Statistics
| Branch: | Tag: | Revision:

root / src / OLD-NEWS @ f0e4b2a4

History | View | Annotate | Download (19 kB)

1
Ganeti-htools release notes
2
===========================
3

    
4

    
5
**Note**: After version 0.3.1, the htools sources have been integrated
6
into the ganeti core repository, and released together with the ganeti
7
releases. Thus this NEWS file is obsolete.
8

    
9
Version 0.3.1 (Fri, 11 Mar 2011)
10
--------------------------------
11

    
12
Minor bugfix release:
13

    
14
- Fixed source archive generation: the hscolour.css file was an invalid
15
  symlink, and the man pages were not correctly timestamped (leading to
16
  unneeded build-time rebuilds)
17
- Improved the Luxi backend to show which attribute fails parsing
18
- Small improvements to the man pages, and also ship the HTML version of
19
  man pages in the source archive
20

    
21

    
22
Version 0.3.0 (Fri, 04 Feb 2011)
23
--------------------------------
24

    
25
A significant release that breaks compatibility with Ganeti versions
26
below 2.4 due to the node group changes. Only the RAPI backend can talk
27
to older clusters, but it is recommended to use this version only with
28
Ganeti 2.4.
29

    
30
All commands are now multi-group aware (but to various degrees), so
31
allocation, balancing and capacity calculation respects the group layout
32
and will not create “broken” instances by using nodes from different
33
groups.
34

    
35
For a regular, single-group cluster, no changes should be directly
36
visible to the users. A multi-group cluster however will change some
37
things slightly:
38

    
39
- hbal will require a target group to operate on (no cluster-wide
40
  balancing yet)
41
- evacuation of (DRBD) instances from a node will be restricted to nodes
42
  in the same group, as inter-group moves are not implemented yet
43
- capacity, while showing correct data, will not give per-group details
44
  yet
45

    
46
There are other changes in this release:
47

    
48
- fixed a long-standing bug in hscan related to node memory data
49
- changed the text backend format, which unfortunately invalidates old
50
  files
51
- error handling improvements, so that invalid input data reports better
52
  where the error is
53
- the simulation backend changes its syntax, now it takes the allocation
54
  policy too, and can generate multiple groups
55
- (internal) man page generation moved to pandoc from hand-written,
56
  which is helpful as it can also generate HTML versions
57
- the balancing algorithm has been changed to work in parallel, if the
58
  code is linked against the multi-threaded runtime; this gives a very
59
  good speedup (~80% on 4 cores, ~60-70% of 12 cores)
60

    
61
Version 0.2.8 (Thu, 23 Dec 2010)
62
--------------------------------
63

    
64
A bug fix release:
65

    
66
- fixed balancing function for big clusters, which will improve corner
67
  cases where hbal didn't see any solution even though the cluster was
68
  obviously not well balanced
69
- fixed exit code of hbal in case of (Luxi) job errors
70
- changed the signal handling in hbal in order to make hbal control
71
  easier: instead of synchronising on the count of signals, make SIGINT
72
  cause graceful termination, and SIGTERM an immediate one
73
- increased the tag exclusion weight so that it has greater importance
74
  during the balancing
75
- slight improvement to the speed of balancing via algorithm tweaks
76

    
77

    
78
Version 0.2.7 (Thu, 07 Oct 2010)
79
--------------------------------
80

    
81
Bug fixes:
82

    
83
- fixed the error message for hail multi-evacuation mode
84
- improve evacuation mode for offline secondary nodes (ignore available
85
  memory)
86

    
87
New features:
88

    
89
- add a new option ``-S`` to hbal and hspace that saves the cluster
90
  state at the end of the processing in the text format used by the
91
  ``-t`` option, for later re-processing
92
- a two new options to hbal, -g and --min-gain-limit, that should help
93
  in limiting the number of balances steps with a low gain in the final
94
  stages
95
- hbal, when executing jobs, will now wait for the current jobs to
96
  finish at the first stop (e.g. ^C); if the user wants immediate exit,
97
  another signal should be sent
98
- added “normalized” physical CPU units in hspace output (NPU), which
99
  represents units of physical CPUs free/used, based on the max-cpu
100
  ratio
101

    
102

    
103
Version 0.2.6 (Mon, 26 Jul 2010)
104
--------------------------------
105

    
106
Exactly three months since the last release. Many internal changes, plus
107
a couple of important changes in the balancing algorithm.
108

    
109
First, the balancing may now introduce N+1 errors, if this solves other,
110
more critical problems. For the moment, this means that moving instances
111
away from offline nodes is allowed even if it creates N+1 errors, and
112
that means evacuation can be done in more cases.
113

    
114
Second, the scoring for N+1 has changed. In previous versions, it simply
115
counted the number of failing N+1 nodes, which means moving an instance
116
away from a N+1 failed node (but without the node 'clearing' the N+1
117
status) was not reflected in the cluster score. As such, the balancing
118
algorithm managed to clear N+1 errors only sometimes, since usually it
119
takes more than one move for this, and the first prerequisite move was
120
not 'rewarded' appropriately and thus it was not selected. Now, it is
121
possible to fix many more error cases than before: on a simulated 40
122
node cluster full with instances (symmetrically allocated on all nodes),
123
around five nodes can be evacuated before N+1 errors can be solved,
124
whereas 0.2.5 could evacuate at best one node.
125

    
126
There were some other internal changes to the scoring algorithm, such
127
that now the metrics have associated weights, and they are not all of
128
the same importance anymore. As of now, the only change is that offline
129
instances have a higher weight, which should favour proper node
130
evacuations.
131

    
132
Among the other changes:
133

    
134
- fixed the hspace KM_POOL_* metrics, which were returned as the final
135
  state and not as the delta between the initial and final states
136
- fixed hspace handling of N+1 failing clusters: before, it used to
137
  generate a 'fake' response, and the structure of this response was not
138
  always in sync with the real responses, leading to missing items;
139
  currently it proceeds correctly through the code (skipping the
140
  computation), and uses the same display mechanisms as the normal case
141
- fixed hscan exit code for RAPI failures: previously it finished with
142
  success even if all the clusters failed, which was creating issues
143
  with the live-test script; now it exits with exit code 2 for RAPI
144
  failures (unfortunately this is still not optimal as LUXI failures
145
  will use exit code 1, the same as the command line)
146
- changed the limit values for CPU/disk, which previously were used
147
  optionally, whereas now they are always used; the default cpu ratio
148
  limit is now 64 VCPUs per PCPU
149
- changed the internal handling of the short name vs. original
150
  (Ganeti-provided) name; now internally we always use the full name,
151
  and only in display routines we show the shortened (called 'alias')
152
  name; as a result, the -O and --excluded-instances options now accept
153
  both the full name and the shortened name
154
- changed internal handling of JSON conversions and errors, such that
155
  now we show a better context for failure messages, which should help
156
  with diagnosing the malformed message
157
- changed the names for a few node fields, and added some more nodes;
158
  this is most likely to help with debugging, and not with regular
159
  operation though
160
- changed the node fields option to allow the '+' prefix to mean 'extend
161
  the default fields list' rather than start from fresh (similar to
162
  Ganeti's implementation)
163
- a few internal changes related to the LUXI protocol implementation,
164
  which should make it more safe against potential bugs, one
165
  optiomization that should help with large messages, and some patches
166
  in preparation for potential expansion of the LUXI backend functionality
167

    
168
And finally, many improvements on unittests and the live-test
169
script. Test coverage is much enhanced, and the test infrastructure has
170
better error reporting; this should lead down-the-road to better code
171
and fewer bugs…
172

    
173

    
174
Version 0.2.5 (Mon, 26 Apr 2010)
175
--------------------------------
176

    
177
Some internal cleanup plus a few user-visible changes:
178

    
179
- new option for marking instances as 'do-not-move' during rebalancing
180
- allow ``hscan`` to scan the local cluster via Luxi
181
- add more metrics to ``hspace`` which show the delta between original
182
  state and final state better (only valid for tiered allocation)
183

    
184

    
185
Version 0.2.4 (Mon, 22 Feb 2010)
186
--------------------------------
187

    
188
Two improvements for node evacuation:
189

    
190
- hbal takes a new parameter ``--evac-mode`` that restricts the
191
  instances to be moved to the ones on offline/drained nodes, which
192
  should reduce the work done
193
- hail supports the new ``multi-evacuate`` mode of the IAllocator
194
  protocol, that will be released in a minor release on the Ganeti 2.1
195
  branch
196

    
197

    
198
Version 0.2.3 (Thu,  4 Feb 2010)
199
--------------------------------
200

    
201
A small release:
202

    
203
- Fixes selection of secondary node: previously, if the cluster had
204
  many N+1 failures, a N+1 failed node could be selected as secondary
205
  even if it did not have enough memory to allow the instance to be
206
  migrated/failed over to it; this is bad for automated tools, since
207
  we can get the cluster in an unhealthy state
208
- Switch the text backend to a single input file, that is generated
209
  now by hscan and shouldn't be generated manually via
210
  gnt-node/instance list anymore; this allows richer information to be
211
  kept in the file, and simplifies a little the internals of the text
212
  backend
213

    
214

    
215
Version 0.2.2 (Tue, 29 Dec 2009)
216
--------------------------------
217

    
218
Small release, 0.2.1 was broken and thus this was released earlier:
219

    
220
- Release 0.2.1 broke the LUXI backend due to a typo, fixed
221
- Added a live-test script that should catch errors like the above one
222
  in the future (needs a working, non-empty cluster)
223
- Changed RAPI and LUXI backends to treat drained nodes as offline,
224
  similar to the IAllocator backend change in 0.2.0 (which was wrongly
225
  marked as affecting all backends)
226
- Changed the metrics for offline instances and N1 score from percent to
227
  count, in order to increase the priority of evacuations
228
- Added a new metric (offline primary instances) which should fix the
229
  evacuation of a offline node in a 2-node cluster
230

    
231

    
232
Version 0.2.1 (Wed,  2 Dec 2009)
233
--------------------------------
234

    
235
- Added instance exclusion defined via instance tags
236
- Fixed the output of hspace to be again parseable from the shell
237

    
238

    
239
Version 0.2.0 (Tue, 10 Nov 2009)
240
--------------------------------
241

    
242
A significant release, with a few new major features:
243

    
244
- Added direct execution of the hbal solution when using the Luxi
245
  backend; the steps for each instance moves are submitted as a single
246
  jobs, and the different jobs are submitted as groups in order to
247
  parallelise the execution of moves
248
- Added support for balancing based on dynamic utilisation data for
249
  instances, fed in via a text file; by default, all instances are
250
  considered equal and this change also improves the equalisation of
251
  secondary instances per node
252
- Added support for tiered capacity calculation in hspace, where we
253
  start from a maximum instance spec and decrease the spec when we run
254
  out of resources; this should give a better measure of available
255
  capacity on 'fragmented' clusters; this is done separately from the
256
  current fixed-mode computation
257

    
258
Also there have been many minor improvements:
259

    
260
- Added option for showing instances (“--print-instances”), similar to
261
  the print nodes option
262
- Added support for customising the node list via an argument to the
263
  print nodes option in the form of a comma-separated list of field
264
  names; currently the field names are not documented, expecting further
265
  changes in a next release
266
- Enhanced the error reporting in the Luxi and Rapi backends
267
- Changed the handling of drained nodes, now being treated the same as
268
  offline nodes, for Ganeti 2.0.4+ compatibility
269
- A number of internal changes, simplifying code and merging some
270
  disparate functions
271
- Simplify the build system in relation to creation of archives
272

    
273

    
274
Version 0.1.8 (Tue, 29 Sep 2009)
275
--------------------------------
276

    
277
- Brown-paper-bag release fixing haddock issues
278

    
279

    
280
Version 0.1.7 (Mon, 28 Sep 2009)
281
--------------------------------
282

    
283
- Fixed a bug in the Luxi backend for big responses
284
- Fixed test suite exit code in presence of test failures
285
- Changed the migrate operation to run instead failover for instances
286
  which were marked as not running in the input data (this could have
287
  been changed since then, but it's better than today's always migrate)
288
- Added support for 'cheap' moves only (only migrate/failover) in
289
  balancing
290
- Added support for building without curl (thus no RAPI backend)
291

    
292

    
293
Version 0.1.6 (Wed, 19 Aug 2009)
294
--------------------------------
295

    
296
- Added support for Luxi (the native Ganeti protocol)
297
- Added support for simulated clusters (for hspace only)
298
- Added timeouts for the RAPI backend
299
- Fixed a few inconsistencies in the command line handling
300
- Fixed handling of errors while loading data
301
- The 'network' is a new dependency due to the Luxi addition
302

    
303

    
304
Version 0.1.5 (Thu, 09 Jul 2009)
305
--------------------------------
306

    
307
- Removed obsolete hn1 program; this allowed removal of a lot of
308
  supporting code
309
- Lots of changes in hspace: the output now is a shell fragment in order
310
  for script to source it or parse it easier; added failure reasons;
311
  optimised to use less memory for large clusters
312
- Optimized the scoring algorithm (used by all tools) so that now
313
  computations should be faster
314

    
315

    
316
Version 0.1.4 (Tue, 16 Jun 2009)
317
--------------------------------
318

    
319
- Added CPU count/ratio of virtual-to-physical CPUs to the cluster
320
  scoring methods; this means that now the balancer, the iallocator
321
  plugin and so on will try to keep the VCPU-to-PCPU ratio equal across
322
  the cluster
323
- Fixed some hscan bugs
324
- Fixed the way iallocator reads the total disk size (was broken and it
325
  was always falling back to summing the disk sizes)
326
- Internals: fixed most compile-time warnings
327

    
328

    
329
Version 0.1.3 (Fri, 05 Jun 2009)
330
--------------------------------
331

    
332
- Fix a bug in the ReplacePrimary instance moves, affecting most of the
333
  tools
334

    
335

    
336
Version 0.1.2 (Tue, 02 Jun 2009)
337
--------------------------------
338

    
339
- Add a new program, “hspace”, which computes the free space on a
340
  cluster (based on a given instance spec)
341
- Improvements in API docs and partially in the user docs
342
- Started adding unittests
343

    
344

    
345
Version 0.1.1 (Tue, 26 May 2009)
346
--------------------------------
347

    
348
- Add a new program, “hail”, which is an iallocator plugin and can
349
  allocate/relocate instances
350
- Experimental support for non-mirrored instances (hail supports them,
351
  hbal should no longer abort when it finds such instances and simply
352
  ignore them)
353
- The RAPI port and/or scheme can be overriden now, and even “file://”
354
  schemes can be used if the message body has been saved under the
355
  appropriate name
356
- Lots of code reorganization, esp. rewritten loading pipeline
357
- Better data checking and better error messages in case validation
358
  fails; tools now consider nodes with error in input data (‘?’ returned
359
  by ganeti) as offline
360
- Small enhancement to the makefile for simpler packaging
361

    
362

    
363
Version 0.1.0 (Tue, 19 May 2009)
364
--------------------------------
365

    
366
- Drop compatibility with Ganeti 1.2
367
- Add a new minimum score option (with a very low default), should help
368
  with very good clusters (but is still not optimal)
369
- Add a --quiet option to hbal
370
- Add support for reading offline nodes directly from the cluster
371

    
372

    
373
Version 0.0.8 (Tue, 21 Apr 2009)
374
--------------------------------
375

    
376
- hbal: prevent mismatches in wrong node names being passed to -O, by
377
  aborting in this case
378
- add the ability to write the commands (-C) to a script via (-C<file>),
379
  so that it can be later executed directly; this has also changed the
380
  commands to include the ncessary -f flags to skip confirmations
381
- add checks for extra argument in hbal and hn1, so that unintended
382
  errors are catched
383
- raise the accepted “missing” memory limit to 512MB, to cover usual Xen
384
  reservations
385

    
386

    
387
Version 0.0.7 (Mon, 23 Mar 2009)
388
--------------------------------
389

    
390
- added support for offline nodes, which are not used as targets for
391
  instance relocation and if they hold instances the hbal algorithm will
392
  attempt to relocate these away
393
- added support for offline instances, which now will no longer skew the
394
  free memory estimation of nodes; the algorithm will no longer create
395
  conditions for N+1 failures when such instances are later started
396
- implemented a complete model of node resources, in order to prevent an
397
  unintended re-occurrence of cases like the offline instance were we
398
  miscalculate some node resource; this gives warning now in case the
399
  node reported free disk or free memory deviates by more than a set
400
  amount from the expected value
401
- a new tool *hscan* that can generate the input text-file for the other
402
  tools by collection via RAPI
403
- some small changes to the build system to make it more friendly; also
404
  included the generated documentation in the source archive
405

    
406

    
407
Version 0.0.6 (Mon, 16 Mar 2009)
408
--------------------------------
409

    
410
- re-factored the hbal algorithm to make it stable in the sense that it
411
  gives the same solution when restarted from the middle; barring
412
  rounding of disk/memory and incomplete reporting from Ganeti (for
413
  1.2), it should be now feasible to rely on its output without
414
  generating moves ad infinitum
415
- the hbal algorithm now uses two more variables: the node N+1 failures
416
  and the amount of reserved memory; the first of which tries to ‘fix’
417
  the N+1 status, the latter tries to distribute secondaries more
418
  equally
419
- the hbal algorithm now uses two more moves at each step:
420
  replace+failover and failover+replace (besides the original failover,
421
  replace, and failover+replace+failover)
422
- slightly changed the build system to embed GIT version/tags into the
423
  binaries so that we know for a binary from which tree it was done,
424
  either via ‘--version’ or via “strings hbal|grep version”
425
- changed the solution list and in general the hbal output to be more
426
  clear by default, and changed “gnt-instance failover” to “gnt-instance
427
  migrate”
428
- added man pages for the two binaries
429

    
430

    
431
Version 0.0.5 (Mon, 09 Mar 2009)
432
--------------------------------
433

    
434
- a few small improvements for hbal (possibly undone by later changes),
435
  hbal is now quite faster
436
- fix documentation building
437
- allow hbal to work on non N+1 compliant clusters, but without
438
  guarantees that the end cluster will be compliant; in any case, this
439
  should give a smaller number of nodes that are not compliant if the
440
  cluster state permits it
441
- strip common domain suffix from nodes and instances, so that output is
442
  shorter and hopefully clearer
443

    
444

    
445
Version 0.0.4 (Sun, 15 Feb 2009)
446
--------------------------------
447

    
448
- better balancing algorithm in hbal
449
- implemented an RAPI collector, now the cluster data can be gathered
450
  automatically via RAPI and doesn't need manual export of node and
451
  instance list
452

    
453

    
454
Version 0.0.3 (Wed, 28 Jan 2009)
455
--------------------------------
456

    
457
- initial release of the hbal, a cluster rebalancing tool
458
- input data format changed due to hbal requirements
459

    
460

    
461
Version 0.0.2 (Tue, 06 Jan 2009)
462
--------------------------------
463

    
464
- fix handling of some common cases (cluster N+1 compliant from the
465
  start, too big depth given, failure to compute solution)
466
- add option to print the needed command list for reaching the proposed
467
  solution
468

    
469

    
470
Version 0.0.1 (Tue, 06 Jan 2009)
471
--------------------------------
472

    
473
- initial release of hn1 tool
474

    
475
.. vim: set textwidth=72 :
476
.. Local Variables:
477
.. mode: rst
478
.. fill-column: 72
479
.. End: