iustin Pop
1589 5c0c1eeb Iustin Pop
Some scripts will need a few more information to work. These will have
1590 5c0c1eeb Iustin Pop
per-script variables, such as for example:
1591 5c0c1eeb Iustin Pop
1592 5c0c1eeb Iustin Pop
OLD_INSTANCE_NAME
1593 5c0c1eeb Iustin Pop
  rename: the name the instance should be renamed from.
1594 5c0c1eeb Iustin Pop
EXPORT_DEVICE
1595 7faf5110 Michael Hanselmann
  export: device to be exported, a snapshot of the actual device. The
1596 7faf5110 Michael Hanselmann
  data must be exported to stdout.
1597 5c0c1eeb Iustin Pop
EXPORT_INDEX
1598 5c0c1eeb Iustin Pop
  export: sequential number of the instance device targeted.
1599 5c0c1eeb Iustin Pop
IMPORT_DEVICE
1600 7faf5110 Michael Hanselmann
  import: device to send the data to, part of the new instance. The data
1601 7faf5110 Michael Hanselmann
  must be imported from stdin.
1602 5c0c1eeb Iustin Pop
IMPORT_INDEX
1603 5c0c1eeb Iustin Pop
  import: sequential number of the instance device targeted.
1604 5c0c1eeb Iustin Pop
1605 7faf5110 Michael Hanselmann
(Rationale for INSTANCE_NAME as an environment variable: the instance
1606 7faf5110 Michael Hanselmann
name is always needed and we could pass it on the command line. On the
1607 7faf5110 Michael Hanselmann
other hand, though, this would force scripts to both access the
1608 7faf5110 Michael Hanselmann
environment and parse the command line, so we'll move it for
1609 7faf5110 Michael Hanselmann
uniformity.)
1610 5c0c1eeb Iustin Pop
1611 5c0c1eeb Iustin Pop
1612 5c0c1eeb Iustin Pop
Output/Behaviour
1613 5c0c1eeb Iustin Pop
________________
1614 5c0c1eeb Iustin Pop
1615 7faf5110 Michael Hanselmann
As discussed scripts should only send user-targeted information to
1616 7faf5110 Michael Hanselmann
stderr. The create and import scripts are supposed to format/initialise
1617 7faf5110 Michael Hanselmann
the given block devices and install the correct instance data. The
1618 7faf5110 Michael Hanselmann
export script is supposed to export instance data to stdout in a format
1619 7faf5110 Michael Hanselmann
understandable by the the import script. The data will be compressed by
1620 7faf5110 Michael Hanselmann
Ganeti, so no compression should be done. The rename script should only
1621 7faf5110 Michael Hanselmann
modify the instance's knowledge of what its name is.
1622 5c0c1eeb Iustin Pop
1623 5c0c1eeb Iustin Pop
Other declarative style features
1624 5c0c1eeb Iustin Pop
++++++++++++++++++++++++++++++++
1625 5c0c1eeb Iustin Pop
1626 5c0c1eeb Iustin Pop
Similar to Ganeti 1.2, OS specifications will need to provide a
1627 6c2d0b44 Iustin Pop
'ganeti_api_version' containing list of numbers matching the
1628 6c2d0b44 Iustin Pop
version(s) of the API they implement. Ganeti itself will always be
1629 6c2d0b44 Iustin Pop
compatible with one version of the API and may maintain backwards
1630 6c2d0b44 Iustin Pop
compatibility if it's feasible to do so. The numbers are one-per-line,
1631 6c2d0b44 Iustin Pop
so an OS supporting both version 5 and version 20 will have a file
1632 6c2d0b44 Iustin Pop
containing two lines. This is different from Ganeti 1.2, which only
1633 6c2d0b44 Iustin Pop
supported one version number.
1634 5c0c1eeb Iustin Pop
1635 7faf5110 Michael Hanselmann
In addition to that an OS will be able to declare that it does support
1636 7faf5110 Michael Hanselmann
only a subset of the Ganeti hypervisors, by declaring them in the
1637 7faf5110 Michael Hanselmann
'hypervisors' file.
1638 5c0c1eeb Iustin Pop
1639 5c0c1eeb Iustin Pop
1640 5c0c1eeb Iustin Pop
Caveats/Notes
1641 5c0c1eeb Iustin Pop
+++++++++++++
1642 5c0c1eeb Iustin Pop
1643 7faf5110 Michael Hanselmann
We might want to have a "default" import/export behaviour that just
1644 7faf5110 Michael Hanselmann
dumps all disks and restores them. This can save work as most systems
1645 7faf5110 Michael Hanselmann
will just do this, while allowing flexibility for different systems.
1646 5c0c1eeb Iustin Pop
1647 7faf5110 Michael Hanselmann
Environment variables are limited in size, but we expect that there will
1648 7faf5110 Michael Hanselmann
be enough space to store the information we need. If we discover that
1649 7faf5110 Michael Hanselmann
this is not the case we may want to go to a more complex API such as
1650 7faf5110 Michael Hanselmann
storing those information on the filesystem and providing the OS script
1651 7faf5110 Michael Hanselmann
with the path to a file where they are encoded in some format.
1652 5c0c1eeb Iustin Pop
1653 5c0c1eeb Iustin Pop
1654 5c0c1eeb Iustin Pop
1655 5c0c1eeb Iustin Pop
Remote API changes
1656 5c0c1eeb Iustin Pop
~~~~~~~~~~~~~~~~~~
1657 5c0c1eeb Iustin Pop
1658 6c2d0b44 Iustin Pop
The first Ganeti remote API (RAPI) was designed and deployed with the
1659 6c2d0b44 Iustin Pop
Ganeti 1.2.5 release.  That version provide read-only access to the
1660 6c2d0b44 Iustin Pop
cluster state. Fully functional read-write API demands significant
1661 6c2d0b44 Iustin Pop
internal changes which will be implemented in version 2.0.
1662 5c0c1eeb Iustin Pop
1663 6c2d0b44 Iustin Pop
We decided to go with implementing the Ganeti RAPI in a RESTful way,
1664 6c2d0b44 Iustin Pop
which is aligned with key features we looking. It is simple,
1665 6c2d0b44 Iustin Pop
stateless, scalable and extensible paradigm of API implementation. As
1666 6c2d0b44 Iustin Pop
transport it uses HTTP over SSL, and we are implementing it with JSON
1667 6c2d0b44 Iustin Pop
encoding, but in a way it possible to extend and provide any other
1668 6c2d0b44 Iustin Pop
one.
1669 5c0c1eeb Iustin Pop
1670 5c0c1eeb Iustin Pop
Design
1671 5c0c1eeb Iustin Pop
++++++
1672 5c0c1eeb Iustin Pop
1673 6c2d0b44 Iustin Pop
The Ganeti RAPI is implemented as independent daemon, running on the
1674 6c2d0b44 Iustin Pop
same node with the same permission level as Ganeti master
1675 6c2d0b44 Iustin Pop
daemon. Communication is done through the LUXI library to the master
1676 6c2d0b44 Iustin Pop
daemon. In order to keep communication asynchronous RAPI processes two
1677 6c2d0b44 Iustin Pop
types of client requests:
1678 5c0c1eeb Iustin Pop
1679 6c2d0b44 Iustin Pop
- queries: server is able to answer immediately
1680 6c2d0b44 Iustin Pop
- job submission: some time is required for a useful response
1681 5c0c1eeb Iustin Pop
1682 6c2d0b44 Iustin Pop
In the query case requested data send back to client in the HTTP
1683 6c2d0b44 Iustin Pop
response body. Typical examples of queries would be: list of nodes,
1684 6c2d0b44 Iustin Pop
instances, cluster info, etc.
1685 5c0c1eeb Iustin Pop
1686 6c2d0b44 Iustin Pop
In the case of job submission, the client receive a job ID, the
1687 6c2d0b44 Iustin Pop
identifier which allows to query the job progress in the job queue
1688 6c2d0b44 Iustin Pop
(see `Job Queue`_).
1689 6c2d0b44 Iustin Pop
1690 6c2d0b44 Iustin Pop
Internally, each exported object has an version identifier, which is
1691 6c2d0b44 Iustin Pop
used as a state identifier in the HTTP header E-Tag field for
1692 6c2d0b44 Iustin Pop
requests/responses to avoid race conditions.
1693 5c0c1eeb Iustin Pop
1694 5c0c1eeb Iustin Pop
1695 5c0c1eeb Iustin Pop
Resource representation
1696 5c0c1eeb Iustin Pop
+++++++++++++++++++++++
1697 5c0c1eeb Iustin Pop
1698 6c2d0b44 Iustin Pop
The key difference of using REST instead of others API is that REST
1699 6c2d0b44 Iustin Pop
requires separation of services via resources with unique URIs. Each
1700 6c2d0b44 Iustin Pop
of them should have limited amount of state and support standard HTTP
1701 5c0c1eeb Iustin Pop
methods: GET, POST, DELETE, PUT.
1702 5c0c1eeb Iustin Pop
1703 6c2d0b44 Iustin Pop
For example in Ganeti's case we can have a set of URI:
1704 6c2d0b44 Iustin Pop
1705 6c2d0b44 Iustin Pop
 - ``/{clustername}/instances``
1706 6c2d0b44 Iustin Pop
 - ``/{clustername}/instances/{instancename}``
1707 6c2d0b44 Iustin Pop
 - ``/{clustername}/instances/{instancename}/tag``
1708 6c2d0b44 Iustin Pop
 - ``/{clustername}/tag``
1709 5c0c1eeb Iustin Pop
1710 6c2d0b44 Iustin Pop
A GET request to ``/{clustername}/instances`` will return the list of
1711 6c2d0b44 Iustin Pop
instances, a POST to ``/{clustername}/instances`` should create a new
1712 6c2d0b44 Iustin Pop
instance, a DELETE ``/{clustername}/instances/{instancename}`` should
1713 6c2d0b44 Iustin Pop
delete the instance, a GET ``/{clustername}/tag`` should return get
1714 6c2d0b44 Iustin Pop
cluster tags.
1715 5c0c1eeb Iustin Pop
1716 6c2d0b44 Iustin Pop
Each resource URI will have a version prefix. The resource IDs are to
1717 6c2d0b44 Iustin Pop
be determined.
1718 5c0c1eeb Iustin Pop
1719 6c2d0b44 Iustin Pop
Internal encoding might be JSON, XML, or any other. The JSON encoding
1720 6c2d0b44 Iustin Pop
fits nicely in Ganeti RAPI needs. The client can request a specific
1721 6c2d0b44 Iustin Pop
representation via the Accept field in the HTTP header.
1722 5c0c1eeb Iustin Pop
1723 6c2d0b44 Iustin Pop
REST uses HTTP as its transport and application protocol for resource
1724 6c2d0b44 Iustin Pop
access. The set of possible responses is a subset of standard HTTP
1725 6c2d0b44 Iustin Pop
responses.
1726 6c2d0b44 Iustin Pop
1727 6c2d0b44 Iustin Pop
The statelessness model provides additional reliability and
1728 6c2d0b44 Iustin Pop
transparency to operations (e.g. only one request needs to be analyzed
1729 6c2d0b44 Iustin Pop
to understand the in-progress operation, not a sequence of multiple
1730 6c2d0b44 Iustin Pop
requests/responses).
1731 5c0c1eeb Iustin Pop
1732 5c0c1eeb Iustin Pop
1733 5c0c1eeb Iustin Pop
Security
1734 5c0c1eeb Iustin Pop
++++++++
1735 5c0c1eeb Iustin Pop
1736 6c2d0b44 Iustin Pop
With the write functionality security becomes a much bigger an issue.
1737 6c2d0b44 Iustin Pop
The Ganeti RAPI uses basic HTTP authentication on top of an
1738 6c2d0b44 Iustin Pop
SSL-secured connection to grant access to an exported resource. The
1739 6c2d0b44 Iustin Pop
password is stored locally in an Apache-style ``.htpasswd`` file. Only
1740 6c2d0b44 Iustin Pop
one level of privileges is supported.
1741 6c2d0b44 Iustin Pop
1742 6c2d0b44 Iustin Pop
Caveats
1743 6c2d0b44 Iustin Pop
+++++++
1744 6c2d0b44 Iustin Pop
1745 6c2d0b44 Iustin Pop
The model detailed above for job submission requires the client to
1746 6c2d0b44 Iustin Pop
poll periodically for updates to the job; an alternative would be to
1747 6c2d0b44 Iustin Pop
allow the client to request a callback, or a 'wait for updates' call.
1748 6c2d0b44 Iustin Pop
1749 6c2d0b44 Iustin Pop
The callback model was not considered due to the following two issues:
1750 5c0c1eeb Iustin Pop
1751 6c2d0b44 Iustin Pop
- callbacks would require a new model of allowed callback URLs,
1752 6c2d0b44 Iustin Pop
  together with a method of managing these
1753 6c2d0b44 Iustin Pop
- callbacks only work when the client and the master are in the same
1754 6c2d0b44 Iustin Pop
  security domain, and they fail in the other cases (e.g. when there is
1755 6c2d0b44 Iustin Pop
  a firewall between the client and the RAPI daemon that only allows
1756 6c2d0b44 Iustin Pop
  client-to-RAPI calls, which is usual in DMZ cases)
1757 6c2d0b44 Iustin Pop
1758 6c2d0b44 Iustin Pop
The 'wait for updates' method is not suited to the HTTP protocol,
1759 6c2d0b44 Iustin Pop
where requests are supposed to be short-lived.
1760 5c0c1eeb Iustin Pop
1761 5c0c1eeb Iustin Pop
Command line changes
1762 5c0c1eeb Iustin Pop
~~~~~~~~~~~~~~~~~~~~
1763 5c0c1eeb Iustin Pop
1764 5c0c1eeb Iustin Pop
Ganeti 2.0 introduces several new features as well as new ways to
1765 5c0c1eeb Iustin Pop
handle instance resources like disks or network interfaces. This
1766 6c2d0b44 Iustin Pop
requires some noticeable changes in the way command line arguments are
1767 5c0c1eeb Iustin Pop
handled.
1768 5c0c1eeb Iustin Pop
1769 6c2d0b44 Iustin Pop
- extend and modify command line syntax to support new features
1770 6c2d0b44 Iustin Pop
- ensure consistent patterns in command line arguments to reduce
1771 6c2d0b44 Iustin Pop
  cognitive load
1772 5c0c1eeb Iustin Pop
1773 5c0c1eeb Iustin Pop
The design changes that require these changes are, in no particular
1774 5c0c1eeb Iustin Pop
order:
1775 5c0c1eeb Iustin Pop
1776 5c0c1eeb Iustin Pop
- flexible instance disk handling: support a variable number of disks
1777 5c0c1eeb Iustin Pop
  with varying properties per instance,
1778 5c0c1eeb Iustin Pop
- flexible instance network interface handling: support a variable
1779 5c0c1eeb Iustin Pop
  number of network interfaces with varying properties per instance
1780 5c0c1eeb Iustin Pop
- multiple hypervisors: multiple hypervisors can be active on the same
1781 5c0c1eeb Iustin Pop
  cluster, each supporting different parameters,
1782 5c0c1eeb Iustin Pop
- support for device type CDROM (via ISO image)
1783 5c0c1eeb Iustin Pop
1784 6c2d0b44 Iustin Pop
As such, there are several areas of Ganeti where the command line
1785 5c0c1eeb Iustin Pop
arguments will change:
1786 5c0c1eeb Iustin Pop
1787 5c0c1eeb Iustin Pop
- Cluster configuration
1788 5c0c1eeb Iustin Pop
1789 5c0c1eeb Iustin Pop
  - cluster initialization
1790 5c0c1eeb Iustin Pop
  - cluster default configuration
1791 5c0c1eeb Iustin Pop
1792 5c0c1eeb Iustin Pop
- Instance configuration
1793 5c0c1eeb Iustin Pop
1794 5c0c1eeb Iustin Pop
  - handling of network cards for instances,
1795 5c0c1eeb Iustin Pop
  - handling of disks for instances,
1796 5c0c1eeb Iustin Pop
  - handling of CDROM devices and
1797 5c0c1eeb Iustin Pop
  - handling of hypervisor specific options.
1798 5c0c1eeb Iustin Pop
1799 6c2d0b44 Iustin Pop
There are several areas of Ganeti where the command line arguments
1800 6c2d0b44 Iustin Pop
will change:
1801 5c0c1eeb Iustin Pop
1802 5c0c1eeb Iustin Pop
- Cluster configuration
1803 5c0c1eeb Iustin Pop
1804 5c0c1eeb Iustin Pop
  - cluster initialization
1805 5c0c1eeb Iustin Pop
  - cluster default configuration
1806 5c0c1eeb Iustin Pop
1807 5c0c1eeb Iustin Pop
- Instance configuration
1808 5c0c1eeb Iustin Pop
1809 5c0c1eeb Iustin Pop
  - handling of network cards for instances,
1810 5c0c1eeb Iustin Pop
  - handling of disks for instances,
1811 5c0c1eeb Iustin Pop
  - handling of CDROM devices and
1812 5c0c1eeb Iustin Pop
  - handling of hypervisor specific options.
1813 5c0c1eeb Iustin Pop
1814 5c0c1eeb Iustin Pop
Notes about device removal/addition
1815 5c0c1eeb Iustin Pop
+++++++++++++++++++++++++++++++++++
1816 5c0c1eeb Iustin Pop
1817 5c0c1eeb Iustin Pop
To avoid problems with device location changes (e.g. second network
1818 5c0c1eeb Iustin Pop
interface of the instance becoming the first or third and the like)
1819 5c0c1eeb Iustin Pop
the list of network/disk devices is treated as a stack, i.e. devices
1820 5c0c1eeb Iustin Pop
can only be added/removed at the end of the list of devices of each
1821 5c0c1eeb Iustin Pop
class (disk or network) for each instance.
1822 5c0c1eeb Iustin Pop
1823 5c0c1eeb Iustin Pop
gnt-instance commands
1824 5c0c1eeb Iustin Pop
+++++++++++++++++++++
1825 5c0c1eeb Iustin Pop
1826 5c0c1eeb Iustin Pop
The commands for gnt-instance will be modified and extended to allow
1827 5c0c1eeb Iustin Pop
for the new functionality:
1828 5c0c1eeb Iustin Pop
1829 5c0c1eeb Iustin Pop
- the add command will be extended to support the new device and
1830 5c0c1eeb Iustin Pop
  hypervisor options,
1831 5c0c1eeb Iustin Pop
- the modify command continues to handle all modifications to
1832 5c0c1eeb Iustin Pop
  instances, but will be extended with new arguments for handling
1833 5c0c1eeb Iustin Pop
  devices.
1834 5c0c1eeb Iustin Pop
1835 5c0c1eeb Iustin Pop
Network Device Options
1836 5c0c1eeb Iustin Pop
++++++++++++++++++++++
1837 5c0c1eeb Iustin Pop
1838 5c0c1eeb Iustin Pop
The generic format of the network device option is:
1839 5c0c1eeb Iustin Pop
1840 5c0c1eeb Iustin Pop
  --net $DEVNUM[:$OPTION=$VALUE][,$OPTION=VALUE]
1841 5c0c1eeb Iustin Pop
1842 5c0c1eeb Iustin Pop
:$DEVNUM: device number, unsigned integer, starting at 0,
1843 5c0c1eeb Iustin Pop
:$OPTION: device option, string,
1844 5c0c1eeb Iustin Pop
:$VALUE: device option value, string.
1845 5c0c1eeb Iustin Pop
1846 5c0c1eeb Iustin Pop
Currently, the following device options will be defined (open to
1847 5c0c1eeb Iustin Pop
further changes):
1848 5c0c1eeb Iustin Pop
1849 5c0c1eeb Iustin Pop
:mac: MAC address of the network interface, accepts either a valid
1850 5c0c1eeb Iustin Pop
  MAC address or the string 'auto'. If 'auto' is specified, a new MAC
1851 5c0c1eeb Iustin Pop
  address will be generated randomly. If the mac device option is not
1852 5c0c1eeb Iustin Pop
  specified, the default value 'auto' is assumed.
1853 5c0c1eeb Iustin Pop
:bridge: network bridge the network interface is connected
1854 5c0c1eeb Iustin Pop
  to. Accepts either a valid bridge name (the specified bridge must
1855 5c0c1eeb Iustin Pop
  exist on the node(s)) as string or the string 'auto'. If 'auto' is
1856 5c0c1eeb Iustin Pop
  specified, the default brigde is used. If the bridge option is not
1857 5c0c1eeb Iustin Pop
  specified, the default value 'auto' is assumed.
1858 5c0c1eeb Iustin Pop
1859 5c0c1eeb Iustin Pop
Disk Device Options
1860 5c0c1eeb Iustin Pop
+++++++++++++++++++
1861 5c0c1eeb Iustin Pop
1862 5c0c1eeb Iustin Pop
The generic format of the disk device option is:
1863 5c0c1eeb Iustin Pop
1864 5c0c1eeb Iustin Pop
  --disk $DEVNUM[:$OPTION=$VALUE][,$OPTION=VALUE]
1865 5c0c1eeb Iustin Pop
1866 5c0c1eeb Iustin Pop
:$DEVNUM: device number, unsigned integer, starting at 0,
1867 5c0c1eeb Iustin Pop
:$OPTION: device option, string,
1868 5c0c1eeb Iustin Pop
:$VALUE: device option value, string.
1869 5c0c1eeb Iustin Pop
1870 5c0c1eeb Iustin Pop
Currently, the following device options will be defined (open to
1871 5c0c1eeb Iustin Pop
further changes):
1872 5c0c1eeb Iustin Pop
1873 5c0c1eeb Iustin Pop
:size: size of the disk device, either a positive number, specifying
1874 5c0c1eeb Iustin Pop
  the disk size in mebibytes, or a number followed by a magnitude suffix
1875 5c0c1eeb Iustin Pop
  (M for mebibytes, G for gibibytes). Also accepts the string 'auto' in
1876 5c0c1eeb Iustin Pop
  which case the default disk size will be used. If the size option is
1877 5c0c1eeb Iustin Pop
  not specified, 'auto' is assumed. This option is not valid for all
1878 5c0c1eeb Iustin Pop
  disk layout types.
1879 5c0c1eeb Iustin Pop
:access: access mode of the disk device, a single letter, valid values
1880 5c0c1eeb Iustin Pop
  are:
1881 5c0c1eeb Iustin Pop
1882 5c0c1eeb Iustin Pop
  - *w*: read/write access to the disk device or
1883 5c0c1eeb Iustin Pop
  - *r*: read-only access to the disk device.
1884 5c0c1eeb Iustin Pop
1885 5c0c1eeb Iustin Pop
  If the access mode is not specified, the default mode of read/write
1886 5c0c1eeb Iustin Pop
  access will be configured.
1887 5c0c1eeb Iustin Pop
:path: path to the image file for the disk device, string. No default
1888 5c0c1eeb Iustin Pop
  exists. This option is not valid for all disk layout types.
1889 5c0c1eeb Iustin Pop
1890 5c0c1eeb Iustin Pop
Adding devices
1891 5c0c1eeb Iustin Pop
++++++++++++++
1892 5c0c1eeb Iustin Pop
1893 5c0c1eeb Iustin Pop
To add devices to an already existing instance, use the device type
1894 5c0c1eeb Iustin Pop
specific option to gnt-instance modify. Currently, there are two
1895 5c0c1eeb Iustin Pop
device type specific options supported:
1896 5c0c1eeb Iustin Pop
1897 5c0c1eeb Iustin Pop
:--net: for network interface cards
1898 5c0c1eeb Iustin Pop
:--disk: for disk devices
1899 5c0c1eeb Iustin Pop
1900 6c2d0b44 Iustin Pop
The syntax to the device specific options is similar to the generic
1901 5c0c1eeb Iustin Pop
device options, but instead of specifying a device number like for
1902 5c0c1eeb Iustin Pop
gnt-instance add, you specify the magic string add. The new device
1903 5c0c1eeb Iustin Pop
will always be appended at the end of the list of devices of this type
1904 5c0c1eeb Iustin Pop
for the specified instance, e.g. if the instance has disk devices 0,1
1905 5c0c1eeb Iustin Pop
and 2, the newly added disk device will be disk device 3.
1906 5c0c1eeb Iustin Pop
1907 5c0c1eeb Iustin Pop
Example: gnt-instance modify --net add:mac=auto test-instance
1908 5c0c1eeb Iustin Pop
1909 5c0c1eeb Iustin Pop
Removing devices
1910 5c0c1eeb Iustin Pop
++++++++++++++++
1911 5c0c1eeb Iustin Pop
1912 5c0c1eeb Iustin Pop
Removing devices from and instance is done via gnt-instance
1913 5c0c1eeb Iustin Pop
modify. The same device specific options as for adding instances are
1914 5c0c1eeb Iustin Pop
used. Instead of a device number and further device options, only the
1915 5c0c1eeb Iustin Pop
magic string remove is specified. It will always remove the last
1916 5c0c1eeb Iustin Pop
device in the list of devices of this type for the instance specified,
1917 5c0c1eeb Iustin Pop
e.g. if the instance has disk devices 0, 1, 2 and 3, the disk device
1918 5c0c1eeb Iustin Pop
number 3 will be removed.
1919 5c0c1eeb Iustin Pop
1920 5c0c1eeb Iustin Pop
Example: gnt-instance modify --net remove test-instance
1921 5c0c1eeb Iustin Pop
1922 5c0c1eeb Iustin Pop
Modifying devices
1923 5c0c1eeb Iustin Pop
+++++++++++++++++
1924 5c0c1eeb Iustin Pop
1925 5c0c1eeb Iustin Pop
Modifying devices is also done with device type specific options to
1926 5c0c1eeb Iustin Pop
the gnt-instance modify command. There are currently two device type
1927 5c0c1eeb Iustin Pop
options supported:
1928 5c0c1eeb Iustin Pop
1929 5c0c1eeb Iustin Pop
:--net: for network interface cards
1930 5c0c1eeb Iustin Pop
:--disk: for disk devices
1931 5c0c1eeb Iustin Pop
1932 6c2d0b44 Iustin Pop
The syntax to the device specific options is similar to the generic
1933 5c0c1eeb Iustin Pop
device options. The device number you specify identifies the device to
1934 5c0c1eeb Iustin Pop
be modified.
1935 5c0c1eeb Iustin Pop
1936 6c2d0b44 Iustin Pop
Example::
1937 6c2d0b44 Iustin Pop
1938 6c2d0b44 Iustin Pop
  gnt-instance modify --disk 2:access=r
1939 5c0c1eeb Iustin Pop
1940 5c0c1eeb Iustin Pop
Hypervisor Options
1941 5c0c1eeb Iustin Pop
++++++++++++++++++
1942 5c0c1eeb Iustin Pop
1943 5c0c1eeb Iustin Pop
Ganeti 2.0 will support more than one hypervisor. Different
1944 5c0c1eeb Iustin Pop
hypervisors have various options that only apply to a specific
1945 5c0c1eeb Iustin Pop
hypervisor. Those hypervisor specific options are treated specially
1946 6c2d0b44 Iustin Pop
via the ``--hypervisor`` option. The generic syntax of the hypervisor
1947 6c2d0b44 Iustin Pop
option is as follows::
1948 5c0c1eeb Iustin Pop
1949 5c0c1eeb Iustin Pop
  --hypervisor $HYPERVISOR:$OPTION=$VALUE[,$OPTION=$VALUE]
1950 5c0c1eeb Iustin Pop
1951 5c0c1eeb Iustin Pop
:$HYPERVISOR: symbolic name of the hypervisor to use, string,
1952 5c0c1eeb Iustin Pop
  has to match the supported hypervisors. Example: xen-pvm
1953 5c0c1eeb Iustin Pop
1954 5c0c1eeb Iustin Pop
:$OPTION: hypervisor option name, string
1955 5c0c1eeb Iustin Pop
:$VALUE: hypervisor option value, string
1956 5c0c1eeb Iustin Pop
1957 5c0c1eeb Iustin Pop
The hypervisor option for an instance can be set on instance creation
1958 6c2d0b44 Iustin Pop
time via the ``gnt-instance add`` command. If the hypervisor for an
1959 5c0c1eeb Iustin Pop
instance is not specified upon instance creation, the default
1960 5c0c1eeb Iustin Pop
hypervisor will be used.
1961 5c0c1eeb Iustin Pop
1962 5c0c1eeb Iustin Pop
Modifying hypervisor parameters
1963 5c0c1eeb Iustin Pop
+++++++++++++++++++++++++++++++
1964 5c0c1eeb Iustin Pop
1965 5c0c1eeb Iustin Pop
The hypervisor parameters of an existing instance can be modified
1966 6c2d0b44 Iustin Pop
using ``--hypervisor`` option of the ``gnt-instance modify``
1967 6c2d0b44 Iustin Pop
command. However, the hypervisor type of an existing instance can not
1968 6c2d0b44 Iustin Pop
be changed, only the particular hypervisor specific option can be
1969 6c2d0b44 Iustin Pop
changed. Therefore, the format of the option parameters has been
1970 6c2d0b44 Iustin Pop
simplified to omit the hypervisor name and only contain the comma
1971 6c2d0b44 Iustin Pop
separated list of option-value pairs.
1972 5c0c1eeb Iustin Pop
1973 6c2d0b44 Iustin Pop
Example::
1974 6c2d0b44 Iustin Pop
1975 6c2d0b44 Iustin Pop
  gnt-instance modify --hypervisor cdrom=/srv/boot.iso,boot_order=cdrom:network test-instance
1976 5c0c1eeb Iustin Pop
1977 5c0c1eeb Iustin Pop
gnt-cluster commands
1978 5c0c1eeb Iustin Pop
++++++++++++++++++++
1979 5c0c1eeb Iustin Pop
1980 5c0c1eeb Iustin Pop
The command for gnt-cluster will be extended to allow setting and
1981 5c0c1eeb Iustin Pop
changing the default parameters of the cluster:
1982 5c0c1eeb Iustin Pop
1983 5c0c1eeb Iustin Pop
- The init command will be extend to support the defaults option to
1984 5c0c1eeb Iustin Pop
  set the cluster defaults upon cluster initialization.
1985 5c0c1eeb Iustin Pop
- The modify command will be added to modify the cluster
1986 5c0c1eeb Iustin Pop
  parameters. It will support the --defaults option to change the
1987 5c0c1eeb Iustin Pop
  cluster defaults.
1988 5c0c1eeb Iustin Pop
1989 5c0c1eeb Iustin Pop
Cluster defaults
1990 5c0c1eeb Iustin Pop
1991 5c0c1eeb Iustin Pop
The generic format of the cluster default setting option is:
1992 5c0c1eeb Iustin Pop
1993 5c0c1eeb Iustin Pop
  --defaults $OPTION=$VALUE[,$OPTION=$VALUE]
1994 5c0c1eeb Iustin Pop
1995 5c0c1eeb Iustin Pop
:$OPTION: cluster default option, string,
1996 5c0c1eeb Iustin Pop
:$VALUE: cluster default option value, string.
1997 5c0c1eeb Iustin Pop
1998 5c0c1eeb Iustin Pop
Currently, the following cluster default options are defined (open to
1999 5c0c1eeb Iustin Pop
further changes):
2000 5c0c1eeb Iustin Pop
2001 5c0c1eeb Iustin Pop
:hypervisor: the default hypervisor to use for new instances,
2002 5c0c1eeb Iustin Pop
  string. Must be a valid hypervisor known to and supported by the
2003 5c0c1eeb Iustin Pop
  cluster.
2004 5c0c1eeb Iustin Pop
:disksize: the disksize for newly created instance disks, where
2005 5c0c1eeb Iustin Pop
  applicable. Must be either a positive number, in which case the unit
2006 5c0c1eeb Iustin Pop
  of megabyte is assumed, or a positive number followed by a supported
2007 5c0c1eeb Iustin Pop
  magnitude symbol (M for megabyte or G for gigabyte).
2008 5c0c1eeb Iustin Pop
:bridge: the default network bridge to use for newly created instance
2009 5c0c1eeb Iustin Pop
  network interfaces, string. Must be a valid bridge name of a bridge
2010 5c0c1eeb Iustin Pop
  existing on the node(s).
2011 5c0c1eeb Iustin Pop
2012 5c0c1eeb Iustin Pop
Hypervisor cluster defaults
2013 5c0c1eeb Iustin Pop
+++++++++++++++++++++++++++
2014 5c0c1eeb Iustin Pop
2015 6c2d0b44 Iustin Pop
The generic format of the hypervisor cluster wide default setting
2016 6c2d0b44 Iustin Pop
option is::
2017 5c0c1eeb Iustin Pop
2018 5c0c1eeb Iustin Pop
  --hypervisor-defaults $HYPERVISOR:$OPTION=$VALUE[,$OPTION=$VALUE]
2019 5c0c1eeb Iustin Pop
2020 5c0c1eeb Iustin Pop
:$HYPERVISOR: symbolic name of the hypervisor whose defaults you want
2021 5c0c1eeb Iustin Pop
  to set, string
2022 5c0c1eeb Iustin Pop
:$OPTION: cluster default option, string,
2023 5c0c1eeb Iustin Pop
:$VALUE: cluster default option value, string.
2024 558fd122 Michael Hanselmann
2025 558fd122 Michael Hanselmann
.. vim: set textwidth=72 :