5 This document describes the major changes in Ganeti 2.0 compared to
8 The 2.0 version will constitute a rewrite of the 'core' architecture,
9 paving the way for additional features in future 2.x versions.
11 .. contents:: :depth: 3
16 Ganeti 1.2 has many scalability issues and restrictions due to its
17 roots as software for managing small and 'static' clusters.
19 Version 2.0 will attempt to remedy first the scalability issues and
20 then the restrictions.
25 While Ganeti 1.2 is usable, it severely limits the flexibility of the
26 cluster administration and imposes a very rigid model. It has the
27 following main scalability issues:
29 - only one operation at a time on the cluster [#]_
30 - poor handling of node failures in the cluster
31 - mixing hypervisors in a cluster not allowed
33 It also has a number of artificial restrictions, due to historical
36 - fixed number of disks (two) per instance
37 - fixed number of NICs
39 .. [#] Replace disks will release the lock, but this is an exception
40 and not a recommended way to operate
42 The 2.0 version is intended to address some of these problems, and
43 create a more flexible code base for future developments.
45 Among these problems, the single-operation at a time restriction is
46 biggest issue with the current version of Ganeti. It is such a big
47 impediment in operating bigger clusters that many times one is tempted
48 to remove the lock just to do a simple operation like start instance
49 while an OS installation is running.
54 Ganeti 1.2 has a single global lock, which is used for all cluster
55 operations. This has been painful at various times, for example:
57 - It is impossible for two people to efficiently interact with a cluster
58 (for example for debugging) at the same time.
59 - When batch jobs are running it's impossible to do other work (for
60 example failovers/fixes) on a cluster.
62 This poses scalability problems: as clusters grow in node and instance
63 size it's a lot more likely that operations which one could conceive
64 should run in parallel (for example because they happen on different
65 nodes) are actually stalling each other while waiting for the global
66 lock, without a real reason for that to happen.
68 One of the main causes of this global lock (beside the higher
69 difficulty of ensuring data consistency in a more granular lock model)
70 is the fact that currently there is no long-lived process in Ganeti
71 that can coordinate multiple operations. Each command tries to acquire
72 the so called *cmd* lock and when it succeeds, it takes complete
73 ownership of the cluster configuration and state.
75 Other scalability problems are due the design of the DRBD device
76 model, which assumed at its creation a low (one to four) number of
77 instances per node, which is no longer true with today's hardware.
79 Artificial restrictions
80 -----------------------
82 Ganeti 1.2 (and previous versions) have a fixed two-disks, one-NIC per
83 instance model. This is a purely artificial restrictions, but it
84 touches multiple areas (configuration, import/export, command line)
85 that it's more fitted to a major release than a minor one.
90 The fact that each command is a separate process that reads the
91 cluster state, executes the command, and saves the new state is also
92 an issue on big clusters where the configuration data for the cluster
93 begins to be non-trivial in size.
98 In order to solve the scalability problems, a rewrite of the core
99 design of Ganeti is required. While the cluster operations themselves
100 won't change (e.g. start instance will do the same things, the way
101 these operations are scheduled internally will change radically.
103 The new design will change the cluster architecture to:
105 .. image:: arch-2.0.png
107 This differs from the 1.2 architecture by the addition of the master
108 daemon, which will be the only entity to talk to the node daemons.
114 The changes for 2.0 can be split into roughly three areas:
116 - core changes that affect the design of the software
117 - features (or restriction removals) but which do not have a wide
119 - user-level and API-level changes which translate into differences for
120 the operation of the cluster
125 The main changes will be switching from a per-process model to a
126 daemon based model, where the individual gnt-* commands will be
127 clients that talk to this daemon (see `Master daemon`_). This will
128 allow us to get rid of the global cluster lock for most operations,
129 having instead a per-object lock (see `Granular locking`_). Also, the
130 daemon will be able to queue jobs, and this will allow the individual
131 clients to submit jobs without waiting for them to finish, and also
132 see the result of old requests (see `Job Queue`_).
134 Beside these major changes, another 'core' change but that will not be
135 as visible to the users will be changing the model of object attribute
136 storage, and separate that into name spaces (such that an Xen PVM
137 instance will not have the Xen HVM parameters). This will allow future
138 flexibility in defining additional parameters. For more details see
139 `Object parameters`_.
141 The various changes brought in by the master daemon model and the
142 read-write RAPI will require changes to the cluster security; we move
143 away from Twisted and use HTTP(s) for intra- and extra-cluster
144 communications. For more details, see the security document in the
150 In Ganeti 2.0, we will have the following *entities*:
152 - the master daemon (on the master node)
153 - the node daemon (on all nodes)
154 - the command line tools (on the master node)
155 - the RAPI daemon (on the master node)
157 The master-daemon related interaction paths are:
159 - (CLI tools/RAPI daemon) and the master daemon, via the so called
161 - the master daemon and the node daemons, via the node RPC
163 There are also some additional interaction paths for exceptional cases:
165 - CLI tools might access via SSH the nodes (for ``gnt-cluster copyfile``
166 and ``gnt-cluster command``)
167 - master failover is a special case when a non-master node will SSH
168 and do node-RPC calls to the current master
170 The protocol between the master daemon and the node daemons will be
171 changed from (Ganeti 1.2) Twisted PB (perspective broker) to HTTP(S),
172 using a simple PUT/GET of JSON-encoded messages. This is done due to
173 difficulties in working with the Twisted framework and its protocols
174 in a multithreaded environment, which we can overcome by using a
175 simpler stack (see the caveats section).
177 The protocol between the CLI/RAPI and the master daemon will be a
178 custom one (called *LUXI*): on a UNIX socket on the master node, with
179 rights restricted by filesystem permissions, the CLI/RAPI will talk to
180 the master daemon using JSON-encoded messages.
182 The operations supported over this internal protocol will be encoded
183 via a python library that will expose a simple API for its
184 users. Internally, the protocol will simply encode all objects in JSON
185 format and decode them on the receiver side.
187 For more details about the RAPI daemon see `Remote API changes`_, and
188 for the node daemon see `Node daemon changes`_.
195 As described above, the protocol for making requests or queries to the
196 master daemon will be a UNIX-socket based simple RPC of JSON-encoded
199 The choice of UNIX was in order to get rid of the need of
200 authentication and authorisation inside Ganeti; for 2.0, the
201 permissions on the Unix socket itself will determine the access
204 We will have two main classes of operations over this API:
206 - cluster query functions
207 - job related functions
209 The cluster query functions are usually short-duration, and are the
210 equivalent of the ``OP_QUERY_*`` opcodes in Ganeti 1.2 (and they are
211 internally implemented still with these opcodes). The clients are
212 guaranteed to receive the response in a reasonable time via a timeout.
214 The job-related functions will be:
217 - query job (which could also be categorized in the query-functions)
218 - archive job (see the job queue design doc)
219 - wait for job change, which allows a client to wait without polling
221 For more details of the actual operation list, see the `Job Queue`_.
223 Both requests and responses will consist of a JSON-encoded message
224 followed by the ``ETX`` character (ASCII decimal 3), which is not a
225 valid character in JSON messages and thus can serve as a message
226 delimiter. The contents of the messages will be a dictionary with two
230 the name of the method called
232 the arguments to the method, as a list (no keyword arguments allowed)
234 Responses will follow the same format, with the two fields being:
237 a boolean denoting the success of the operation
239 the actual result, or error message in case of failure
241 There are two special value for the result field:
243 - in the case that the operation failed, and this field is a list of
244 length two, the client library will try to interpret is as an
245 exception, the first element being the exception type and the second
246 one the actual exception arguments; this will allow a simple method of
247 passing Ganeti-related exception across the interface
248 - for the *WaitForChange* call (that waits on the server for a job to
249 change status), if the result is equal to ``nochange`` instead of the
250 usual result for this call (a list of changes), then the library will
251 internally retry the call; this is done in order to differentiate
252 internally between master daemon hung and job simply not changed
254 Users of the API that don't use the provided python library should
255 take care of the above two cases.
258 Master daemon implementation
259 ++++++++++++++++++++++++++++
261 The daemon will be based around a main I/O thread that will wait for
262 new requests from the clients, and that does the setup/shutdown of the
263 other thread (pools).
265 There will two other classes of threads in the daemon:
267 - job processing threads, part of a thread pool, and which are
268 long-lived, started at daemon startup and terminated only at shutdown
270 - client I/O threads, which are the ones that talk the local protocol
271 (LUXI) to the clients, and are short-lived
273 Master startup/failover
274 +++++++++++++++++++++++
276 In Ganeti 1.x there is no protection against failing over the master
277 to a node with stale configuration. In effect, the responsibility of
278 correct failovers falls on the admin. This is true both for the new
279 master and for when an old, offline master startup.
281 Since in 2.x we are extending the cluster state to cover the job queue
282 and have a daemon that will execute by itself the job queue, we want
283 to have more resilience for the master role.
285 The following algorithm will happen whenever a node is ready to
286 transition to the master role, either at startup time or at node
289 #. read the configuration file and parse the node list
292 #. query all the nodes and make sure we obtain an agreement via
293 a quorum of at least half plus one nodes for the following:
295 - we have the latest configuration and job list (as
296 determined by the serial number on the configuration and
297 highest job ID on the job queue)
299 - if we are not failing over (but just starting), the
300 quorum agrees that we are the designated master
302 - if any of the above is false, we prevent the current operation
303 (i.e. we don't become the master)
305 #. at this point, the node transitions to the master role
307 #. for all the in-progress jobs, mark them as failed, with
308 reason unknown or something similar (master failed, etc.)
310 Since due to exceptional conditions we could have a situation in which
311 no node can become the master due to inconsistent data, we will have
312 an override switch for the master daemon startup that will assume the
313 current node has the right data and will replicate all the
314 configuration files to the other nodes.
316 **Note**: the above algorithm is by no means an election algorithm; it
317 is a *confirmation* of the master role currently held by a node.
322 The logging system will be switched completely to the standard python
323 logging module; currently it's logging-based, but exposes a different
324 API, which is just overhead. As such, the code will be switched over
325 to standard logging calls, and only the setup will be custom.
327 With this change, we will remove the separate debug/info/error logs,
328 and instead have always one logfile per daemon model:
330 - master-daemon.log for the master daemon
331 - node-daemon.log for the node daemon (this is the same as in 1.2)
332 - rapi-daemon.log for the RAPI daemon logs
333 - rapi-access.log, an additional log file for the RAPI that will be
334 in the standard HTTP log format for possible parsing by other tools
336 Since the :term:`watcher` will only submit jobs to the master for
337 startup of the instances, its log file will contain less information
338 than before, mainly that it will start the instance, but not the
344 The only change to the node daemon is that, since we need better
345 concurrency, we don't process the inter-node RPC calls in the node
346 daemon itself, but we fork and process each request in a separate
349 Since we don't have many calls, and we only fork (not exec), the
350 overhead should be minimal.
355 A discussed alternative is to keep the current individual processes
356 touching the cluster configuration model. The reasons we have not
357 chosen this approach is:
359 - the speed of reading and unserializing the cluster state
360 today is not small enough that we can ignore it; the addition of
361 the job queue will make the startup cost even higher. While this
362 runtime cost is low, it can be on the order of a few seconds on
363 bigger clusters, which for very quick commands is comparable to
364 the actual duration of the computation itself
366 - individual commands would make it harder to implement a
367 fire-and-forget job request, along the lines "start this
368 instance but do not wait for it to finish"; it would require a
369 model of backgrounding the operation and other things that are
370 much better served by a daemon-based model
372 Another area of discussion is moving away from Twisted in this new
373 implementation. While Twisted has its advantages, there are also many
374 disadvantages to using it:
376 - first and foremost, it's not a library, but a framework; thus, if
377 you use twisted, all the code needs to be 'twiste-ized' and written
378 in an asynchronous manner, using deferreds; while this method works,
379 it's not a common way to code and it requires that the entire process
380 workflow is based around a single *reactor* (Twisted name for a main
382 - the more advanced granular locking that we want to implement would
383 require, if written in the async-manner, deep integration with the
384 Twisted stack, to such an extend that business-logic is inseparable
385 from the protocol coding; we felt that this is an unreasonable
386 request, and that a good protocol library should allow complete
387 separation of low-level protocol calls and business logic; by
388 comparison, the threaded approach combined with HTTPs protocol
389 required (for the first iteration) absolutely no changes from the 1.2
390 code, and later changes for optimizing the inter-node RPC calls
391 required just syntactic changes (e.g. ``rpc.call_...`` to
392 ``self.rpc.call_...``)
394 Another issue is with the Twisted API stability - during the Ganeti
395 1.x lifetime, we had to to implement many times workarounds to changes
396 in the Twisted version, so that for example 1.2 is able to use both
399 In the end, since we already had an HTTP server library for the RAPI,
400 we just reused that for inter-node communication.
406 We want to make sure that multiple operations can run in parallel on a
407 Ganeti Cluster. In order for this to happen we need to make sure
408 concurrently run operations don't step on each other toes and break the
411 This design addresses how we are going to deal with locking so that:
413 - we preserve data coherency
414 - we prevent deadlocks
415 - we prevent job starvation
417 Reaching the maximum possible parallelism is a Non-Goal. We have
418 identified a set of operations that are currently bottlenecks and need
419 to be parallelised and have worked on those. In the future it will be
420 possible to address other needs, thus making the cluster more and more
421 parallel one step at a time.
423 This section only talks about parallelising Ganeti level operations, aka
424 Logical Units, and the locking needed for that. Any other
425 synchronization lock needed internally by the code is outside its scope.
430 The proposed library has these features:
432 - internally managing all the locks, making the implementation
433 transparent from their usage
434 - automatically grabbing multiple locks in the right order (avoid
436 - ability to transparently handle conversion to more granularity
437 - support asynchronous operation (future goal)
439 Locking will be valid only on the master node and will not be a
440 distributed operation. Therefore, in case of master failure, the
441 operations currently running will be aborted and the locks will be
442 lost; it remains to the administrator to cleanup (if needed) the
443 operation result (e.g. make sure an instance is either installed
444 correctly or removed).
446 A corollary of this is that a master-failover operation with both
447 masters alive needs to happen while no operations are running, and
448 therefore no locks are held.
450 All the locks will be represented by objects (like
451 ``lockings.SharedLock``), and the individual locks for each object
452 will be created at initialisation time, from the config file.
454 The API will have a way to grab one or more than one locks at the same
455 time. Any attempt to grab a lock while already holding one in the wrong
456 order will be checked for, and fail.
462 At the first stage we have decided to provide the following locks:
464 - One "config file" lock
465 - One lock per node in the cluster
466 - One lock per instance in the cluster
468 All the instance locks will need to be taken before the node locks, and
469 the node locks before the config lock. Locks will need to be acquired at
470 the same time for multiple instances and nodes, and internal ordering
471 will be dealt within the locking library, which, for simplicity, will
472 just use alphabetical order.
474 Each lock has the following three possible statuses:
476 - unlocked (anyone can grab the lock)
477 - shared (anyone can grab/have the lock but only in shared mode)
478 - exclusive (no one else can grab/have the lock)
480 Handling conversion to more granularity
481 +++++++++++++++++++++++++++++++++++++++
483 In order to convert to a more granular approach transparently each time
484 we split a lock into more we'll create a "metalock", which will depend
485 on those sub-locks and live for the time necessary for all the code to
486 convert (or forever, in some conditions). When a metalock exists all
487 converted code must acquire it in shared mode, so it can run
488 concurrently, but still be exclusive with old code, which acquires it
491 In the beginning the only such lock will be what replaces the current
492 "command" lock, and will acquire all the locks in the system, before
493 proceeding. This lock will be called the "Big Ganeti Lock" because
494 holding that one will avoid any other concurrent Ganeti operations.
496 We might also want to devise more metalocks (eg. all nodes, all
497 nodes+config) in order to make it easier for some parts of the code to
498 acquire what it needs without specifying it explicitly.
500 In the future things like the node locks could become metalocks, should
501 we decide to split them into an even more fine grained approach, but
502 this will probably be only after the first 2.0 version has been
505 Adding/Removing locks
506 +++++++++++++++++++++
508 When a new instance or a new node is created an associated lock must be
509 added to the list. The relevant code will need to inform the locking
510 library of such a change.
512 This needs to be compatible with every other lock in the system,
513 especially metalocks that guarantee to grab sets of resources without
514 specifying them explicitly. The implementation of this will be handled
515 in the locking library itself.
517 When instances or nodes disappear from the cluster the relevant locks
518 must be removed. This is easier than adding new elements, as the code
519 which removes them must own them exclusively already, and thus deals
520 with metalocks exactly as normal code acquiring those locks. Any
521 operation queuing on a removed lock will fail after its removal.
523 Asynchronous operations
524 +++++++++++++++++++++++
526 For the first version the locking library will only export synchronous
527 operations, which will block till the needed lock are held, and only
528 fail if the request is impossible or somehow erroneous.
530 In the future we may want to implement different types of asynchronous
533 - try to acquire this lock set and fail if not possible
534 - try to acquire one of these lock sets and return the first one you
535 were able to get (or after a timeout) (select/poll like)
537 These operations can be used to prioritize operations based on available
538 locks, rather than making them just blindly queue for acquiring them.
539 The inherent risk, though, is that any code using the first operation,
540 or setting a timeout for the second one, is susceptible to starvation
541 and thus may never be able to get the required locks and complete
542 certain tasks. Considering this providing/using these operations should
543 not be among our first priorities.
548 For the first version of this code we'll convert each Logical Unit to
549 acquire/release the locks it needs, so locking will be at the Logical
550 Unit level. In the future we may want to split logical units in
551 independent "tasklets" with their own locking requirements. A different
552 design doc (or mini design doc) will cover the move from Logical Units
558 In general when acquiring locks we should use a code path equivalent
568 This makes sure we release all locks, and avoid possible deadlocks. Of
569 course extra care must be used not to leave, if possible locked
570 structures in an unusable state. Note that with Python 2.5 a simpler
571 syntax will be possible, but we want to keep compatibility with Python
572 2.4 so the new constructs should not be used.
574 In order to avoid this extra indentation and code changes everywhere in
575 the Logical Units code, we decided to allow LUs to declare locks, and
576 then execute their code with their locks acquired. In the new world LUs
577 are called like this::
579 # user passed names are expanded to the internal lock/resource name,
580 # then known needed locks are declared
582 ... some locking/adding of locks may happen ...
583 # late declaration of locks for one level: this is useful because sometimes
584 # we can't know which resource we need before locking the previous level
585 lu.DeclareLocks() # for each level (cluster, instance, node)
586 ... more locking/adding of locks can happen ...
587 # these functions are called with the proper locks held
590 ... locks declared for removal are removed, all acquired locks released ...
592 The Processor and the LogicalUnit class will contain exact documentation
593 on how locks are supposed to be declared.
598 This library will provide an easy upgrade path to bring all the code to
599 granular locking without breaking everything, and it will also guarantee
600 against a lot of common errors. Code switching from the old "lock
601 everything" lock to the new system, though, needs to be carefully
602 scrutinised to be sure it is really acquiring all the necessary locks,
603 and none has been overlooked or forgotten.
605 The code can contain other locks outside of this library, to synchronise
606 other threaded code (eg for the job queue) but in general these should
607 be leaf locks or carefully structured non-leaf ones, to avoid deadlock
611 .. _jqueue-original-design:
616 Granular locking is not enough to speed up operations, we also need a
617 queue to store these and to be able to process as many as possible in
620 A Ganeti job will consist of multiple ``OpCodes`` which are the basic
621 element of operation in Ganeti 1.2 (and will remain as such). Most
622 command-level commands are equivalent to one OpCode, or in some cases
623 to a sequence of opcodes, all of the same type (e.g. evacuating a node
624 will generate N opcodes of type replace disks).
627 Job execution—“Life of a Ganeti job”
628 ++++++++++++++++++++++++++++++++++++
630 #. Job gets submitted by the client. A new job identifier is generated
631 and assigned to the job. The job is then automatically replicated
632 [#replic]_ to all nodes in the cluster. The identifier is returned to
634 #. A pool of worker threads waits for new jobs. If all are busy, the job
635 has to wait and the first worker finishing its work will grab it.
636 Otherwise any of the waiting threads will pick up the new job.
637 #. Client waits for job status updates by calling a waiting RPC
638 function. Log message may be shown to the user. Until the job is
639 started, it can also be canceled.
640 #. As soon as the job is finished, its final result and status can be
641 retrieved from the server.
642 #. If the client archives the job, it gets moved to a history directory.
643 There will be a method to archive all jobs older than a a given age.
645 .. [#replic] We need replication in order to maintain the consistency
646 across all nodes in the system; the master node only differs in the
647 fact that now it is running the master daemon, but it if fails and we
648 do a master failover, the jobs are still visible on the new master
649 (though marked as failed).
651 Failures to replicate a job to other nodes will be only flagged as
652 errors in the master daemon log if more than half of the nodes failed,
653 otherwise we ignore the failure, and rely on the fact that the next
654 update (for still running jobs) will retry the update. For finished
655 jobs, it is less of a problem.
657 Future improvements will look into checking the consistency of the job
658 list and jobs themselves at master daemon startup.
664 Jobs are stored in the filesystem as individual files, serialized
665 using JSON (standard serialization mechanism in Ganeti).
667 The choice of storing each job in its own file was made because:
669 - a file can be atomically replaced
670 - a file can easily be replicated to other nodes
671 - checking consistency across nodes can be implemented very easily,
672 since all job files should be (at a given moment in time) identical
674 The other possible choices that were discussed and discounted were:
676 - single big file with all job data: not feasible due to difficult
678 - in-process databases: hard to replicate the entire database to the
679 other nodes, and replicating individual operations does not mean wee
686 All file operations have to be done atomically by writing to a temporary
687 file and subsequent renaming. Except for log messages, every change in a
688 job is stored and replicated to other nodes.
692 /var/lib/ganeti/queue/
693 job-1 (JSON encoded job description and status)
698 lock (Queue managing process opens this file in exclusive mode)
699 serial (Last job ID used)
700 version (Queue format version)
706 Locking in the job queue is a complicated topic. It is called from more
707 than one thread and must be thread-safe. For simplicity, a single lock
708 is used for the whole job queue.
710 A more detailed description can be found in doc/locking.rst.
716 RPC calls available between Ganeti master and node daemons:
718 jobqueue_update(file_name, content)
719 Writes a file in the job queue directory.
721 Cleans the job queue directory completely, including archived job.
722 jobqueue_rename(old, new)
723 Renames a file in the job queue directory.
729 RPC between Ganeti clients and the Ganeti master daemon supports the
730 following operations:
733 Submits a list of opcodes and returns the job identifier. The
734 identifier is guaranteed to be unique during the lifetime of a
736 WaitForJobChange(job_id, fields, […], timeout)
737 This function waits until a job changes or a timeout expires. The
738 condition for when a job changed is defined by the fields passed and
739 the last log message received.
740 QueryJobs(job_ids, fields)
741 Returns field values for the job identifiers passed.
743 Cancels the job specified by identifier. This operation may fail if
744 the job is already running, canceled or finished.
746 Moves a job into the …/archive/ directory. This operation will fail if
747 the job has not been canceled or finished.
750 Job and opcode status
751 +++++++++++++++++++++
753 Each job and each opcode has, at any time, one of the following states:
756 The job/opcode was submitted, but did not yet start.
758 The job/opcode is waiting for a lock to proceed.
760 The job/opcode is running.
762 The job/opcode was canceled before it started.
764 The job/opcode ran and finished successfully.
766 The job/opcode was aborted with an error.
768 If the master is aborted while a job is running, the job will be set to
769 the Error status once the master started again.
775 Archived jobs are kept in a separate directory,
776 ``/var/lib/ganeti/queue/archive/``. This is done in order to speed up
777 the queue handling: by default, the jobs in the archive are not
778 touched by any functions. Only the current (unarchived) jobs are
779 parsed, loaded, and verified (if implemented) by the master daemon.
785 The queue has to be completely empty for Ganeti updates with changes
786 in the job queue structure. In order to allow this, there will be a
787 way to prevent new jobs entering the queue.
793 Across all cluster configuration data, we have multiple classes of
796 A. cluster-wide parameters (e.g. name of the cluster, the master);
797 these are the ones that we have today, and are unchanged from the
802 #. instance specific parameters, e.g. the name of disks (LV), that
803 cannot be shared with other instances
805 #. instance parameters, that are or can be the same for many
806 instances, but are not hypervisor related; e.g. the number of VCPUs,
807 or the size of memory
809 #. instance parameters that are hypervisor specific (e.g. kernel_path
813 The following definitions for instance parameters will be used below:
815 :hypervisor parameter:
816 a hypervisor parameter (or hypervisor specific parameter) is defined
817 as a parameter that is interpreted by the hypervisor support code in
818 Ganeti and usually is specific to a particular hypervisor (like the
819 kernel path for :term:`PVM` which makes no sense for :term:`HVM`).
822 a backend parameter is defined as an instance parameter that can be
823 shared among a list of instances, and is either generic enough not
824 to be tied to a given hypervisor or cannot influence at all the
825 hypervisor behaviour.
827 For example: memory, vcpus, auto_balance
829 All these parameters will be encoded into constants.py with the prefix
830 "BE\_" and the whole list of parameters will exist in the set
834 a parameter whose value is unique to the instance (e.g. the name of a
835 LV, or the MAC of a NIC)
837 As a general rule, for all kind of parameters, “None” (or in
838 JSON-speak, “nil”) will no longer be a valid value for a parameter. As
839 such, only non-default parameters will be saved as part of objects in
840 the serialization step, reducing the size of the serialized format.
845 Cluster parameters remain as today, attributes at the top level of the
846 Cluster object. In addition, two new attributes at this level will
847 hold defaults for the instances:
849 - hvparams, a dictionary indexed by hypervisor type, holding default
850 values for hypervisor parameters that are not defined/overridden by
851 the instances of this hypervisor type
853 - beparams, a dictionary holding (for 2.0) a single element 'default',
854 which holds the default value for backend parameters
859 Node-related parameters are very few, and we will continue using the
860 same model for these as previously (attributes on the Node object).
862 There are three new node flags, described in a separate section "node
868 As described before, the instance parameters are split in three:
869 instance proper parameters, unique to each instance, instance
870 hypervisor parameters and instance backend parameters.
872 The “hvparams” and “beparams” are kept in two dictionaries at instance
873 level. Only non-default parameters are stored (but once customized, a
874 parameter will be kept, even with the same value as the default one,
877 The names for hypervisor parameters in the instance.hvparams subtree
878 should be choosen as generic as possible, especially if specific
879 parameters could conceivably be useful for more than one hypervisor,
880 e.g. ``instance.hvparams.vnc_console_port`` instead of using both
881 ``instance.hvparams.hvm_vnc_console_port`` and
882 ``instance.hvparams.kvm_vnc_console_port``.
884 There are some special cases related to disks and NICs (for example):
885 a disk has both Ganeti-related parameters (e.g. the name of the LV)
886 and hypervisor-related parameters (how the disk is presented to/named
887 in the instance). The former parameters remain as proper-instance
888 parameters, while the latter value are migrated to the hvparams
889 structure. In 2.0, we will have only globally-per-instance such
890 hypervisor parameters, and not per-disk ones (e.g. all NICs will be
891 exported as of the same type).
893 Starting from the 1.2 list of instance parameters, here is how they
894 will be mapped to the three classes of parameters:
909 - hvm_boot_order (HV)
912 - hvm_cdrom_image_path (HV)
915 - vnc_bind_address (HV)
922 To support the new cluster parameter design, additional features will
923 be required from the hypervisor support implementations in Ganeti.
925 The hypervisor support implementation API will be extended with the
928 :PARAMETERS: class-level attribute holding the list of valid parameters
930 :CheckParamSyntax(hvparams): checks that the given parameters are
931 valid (as in the names are valid) for this hypervisor; usually just
932 comparing ``hvparams.keys()`` and ``cls.PARAMETERS``; this is a class
933 method that can be called from within master code (i.e. cmdlib) and
934 should be safe to do so
935 :ValidateParameters(hvparams): verifies the values of the provided
936 parameters against this hypervisor; this is a method that will be
937 called on the target node, from backend.py code, and as such can
938 make node-specific checks (e.g. kernel_path checking)
940 Default value application
941 +++++++++++++++++++++++++
943 The application of defaults to an instance is done in the Cluster
944 object, via two new methods as follows:
946 - ``Cluster.FillHV(instance)``, returns 'filled' hvparams dict, based on
947 instance's hvparams and cluster's ``hvparams[instance.hypervisor]``
949 - ``Cluster.FillBE(instance, be_type="default")``, which returns the
950 beparams dict, based on the instance and cluster beparams
952 The FillHV/BE transformations will be used, for example, in the
953 RpcRunner when sending an instance for activation/stop, and the sent
954 instance hvparams/beparams will have the final value (noded code doesn't
955 know about defaults).
957 LU code will need to self-call the transformation, if needed.
962 The parameter changes will have impact on the OpCodes, especially on
965 - ``OpInstanceCreate``, where the new hv and be parameters will be sent
966 as dictionaries; note that all hv and be parameters are now optional,
967 as the values can be instead taken from the cluster
968 - ``OpInstanceQuery``, where we have to be able to query these new
969 parameters; the syntax for names will be ``hvparam/$NAME`` and
970 ``beparam/$NAME`` for querying an individual parameter out of one
971 dictionary, and ``hvparams``, respectively ``beparams``, for the whole
973 - ``OpModifyInstance``, where the the modified parameters are sent as
976 Additionally, we will need new OpCodes to modify the cluster-level
977 defaults for the be/hv sets of parameters.
982 One problem that might appear is that our classification is not
983 complete or not good enough, and we'll need to change this model. As
984 the last resort, we will need to rollback and keep 1.2 style.
986 Another problem is that classification of one parameter is unclear
987 (e.g. ``network_port``, is this BE or HV?); in this case we'll take
988 the risk of having to move parameters later between classes.
993 The only security issue that we foresee is if some new parameters will
994 have sensitive value. If so, we will need to have a way to export the
995 config data while purging the sensitive value.
997 E.g. for the drbd shared secrets, we could export these with the
998 values replaced by an empty string.
1003 Ganeti 2.0 adds three node flags that change the way nodes are handled
1004 within Ganeti and the related infrastructure (iallocator interaction,
1007 *master candidate* flag
1008 +++++++++++++++++++++++
1010 Ganeti 2.0 allows more scalability in operation by introducing
1011 parallelization. However, a new bottleneck is reached that is the
1012 synchronization and replication of cluster configuration to all nodes
1015 This breaks scalability as the speed of the replication decreases
1016 roughly with the size of the nodes in the cluster. The goal of the
1017 master candidate flag is to change this O(n) into O(1) with respect to
1018 job and configuration data propagation.
1020 Only nodes having this flag set (let's call this set of nodes the
1021 *candidate pool*) will have jobs and configuration data replicated.
1023 The cluster will have a new parameter (runtime changeable) called
1024 ``candidate_pool_size`` which represents the number of candidates the
1025 cluster tries to maintain (preferably automatically).
1027 This will impact the cluster operations as follows:
1029 - jobs and config data will be replicated only to a fixed set of nodes
1030 - master fail-over will only be possible to a node in the candidate pool
1031 - cluster verify needs changing to account for these two roles
1032 - external scripts will no longer have access to the configuration
1033 file (this is not recommended anyway)
1036 The caveats of this change are:
1038 - if all candidates are lost (completely), cluster configuration is
1039 lost (but it should be backed up external to the cluster anyway)
1041 - failed nodes which are candidate must be dealt with properly, so
1042 that we don't lose too many candidates at the same time; this will be
1043 reported in cluster verify
1045 - the 'all equal' concept of ganeti is no longer true
1047 - the partial distribution of config data means that all nodes will
1048 have to revert to ssconf files for master info (as in 1.2)
1052 - speed on a 100+ nodes simulated cluster is greatly enhanced, even
1053 for a simple operation; ``gnt-instance remove`` on a diskless instance
1054 remove goes from ~9seconds to ~2 seconds
1056 - node failure of non-candidates will be less impacting on the cluster
1058 The default value for the candidate pool size will be set to 10 but
1059 this can be changed at cluster creation and modified any time later.
1061 Testing on simulated big clusters with sequential and parallel jobs
1062 show that this value (10) is a sweet-spot from performance and load
1068 In order to support better the situation in which nodes are offline
1069 (e.g. for repair) without altering the cluster configuration, Ganeti
1070 needs to be told and needs to properly handle this state for nodes.
1072 This will result in simpler procedures, and less mistakes, when the
1073 amount of node failures is high on an absolute scale (either due to
1074 high failure rate or simply big clusters).
1076 Nodes having this attribute set will not be contacted for inter-node
1077 RPC calls, will not be master candidates, and will not be able to host
1078 instances as primaries.
1080 Setting this attribute on a node:
1082 - will not be allowed if the node is the master
1083 - will not be allowed if the node has primary instances
1084 - will cause the node to be demoted from the master candidate role (if
1085 it was), possibly causing another node to be promoted to that role
1087 This attribute will impact the cluster operations as follows:
1089 - querying these nodes for anything will fail instantly in the RPC
1090 library, with a specific RPC error (RpcResult.offline == True)
1092 - they will be listed in the Other section of cluster verify
1094 The code is changed in the following ways:
1096 - RPC calls were be converted to skip such nodes:
1098 - RpcRunner-instance-based RPC calls are easy to convert
1100 - static/classmethod RPC calls are harder to convert, and were left
1103 - the RPC results were unified so that this new result state (offline)
1104 can be differentiated
1106 - master voting still queries in repair nodes, as we need to ensure
1107 consistency in case the (wrong) masters have old data, and nodes have
1108 come back from repairs
1112 - some operation semantics are less clear (e.g. what to do on instance
1113 start with offline secondary?); for now, these will just fail as if
1114 the flag is not set (but faster)
1115 - 2-node cluster with one node offline needs manual startup of the
1116 master with a special flag to skip voting (as the master can't get a
1119 One of the advantages of implementing this flag is that it will allow
1120 in the future automation tools to automatically put the node in
1121 repairs and recover from this state, and the code (should/will) handle
1122 this much better than just timing out. So, future possible
1123 improvements (for later versions):
1125 - watcher will detect nodes which fail RPC calls, will attempt to ssh
1126 to them, if failure will put them offline
1127 - watcher will try to ssh and query the offline nodes, if successful
1128 will take them off the repair list
1130 Alternatives considered: The RPC call model in 2.0 is, by default,
1131 much nicer - errors are logged in the background, and job/opcode
1132 execution is clearer, so we could simply not introduce this. However,
1133 having this state will make both the codepaths clearer (offline
1134 vs. temporary failure) and the operational model (it's not a node with
1135 errors, but an offline node).
1141 Due to parallel execution of jobs in Ganeti 2.0, we could have the
1142 following situation:
1144 - gnt-node migrate + failover is run
1145 - gnt-node evacuate is run, which schedules a long-running 6-opcode
1147 - partway through, a new job comes in that runs an iallocator script,
1148 which finds the above node as empty and a very good candidate
1149 - gnt-node evacuate has finished, but now it has to be run again, to
1150 clean the above instance(s)
1152 In order to prevent this situation, and to be able to get nodes into
1153 proper offline status easily, a new *drained* flag was added to the
1156 This flag (which actually means "is being, or was drained, and is
1157 expected to go offline"), will prevent allocations on the node, but
1158 otherwise all other operations (start/stop instance, query, etc.) are
1159 working without any restrictions.
1161 Interaction between flags
1162 +++++++++++++++++++++++++
1164 While these flags are implemented as separate flags, they are
1165 mutually-exclusive and are acting together with the master node role
1166 as a single *node status* value. In other words, a flag is only in one
1167 of these roles at a given time. The lack of any of these flags denote
1170 The current node status is visible in the ``gnt-cluster verify``
1171 output, and the individual flags can be examined via separate flags in
1172 the ``gnt-node list`` output.
1174 These new flags will be exported in both the iallocator input message
1175 and via RAPI, see the respective man pages for the exact names.
1180 The main feature-level changes will be:
1182 - a number of disk related changes
1183 - removal of fixed two-disk, one-nic per instance limitation
1185 Disk handling changes
1186 ~~~~~~~~~~~~~~~~~~~~~
1188 The storage options available in Ganeti 1.x were introduced based on
1189 then-current software (first DRBD 0.7 then later DRBD 8) and the
1190 estimated usage patters. However, experience has later shown that some
1191 assumptions made initially are not true and that more flexibility is
1194 One main assumption made was that disk failures should be treated as
1195 'rare' events, and that each of them needs to be manually handled in
1196 order to ensure data safety; however, both these assumptions are false:
1198 - disk failures can be a common occurrence, based on usage patterns or
1200 - our disk setup is robust enough (referring to DRBD8 + LVM) that we
1201 could automate more of the recovery
1203 Note that we still don't have fully-automated disk recovery as a goal,
1204 but our goal is to reduce the manual work needed.
1206 As such, we plan the following main changes:
1208 - DRBD8 is much more flexible and stable than its previous version
1209 (0.7), such that removing the support for the ``remote_raid1``
1210 template and focusing only on DRBD8 is easier
1212 - dynamic discovery of DRBD devices is not actually needed in a cluster
1213 that where the DRBD namespace is controlled by Ganeti; switching to a
1214 static assignment (done at either instance creation time or change
1215 secondary time) will change the disk activation time from O(n) to
1216 O(1), which on big clusters is a significant gain
1218 - remove the hard dependency on LVM (currently all available storage
1219 types are ultimately backed by LVM volumes) by introducing file-based
1222 Additionally, a number of smaller enhancements are also planned:
1223 - support variable number of disks
1224 - support read-only disks
1226 Future enhancements in the 2.x series, which do not require base design
1227 changes, might include:
1229 - enhancement of the LVM allocation method in order to try to keep
1230 all of an instance's virtual disks on the same physical
1233 - add support for DRBD8 authentication at handshake time in
1234 order to ensure each device connects to the correct peer
1236 - remove the restrictions on failover only to the secondary
1237 which creates very strict rules on cluster allocation
1239 DRBD minor allocation
1240 +++++++++++++++++++++
1242 Currently, when trying to identify or activate a new DRBD (or MD)
1243 device, the code scans all in-use devices in order to see if we find
1244 one that looks similar to our parameters and is already in the desired
1245 state or not. Since this needs external commands to be run, it is very
1246 slow when more than a few devices are already present.
1248 Therefore, we will change the discovery model from dynamic to
1249 static. When a new device is logically created (added to the
1250 configuration) a free minor number is computed from the list of
1251 devices that should exist on that node and assigned to that
1254 At device activation, if the minor is already in use, we check if
1255 it has our parameters; if not so, we just destroy the device (if
1256 possible, otherwise we abort) and start it with our own
1259 This means that we in effect take ownership of the minor space for
1260 that device type; if there's a user-created DRBD minor, it will be
1261 automatically removed.
1263 The change will have the effect of reducing the number of external
1264 commands run per device from a constant number times the index of the
1265 first free DRBD minor to just a constant number.
1267 Removal of obsolete device types (MD, DRBD7)
1268 ++++++++++++++++++++++++++++++++++++++++++++
1270 We need to remove these device types because of two issues. First,
1271 DRBD7 has bad failure modes in case of dual failures (both network and
1272 disk - it cannot propagate the error up the device stack and instead
1273 just panics. Second, due to the asymmetry between primary and
1274 secondary in MD+DRBD mode, we cannot do live failover (not even if we
1277 File-based storage support
1278 ++++++++++++++++++++++++++
1280 Using files instead of logical volumes for instance storage would
1281 allow us to get rid of the hard requirement for volume groups for
1282 testing clusters and it would also allow usage of SAN storage to do
1283 live failover taking advantage of this storage solution.
1285 Better LVM allocation
1286 +++++++++++++++++++++
1288 Currently, the LV to PV allocation mechanism is a very simple one: at
1289 each new request for a logical volume, tell LVM to allocate the volume
1290 in order based on the amount of free space. This is good for
1291 simplicity and for keeping the usage equally spread over the available
1292 physical disks, however it introduces a problem that an instance could
1293 end up with its (currently) two drives on two physical disks, or
1294 (worse) that the data and metadata for a DRBD device end up on
1297 This is bad because it causes unneeded ``replace-disks`` operations in
1298 case of a physical failure.
1300 The solution is to batch allocations for an instance and make the LVM
1301 handling code try to allocate as close as possible all the storage of
1302 one instance. We will still allow the logical volumes to spill over to
1303 additional disks as needed.
1305 Note that this clustered allocation can only be attempted at initial
1306 instance creation, or at change secondary node time. At add disk time,
1307 or at replacing individual disks, it's not easy enough to compute the
1308 current disk map so we'll not attempt the clustering.
1310 DRBD8 peer authentication at handshake
1311 ++++++++++++++++++++++++++++++++++++++
1313 DRBD8 has a new feature that allow authentication of the peer at
1314 connect time. We can use this to prevent connecting to the wrong peer
1315 more that securing the connection. Even though we never had issues
1316 with wrong connections, it would be good to implement this.
1319 LVM self-repair (optional)
1320 ++++++++++++++++++++++++++
1322 The complete failure of a physical disk is very tedious to
1323 troubleshoot, mainly because of the many failure modes and the many
1324 steps needed. We can safely automate some of the steps, more
1325 specifically the ``vgreduce --removemissing`` using the following
1328 #. check if all nodes have consistent volume groups
1329 #. if yes, and previous status was yes, do nothing
1330 #. if yes, and previous status was no, save status and restart
1331 #. if no, and previous status was no, do nothing
1332 #. if no, and previous status was yes:
1333 #. if more than one node is inconsistent, do nothing
1334 #. if only one node is inconsistent:
1335 #. run ``vgreduce --removemissing``
1336 #. log this occurrence in the Ganeti log in a form that
1337 can be used for monitoring
1338 #. [FUTURE] run ``replace-disks`` for all
1341 Failover to any node
1342 ++++++++++++++++++++
1344 With a modified disk activation sequence, we can implement the
1345 *failover to any* functionality, removing many of the layout
1346 restrictions of a cluster:
1348 - the need to reserve memory on the current secondary: this gets reduced
1349 to a must to reserve memory anywhere on the cluster
1351 - the need to first failover and then replace secondary for an
1352 instance: with failover-to-any, we can directly failover to
1353 another node, which also does the replace disks at the same
1356 In the following, we denote the current primary by P1, the current
1357 secondary by S1, and the new primary and secondaries by P2 and S2. P2
1358 is fixed to the node the user chooses, but the choice of S2 can be
1359 made between P1 and S1. This choice can be constrained, depending on
1360 which of P1 and S1 has failed.
1362 - if P1 has failed, then S1 must become S2, and live migration is not
1364 - if S1 has failed, then P1 must become S2, and live migration could be
1365 possible (in theory, but this is not a design goal for 2.0)
1367 The algorithm for performing the failover is straightforward:
1369 - verify that S2 (the node the user has chosen to keep as secondary) has
1370 valid data (is consistent)
1372 - tear down the current DRBD association and setup a DRBD pairing
1373 between P2 (P2 is indicated by the user) and S2; since P2 has no data,
1374 it will start re-syncing from S2
1376 - as soon as P2 is in state SyncTarget (i.e. after the resync has
1377 started but before it has finished), we can promote it to primary role
1378 (r/w) and start the instance on P2
1380 - as soon as the P2?S2 sync has finished, we can remove
1381 the old data on the old node that has not been chosen for
1384 Caveats: during the P2?S2 sync, a (non-transient) network error
1385 will cause I/O errors on the instance, so (if a longer instance
1386 downtime is acceptable) we can postpone the restart of the instance
1387 until the resync is done. However, disk I/O errors on S2 will cause
1388 data loss, since we don't have a good copy of the data anymore, so in
1389 this case waiting for the sync to complete is not an option. As such,
1390 it is recommended that this feature is used only in conjunction with
1391 proper disk monitoring.
1394 Live migration note: While failover-to-any is possible for all choices
1395 of S2, migration-to-any is possible only if we keep P1 as S2.
1400 The dynamic device model, while more complex, has an advantage: it
1401 will not reuse by mistake the DRBD device of another instance, since
1402 it always looks for either our own or a free one.
1404 The static one, in contrast, will assume that given a minor number N,
1405 it's ours and we can take over. This needs careful implementation such
1406 that if the minor is in use, either we are able to cleanly shut it
1407 down, or we abort the startup. Otherwise, it could be that we start
1408 syncing between two instance's disks, causing data loss.
1411 Variable number of disk/NICs per instance
1412 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1414 Variable number of disks
1415 ++++++++++++++++++++++++
1417 In order to support high-security scenarios (for example read-only sda
1418 and read-write sdb), we need to make a fully flexibly disk
1419 definition. This has less impact that it might look at first sight:
1420 only the instance creation has hard coded number of disks, not the disk
1421 handling code. The block device handling and most of the instance
1422 handling code is already working with "the instance's disks" as
1423 opposed to "the two disks of the instance", but some pieces are not
1424 (e.g. import/export) and the code needs a review to ensure safety.
1426 The objective is to be able to specify the number of disks at
1427 instance creation, and to be able to toggle from read-only to
1428 read-write a disk afterward.
1430 Variable number of NICs
1431 +++++++++++++++++++++++
1433 Similar to the disk change, we need to allow multiple network
1434 interfaces per instance. This will affect the internal code (some
1435 function will have to stop assuming that ``instance.nics`` is a list
1436 of length one), the OS API which currently can export/import only one
1437 instance, and the command line interface.
1442 There are two areas of interface changes: API-level changes (the OS
1443 interface and the RAPI interface) and the command line interface
1449 The current Ganeti OS interface, version 5, is tailored for Ganeti 1.2.
1450 The interface is composed by a series of scripts which get called with
1451 certain parameters to perform OS-dependent operations on the cluster.
1452 The current scripts are:
1455 called when a new instance is added to the cluster
1457 called to export an instance disk to a stream
1459 called to import from a stream to a new instance
1461 called to perform the os-specific operations necessary for renaming an
1464 Currently these scripts suffer from the limitations of Ganeti 1.2: for
1465 example they accept exactly one block and one swap devices to operate
1466 on, rather than any amount of generic block devices, they blindly assume
1467 that an instance will have just one network interface to operate, they
1468 can not be configured to optimise the instance for a particular
1471 Since in Ganeti 2.0 we want to support multiple hypervisors, and a
1472 non-fixed number of network and disks the OS interface need to change to
1473 transmit the appropriate amount of information about an instance to its
1474 managing operating system, when operating on it. Moreover since some old
1475 assumptions usually used in OS scripts are no longer valid we need to
1476 re-establish a common knowledge on what can be assumed and what cannot
1477 be regarding Ganeti environment.
1480 When designing the new OS API our priorities are:
1482 - future extensibility
1483 - ease of porting from the old API
1486 As such we want to limit the number of scripts that must be written to
1487 support an OS, and make it easy to share code between them by uniforming
1488 their input. We also will leave the current script structure unchanged,
1489 as far as we can, and make a few of the scripts (import, export and
1490 rename) optional. Most information will be passed to the script through
1491 environment variables, for ease of access and at the same time ease of
1492 using only the information a script needs.
1498 As in Ganeti 1.2, every OS which wants to be installed in Ganeti needs
1499 to support the following functionality, through scripts:
1502 used to create a new instance running that OS. This script should
1503 prepare the block devices, and install them so that the new OS can
1504 boot under the specified hypervisor.
1506 used to export an installed instance using the given OS to a format
1507 which can be used to import it back into a new instance.
1509 used to import an exported instance into a new one. This script is
1510 similar to create, but the new instance should have the content of the
1511 export, rather than contain a pristine installation.
1513 used to perform the internal OS-specific operations needed to rename
1516 If any optional script is not implemented Ganeti will refuse to perform
1517 the given operation on instances using the non-implementing OS. Of
1518 course the create script is mandatory, and it doesn't make sense to
1519 support the either the export or the import operation but not both.
1521 Incompatibilities with 1.2
1522 __________________________
1524 We expect the following incompatibilities between the OS scripts for 1.2
1525 and the ones for 2.0:
1527 - Input parameters: in 1.2 those were passed on the command line, in 2.0
1528 we'll use environment variables, as there will be a lot more
1529 information and not all OSes may care about all of it.
1530 - Number of calls: export scripts will be called once for each device
1531 the instance has, and import scripts once for every exported disk.
1532 Imported instances will be forced to have a number of disks greater or
1533 equal to the one of the export.
1534 - Some scripts are not compulsory: if such a script is missing the
1535 relevant operations will be forbidden for instances of that OS. This
1536 makes it easier to distinguish between unsupported operations and
1537 no-op ones (if any).
1543 Rather than using command line flags, as they do now, scripts will
1544 accept inputs from environment variables. We expect the following input
1548 The version of the OS API that the following parameters comply with;
1549 this is used so that in the future we could have OSes supporting
1550 multiple versions and thus Ganeti send the proper version in this
1553 Name of the instance acted on
1555 The hypervisor the instance should run on (e.g. 'xen-pvm', 'xen-hvm',
1558 The number of disks this instance will have
1560 The number of NICs this instance will have
1562 Path to the Nth disk.
1564 W if read/write, R if read only. OS scripts are not supposed to touch
1565 read-only disks, but will be passed them to know.
1566 DISK_<N>_FRONTEND_TYPE
1567 Type of the disk as seen by the instance. Can be 'scsi', 'ide',
1569 DISK_<N>_BACKEND_TYPE
1570 Type of the disk as seen from the node. Can be 'block', 'file:loop' or
1573 Mac address for the Nth network interface
1575 Ip address for the Nth network interface, if available
1577 Node bridge the Nth network interface will be connected to
1578 NIC_<N>_FRONTEND_TYPE
1579 Type of the Nth NIC as seen by the instance. For example 'virtio',
1582 Whether more out should be produced, for debugging purposes. Currently
1583 the only valid values are 0 and 1.
1585 These are only the basic variables we are thinking of now, but more
1586 may come during the implementation and they will be documented in the
1587 :manpage:`ganeti-os-api` man page. All these variables will be
1588 available to all scripts.
1590 Some scripts will need a few more information to work. These will have
1591 per-script variables, such as for example:
1594 rename: the name the instance should be renamed from.
1596 export: device to be exported, a snapshot of the actual device. The
1597 data must be exported to stdout.
1599 export: sequential number of the instance device targeted.
1601 import: device to send the data to, part of the new instance. The data
1602 must be imported from stdin.
1604 import: sequential number of the instance device targeted.
1606 (Rationale for INSTANCE_NAME as an environment variable: the instance
1607 name is always needed and we could pass it on the command line. On the
1608 other hand, though, this would force scripts to both access the
1609 environment and parse the command line, so we'll move it for
1616 As discussed scripts should only send user-targeted information to
1617 stderr. The create and import scripts are supposed to format/initialise
1618 the given block devices and install the correct instance data. The
1619 export script is supposed to export instance data to stdout in a format
1620 understandable by the the import script. The data will be compressed by
1621 Ganeti, so no compression should be done. The rename script should only
1622 modify the instance's knowledge of what its name is.
1624 Other declarative style features
1625 ++++++++++++++++++++++++++++++++
1627 Similar to Ganeti 1.2, OS specifications will need to provide a
1628 'ganeti_api_version' containing list of numbers matching the
1629 version(s) of the API they implement. Ganeti itself will always be
1630 compatible with one version of the API and may maintain backwards
1631 compatibility if it's feasible to do so. The numbers are one-per-line,
1632 so an OS supporting both version 5 and version 20 will have a file
1633 containing two lines. This is different from Ganeti 1.2, which only
1634 supported one version number.
1636 In addition to that an OS will be able to declare that it does support
1637 only a subset of the Ganeti hypervisors, by declaring them in the
1644 We might want to have a "default" import/export behaviour that just
1645 dumps all disks and restores them. This can save work as most systems
1646 will just do this, while allowing flexibility for different systems.
1648 Environment variables are limited in size, but we expect that there will
1649 be enough space to store the information we need. If we discover that
1650 this is not the case we may want to go to a more complex API such as
1651 storing those information on the filesystem and providing the OS script
1652 with the path to a file where they are encoded in some format.
1659 The first Ganeti remote API (RAPI) was designed and deployed with the
1660 Ganeti 1.2.5 release. That version provide read-only access to the
1661 cluster state. Fully functional read-write API demands significant
1662 internal changes which will be implemented in version 2.0.
1664 We decided to go with implementing the Ganeti RAPI in a RESTful way,
1665 which is aligned with key features we looking. It is simple,
1666 stateless, scalable and extensible paradigm of API implementation. As
1667 transport it uses HTTP over SSL, and we are implementing it with JSON
1668 encoding, but in a way it possible to extend and provide any other
1674 The Ganeti RAPI is implemented as independent daemon, running on the
1675 same node with the same permission level as Ganeti master
1676 daemon. Communication is done through the LUXI library to the master
1677 daemon. In order to keep communication asynchronous RAPI processes two
1678 types of client requests:
1680 - queries: server is able to answer immediately
1681 - job submission: some time is required for a useful response
1683 In the query case requested data send back to client in the HTTP
1684 response body. Typical examples of queries would be: list of nodes,
1685 instances, cluster info, etc.
1687 In the case of job submission, the client receive a job ID, the
1688 identifier which allows one to query the job progress in the job queue
1691 Internally, each exported object has an version identifier, which is
1692 used as a state identifier in the HTTP header E-Tag field for
1693 requests/responses to avoid race conditions.
1696 Resource representation
1697 +++++++++++++++++++++++
1699 The key difference of using REST instead of others API is that REST
1700 requires separation of services via resources with unique URIs. Each
1701 of them should have limited amount of state and support standard HTTP
1702 methods: GET, POST, DELETE, PUT.
1704 For example in Ganeti's case we can have a set of URI:
1706 - ``/{clustername}/instances``
1707 - ``/{clustername}/instances/{instancename}``
1708 - ``/{clustername}/instances/{instancename}/tag``
1709 - ``/{clustername}/tag``
1711 A GET request to ``/{clustername}/instances`` will return the list of
1712 instances, a POST to ``/{clustername}/instances`` should create a new
1713 instance, a DELETE ``/{clustername}/instances/{instancename}`` should
1714 delete the instance, a GET ``/{clustername}/tag`` should return get
1717 Each resource URI will have a version prefix. The resource IDs are to
1720 Internal encoding might be JSON, XML, or any other. The JSON encoding
1721 fits nicely in Ganeti RAPI needs. The client can request a specific
1722 representation via the Accept field in the HTTP header.
1724 REST uses HTTP as its transport and application protocol for resource
1725 access. The set of possible responses is a subset of standard HTTP
1728 The statelessness model provides additional reliability and
1729 transparency to operations (e.g. only one request needs to be analyzed
1730 to understand the in-progress operation, not a sequence of multiple
1731 requests/responses).
1737 With the write functionality security becomes a much bigger an issue.
1738 The Ganeti RAPI uses basic HTTP authentication on top of an
1739 SSL-secured connection to grant access to an exported resource. The
1740 password is stored locally in an Apache-style ``.htpasswd`` file. Only
1741 one level of privileges is supported.
1746 The model detailed above for job submission requires the client to
1747 poll periodically for updates to the job; an alternative would be to
1748 allow the client to request a callback, or a 'wait for updates' call.
1750 The callback model was not considered due to the following two issues:
1752 - callbacks would require a new model of allowed callback URLs,
1753 together with a method of managing these
1754 - callbacks only work when the client and the master are in the same
1755 security domain, and they fail in the other cases (e.g. when there is
1756 a firewall between the client and the RAPI daemon that only allows
1757 client-to-RAPI calls, which is usual in DMZ cases)
1759 The 'wait for updates' method is not suited to the HTTP protocol,
1760 where requests are supposed to be short-lived.
1762 Command line changes
1763 ~~~~~~~~~~~~~~~~~~~~
1765 Ganeti 2.0 introduces several new features as well as new ways to
1766 handle instance resources like disks or network interfaces. This
1767 requires some noticeable changes in the way command line arguments are
1770 - extend and modify command line syntax to support new features
1771 - ensure consistent patterns in command line arguments to reduce
1774 The design changes that require these changes are, in no particular
1777 - flexible instance disk handling: support a variable number of disks
1778 with varying properties per instance,
1779 - flexible instance network interface handling: support a variable
1780 number of network interfaces with varying properties per instance
1781 - multiple hypervisors: multiple hypervisors can be active on the same
1782 cluster, each supporting different parameters,
1783 - support for device type CDROM (via ISO image)
1785 As such, there are several areas of Ganeti where the command line
1786 arguments will change:
1788 - Cluster configuration
1790 - cluster initialization
1791 - cluster default configuration
1793 - Instance configuration
1795 - handling of network cards for instances,
1796 - handling of disks for instances,
1797 - handling of CDROM devices and
1798 - handling of hypervisor specific options.
1800 There are several areas of Ganeti where the command line arguments
1803 - Cluster configuration
1805 - cluster initialization
1806 - cluster default configuration
1808 - Instance configuration
1810 - handling of network cards for instances,
1811 - handling of disks for instances,
1812 - handling of CDROM devices and
1813 - handling of hypervisor specific options.
1815 Notes about device removal/addition
1816 +++++++++++++++++++++++++++++++++++
1818 To avoid problems with device location changes (e.g. second network
1819 interface of the instance becoming the first or third and the like)
1820 the list of network/disk devices is treated as a stack, i.e. devices
1821 can only be added/removed at the end of the list of devices of each
1822 class (disk or network) for each instance.
1824 gnt-instance commands
1825 +++++++++++++++++++++
1827 The commands for gnt-instance will be modified and extended to allow
1828 for the new functionality:
1830 - the add command will be extended to support the new device and
1832 - the modify command continues to handle all modifications to
1833 instances, but will be extended with new arguments for handling
1836 Network Device Options
1837 ++++++++++++++++++++++
1839 The generic format of the network device option is:
1841 --net $DEVNUM[:$OPTION=$VALUE][,$OPTION=VALUE]
1843 :$DEVNUM: device number, unsigned integer, starting at 0,
1844 :$OPTION: device option, string,
1845 :$VALUE: device option value, string.
1847 Currently, the following device options will be defined (open to
1850 :mac: MAC address of the network interface, accepts either a valid
1851 MAC address or the string 'auto'. If 'auto' is specified, a new MAC
1852 address will be generated randomly. If the mac device option is not
1853 specified, the default value 'auto' is assumed.
1854 :bridge: network bridge the network interface is connected
1855 to. Accepts either a valid bridge name (the specified bridge must
1856 exist on the node(s)) as string or the string 'auto'. If 'auto' is
1857 specified, the default brigde is used. If the bridge option is not
1858 specified, the default value 'auto' is assumed.
1863 The generic format of the disk device option is:
1865 --disk $DEVNUM[:$OPTION=$VALUE][,$OPTION=VALUE]
1867 :$DEVNUM: device number, unsigned integer, starting at 0,
1868 :$OPTION: device option, string,
1869 :$VALUE: device option value, string.
1871 Currently, the following device options will be defined (open to
1874 :size: size of the disk device, either a positive number, specifying
1875 the disk size in mebibytes, or a number followed by a magnitude suffix
1876 (M for mebibytes, G for gibibytes). Also accepts the string 'auto' in
1877 which case the default disk size will be used. If the size option is
1878 not specified, 'auto' is assumed. This option is not valid for all
1880 :access: access mode of the disk device, a single letter, valid values
1883 - *w*: read/write access to the disk device or
1884 - *r*: read-only access to the disk device.
1886 If the access mode is not specified, the default mode of read/write
1887 access will be configured.
1888 :path: path to the image file for the disk device, string. No default
1889 exists. This option is not valid for all disk layout types.
1894 To add devices to an already existing instance, use the device type
1895 specific option to gnt-instance modify. Currently, there are two
1896 device type specific options supported:
1898 :--net: for network interface cards
1899 :--disk: for disk devices
1901 The syntax to the device specific options is similar to the generic
1902 device options, but instead of specifying a device number like for
1903 gnt-instance add, you specify the magic string add. The new device
1904 will always be appended at the end of the list of devices of this type
1905 for the specified instance, e.g. if the instance has disk devices 0,1
1906 and 2, the newly added disk device will be disk device 3.
1908 Example: gnt-instance modify --net add:mac=auto test-instance
1913 Removing devices from and instance is done via gnt-instance
1914 modify. The same device specific options as for adding instances are
1915 used. Instead of a device number and further device options, only the
1916 magic string remove is specified. It will always remove the last
1917 device in the list of devices of this type for the instance specified,
1918 e.g. if the instance has disk devices 0, 1, 2 and 3, the disk device
1919 number 3 will be removed.
1921 Example: gnt-instance modify --net remove test-instance
1926 Modifying devices is also done with device type specific options to
1927 the gnt-instance modify command. There are currently two device type
1930 :--net: for network interface cards
1931 :--disk: for disk devices
1933 The syntax to the device specific options is similar to the generic
1934 device options. The device number you specify identifies the device to
1939 gnt-instance modify --disk 2:access=r
1944 Ganeti 2.0 will support more than one hypervisor. Different
1945 hypervisors have various options that only apply to a specific
1946 hypervisor. Those hypervisor specific options are treated specially
1947 via the ``--hypervisor`` option. The generic syntax of the hypervisor
1948 option is as follows::
1950 --hypervisor $HYPERVISOR:$OPTION=$VALUE[,$OPTION=$VALUE]
1952 :$HYPERVISOR: symbolic name of the hypervisor to use, string,
1953 has to match the supported hypervisors. Example: xen-pvm
1955 :$OPTION: hypervisor option name, string
1956 :$VALUE: hypervisor option value, string
1958 The hypervisor option for an instance can be set on instance creation
1959 time via the ``gnt-instance add`` command. If the hypervisor for an
1960 instance is not specified upon instance creation, the default
1961 hypervisor will be used.
1963 Modifying hypervisor parameters
1964 +++++++++++++++++++++++++++++++
1966 The hypervisor parameters of an existing instance can be modified
1967 using ``--hypervisor`` option of the ``gnt-instance modify``
1968 command. However, the hypervisor type of an existing instance can not
1969 be changed, only the particular hypervisor specific option can be
1970 changed. Therefore, the format of the option parameters has been
1971 simplified to omit the hypervisor name and only contain the comma
1972 separated list of option-value pairs.
1976 gnt-instance modify --hypervisor cdrom=/srv/boot.iso,boot_order=cdrom:network test-instance
1978 gnt-cluster commands
1979 ++++++++++++++++++++
1981 The command for gnt-cluster will be extended to allow setting and
1982 changing the default parameters of the cluster:
1984 - The init command will be extend to support the defaults option to
1985 set the cluster defaults upon cluster initialization.
1986 - The modify command will be added to modify the cluster
1987 parameters. It will support the --defaults option to change the
1992 The generic format of the cluster default setting option is:
1994 --defaults $OPTION=$VALUE[,$OPTION=$VALUE]
1996 :$OPTION: cluster default option, string,
1997 :$VALUE: cluster default option value, string.
1999 Currently, the following cluster default options are defined (open to
2002 :hypervisor: the default hypervisor to use for new instances,
2003 string. Must be a valid hypervisor known to and supported by the
2005 :disksize: the disksize for newly created instance disks, where
2006 applicable. Must be either a positive number, in which case the unit
2007 of megabyte is assumed, or a positive number followed by a supported
2008 magnitude symbol (M for megabyte or G for gigabyte).
2009 :bridge: the default network bridge to use for newly created instance
2010 network interfaces, string. Must be a valid bridge name of a bridge
2011 existing on the node(s).
2013 Hypervisor cluster defaults
2014 +++++++++++++++++++++++++++
2016 The generic format of the hypervisor cluster wide default setting
2019 --hypervisor-defaults $HYPERVISOR:$OPTION=$VALUE[,$OPTION=$VALUE]
2021 :$HYPERVISOR: symbolic name of the hypervisor whose defaults you want
2023 :$OPTION: cluster default option, string,
2024 :$VALUE: cluster default option value, string.
2026 .. vim: set textwidth=72 :