1 Ganeti-htools release notes
2 ===========================
5 Version 0.3.0 (Fri, 04 Feb 2011)
6 --------------------------------
8 A significant release that breaks compatibility with Ganeti versions
9 below 2.4 due to the node group changes. Only the RAPI backend can talk
10 to older clusters, but it is recommended to use this version only with
13 All commands are now multi-group aware (but to various degrees), so
14 allocation, balancing and capacity calculation respects the group layout
15 and will not create “broken” instances by using nodes from different
18 For a regular, single-group cluster, no changes should be directly
19 visible to the users. A multi-group cluster however will change some
22 - hbal will require a target group to operate on (no cluster-wide
24 - evacuation of (DRBD) instances from a node will be restricted to nodes
25 in the same group, as inter-group moves are not implemented yet
26 - capacity, while showing correct data, will not give per-group details
29 There are other changes in this release:
31 - fixed a long-standing bug in hscan related to node memory data
32 - changed the text backend format, which unfortunately invalidates old
34 - error handling improvements, so that invalid input data reports better
36 - the simulation backend changes its syntax, now it takes the allocation
37 policy too, and can generate multiple groups
38 - (internal) man page generation moved to pandoc from hand-written,
39 which is helpful as it can also generate HTML versions
40 - the balancing algorithm has been changed to work in parallel, if the
41 code is linked against the multi-threaded runtime; this gives a very
42 good speedup (~80% on 4 cores, ~60-70% of 12 cores)
44 Version 0.2.8 (Thu, 23 Dec 2010)
45 --------------------------------
49 - fixed balancing function for big clusters, which will improve corner
50 cases where hbal didn't see any solution even though the cluster was
51 obviously not well balanced
52 - fixed exit code of hbal in case of (Luxi) job errors
53 - changed the signal handling in hbal in order to make hbal control
54 easier: instead of synchronising on the count of signals, make SIGINT
55 cause graceful termination, and SIGTERM an immediate one
56 - increased the tag exclusion weight so that it has greater importance
58 - slight improvement to the speed of balancing via algorithm tweaks
61 Version 0.2.7 (Thu, 07 Oct 2010)
62 --------------------------------
66 - fixed the error message for hail multi-evacuation mode
67 - improve evacuation mode for offline secondary nodes (ignore available
72 - add a new option ``-S`` to hbal and hspace that saves the cluster
73 state at the end of the processing in the text format used by the
74 ``-t`` option, for later re-processing
75 - a two new options to hbal, -g and --min-gain-limit, that should help
76 in limiting the number of balances steps with a low gain in the final
78 - hbal, when executing jobs, will now wait for the current jobs to
79 finish at the first stop (e.g. ^C); if the user wants immediate exit,
80 another signal should be sent
81 - added “normalized” physical CPU units in hspace output (NPU), which
82 represents units of physical CPUs free/used, based on the max-cpu
86 Version 0.2.6 (Mon, 26 Jul 2010)
87 --------------------------------
89 Exactly three months since the last release. Many internal changes, plus
90 a couple of important changes in the balancing algorithm.
92 First, the balancing may now introduce N+1 errors, if this solves other,
93 more critical problems. For the moment, this means that moving instances
94 away from offline nodes is allowed even if it creates N+1 errors, and
95 that means evacuation can be done in more cases.
97 Second, the scoring for N+1 has changed. In previous versions, it simply
98 counted the number of failing N+1 nodes, which means moving an instance
99 away from a N+1 failed node (but without the node 'clearing' the N+1
100 status) was not reflected in the cluster score. As such, the balancing
101 algorithm managed to clear N+1 errors only sometimes, since usually it
102 takes more than one move for this, and the first prerequisite move was
103 not 'rewarded' appropriately and thus it was not selected. Now, it is
104 possible to fix many more error cases than before: on a simulated 40
105 node cluster full with instances (symmetrically allocated on all nodes),
106 around five nodes can be evacuated before N+1 errors can be solved,
107 whereas 0.2.5 could evacuate at best one node.
109 There were some other internal changes to the scoring algorithm, such
110 that now the metrics have associated weights, and they are not all of
111 the same importance anymore. As of now, the only change is that offline
112 instances have a higher weight, which should favour proper node
115 Among the other changes:
117 - fixed the hspace KM_POOL_* metrics, which were returned as the final
118 state and not as the delta between the initial and final states
119 - fixed hspace handling of N+1 failing clusters: before, it used to
120 generate a 'fake' response, and the structure of this response was not
121 always in sync with the real responses, leading to missing items;
122 currently it proceeds correctly through the code (skipping the
123 computation), and uses the same display mechanisms as the normal case
124 - fixed hscan exit code for RAPI failures: previously it finished with
125 success even if all the clusters failed, which was creating issues
126 with the live-test script; now it exits with exit code 2 for RAPI
127 failures (unfortunately this is still not optimal as LUXI failures
128 will use exit code 1, the same as the command line)
129 - changed the limit values for CPU/disk, which previously were used
130 optionally, whereas now they are always used; the default cpu ratio
131 limit is now 64 VCPUs per PCPU
132 - changed the internal handling of the short name vs. original
133 (Ganeti-provided) name; now internally we always use the full name,
134 and only in display routines we show the shortened (called 'alias')
135 name; as a result, the -O and --excluded-instances options now accept
136 both the full name and the shortened name
137 - changed internal handling of JSON conversions and errors, such that
138 now we show a better context for failure messages, which should help
139 with diagnosing the malformed message
140 - changed the names for a few node fields, and added some more nodes;
141 this is most likely to help with debugging, and not with regular
143 - changed the node fields option to allow the '+' prefix to mean 'extend
144 the default fields list' rather than start from fresh (similar to
145 Ganeti's implementation)
146 - a few internal changes related to the LUXI protocol implementation,
147 which should make it more safe against potential bugs, one
148 optiomization that should help with large messages, and some patches
149 in preparation for potential expansion of the LUXI backend functionality
151 And finally, many improvements on unittests and the live-test
152 script. Test coverage is much enhanced, and the test infrastructure has
153 better error reporting; this should lead down-the-road to better code
157 Version 0.2.5 (Mon, 26 Apr 2010)
158 --------------------------------
160 Some internal cleanup plus a few user-visible changes:
162 - new option for marking instances as 'do-not-move' during rebalancing
163 - allow ``hscan`` to scan the local cluster via Luxi
164 - add more metrics to ``hspace`` which show the delta between original
165 state and final state better (only valid for tiered allocation)
168 Version 0.2.4 (Mon, 22 Feb 2010)
169 --------------------------------
171 Two improvements for node evacuation:
173 - hbal takes a new parameter ``--evac-mode`` that restricts the
174 instances to be moved to the ones on offline/drained nodes, which
175 should reduce the work done
176 - hail supports the new ``multi-evacuate`` mode of the IAllocator
177 protocol, that will be released in a minor release on the Ganeti 2.1
181 Version 0.2.3 (Thu, 4 Feb 2010)
182 --------------------------------
186 - Fixes selection of secondary node: previously, if the cluster had
187 many N+1 failures, a N+1 failed node could be selected as secondary
188 even if it did not have enough memory to allow the instance to be
189 migrated/failed over to it; this is bad for automated tools, since
190 we can get the cluster in an unhealthy state
191 - Switch the text backend to a single input file, that is generated
192 now by hscan and shouldn't be generated manually via
193 gnt-node/instance list anymore; this allows richer information to be
194 kept in the file, and simplifies a little the internals of the text
198 Version 0.2.2 (Tue, 29 Dec 2009)
199 --------------------------------
201 Small release, 0.2.1 was broken and thus this was released earlier:
203 - Release 0.2.1 broke the LUXI backend due to a typo, fixed
204 - Added a live-test script that should catch errors like the above one
205 in the future (needs a working, non-empty cluster)
206 - Changed RAPI and LUXI backends to treat drained nodes as offline,
207 similar to the IAllocator backend change in 0.2.0 (which was wrongly
208 marked as affecting all backends)
209 - Changed the metrics for offline instances and N1 score from percent to
210 count, in order to increase the priority of evacuations
211 - Added a new metric (offline primary instances) which should fix the
212 evacuation of a offline node in a 2-node cluster
215 Version 0.2.1 (Wed, 2 Dec 2009)
216 --------------------------------
218 - Added instance exclusion defined via instance tags
219 - Fixed the output of hspace to be again parseable from the shell
222 Version 0.2.0 (Tue, 10 Nov 2009)
223 --------------------------------
225 A significant release, with a few new major features:
227 - Added direct execution of the hbal solution when using the Luxi
228 backend; the steps for each instance moves are submitted as a single
229 jobs, and the different jobs are submitted as groups in order to
230 parallelise the execution of moves
231 - Added support for balancing based on dynamic utilisation data for
232 instances, fed in via a text file; by default, all instances are
233 considered equal and this change also improves the equalisation of
234 secondary instances per node
235 - Added support for tiered capacity calculation in hspace, where we
236 start from a maximum instance spec and decrease the spec when we run
237 out of resources; this should give a better measure of available
238 capacity on 'fragmented' clusters; this is done separately from the
239 current fixed-mode computation
241 Also there have been many minor improvements:
243 - Added option for showing instances (“--print-instances”), similar to
244 the print nodes option
245 - Added support for customising the node list via an argument to the
246 print nodes option in the form of a comma-separated list of field
247 names; currently the field names are not documented, expecting further
248 changes in a next release
249 - Enhanced the error reporting in the Luxi and Rapi backends
250 - Changed the handling of drained nodes, now being treated the same as
251 offline nodes, for Ganeti 2.0.4+ compatibility
252 - A number of internal changes, simplifying code and merging some
254 - Simplify the build system in relation to creation of archives
257 Version 0.1.8 (Tue, 29 Sep 2009)
258 --------------------------------
260 - Brown-paper-bag release fixing haddock issues
263 Version 0.1.7 (Mon, 28 Sep 2009)
264 --------------------------------
266 - Fixed a bug in the Luxi backend for big responses
267 - Fixed test suite exit code in presence of test failures
268 - Changed the migrate operation to run instead failover for instances
269 which were marked as not running in the input data (this could have
270 been changed since then, but it's better than today's always migrate)
271 - Added support for 'cheap' moves only (only migrate/failover) in
273 - Added support for building without curl (thus no RAPI backend)
276 Version 0.1.6 (Wed, 19 Aug 2009)
277 --------------------------------
279 - Added support for Luxi (the native Ganeti protocol)
280 - Added support for simulated clusters (for hspace only)
281 - Added timeouts for the RAPI backend
282 - Fixed a few inconsistencies in the command line handling
283 - Fixed handling of errors while loading data
284 - The 'network' is a new dependency due to the Luxi addition
287 Version 0.1.5 (Thu, 09 Jul 2009)
288 --------------------------------
290 - Removed obsolete hn1 program; this allowed removal of a lot of
292 - Lots of changes in hspace: the output now is a shell fragment in order
293 for script to source it or parse it easier; added failure reasons;
294 optimised to use less memory for large clusters
295 - Optimized the scoring algorithm (used by all tools) so that now
296 computations should be faster
299 Version 0.1.4 (Tue, 16 Jun 2009)
300 --------------------------------
302 - Added CPU count/ratio of virtual-to-physical CPUs to the cluster
303 scoring methods; this means that now the balancer, the iallocator
304 plugin and so on will try to keep the VCPU-to-PCPU ratio equal across
306 - Fixed some hscan bugs
307 - Fixed the way iallocator reads the total disk size (was broken and it
308 was always falling back to summing the disk sizes)
309 - Internals: fixed most compile-time warnings
312 Version 0.1.3 (Fri, 05 Jun 2009)
313 --------------------------------
315 - Fix a bug in the ReplacePrimary instance moves, affecting most of the
319 Version 0.1.2 (Tue, 02 Jun 2009)
320 --------------------------------
322 - Add a new program, “hspace”, which computes the free space on a
323 cluster (based on a given instance spec)
324 - Improvements in API docs and partially in the user docs
325 - Started adding unittests
328 Version 0.1.1 (Tue, 26 May 2009)
329 --------------------------------
331 - Add a new program, “hail”, which is an iallocator plugin and can
332 allocate/relocate instances
333 - Experimental support for non-mirrored instances (hail supports them,
334 hbal should no longer abort when it finds such instances and simply
336 - The RAPI port and/or scheme can be overriden now, and even “file://”
337 schemes can be used if the message body has been saved under the
339 - Lots of code reorganization, esp. rewritten loading pipeline
340 - Better data checking and better error messages in case validation
341 fails; tools now consider nodes with error in input data (‘?’ returned
342 by ganeti) as offline
343 - Small enhancement to the makefile for simpler packaging
346 Version 0.1.0 (Tue, 19 May 2009)
347 --------------------------------
349 - Drop compatibility with Ganeti 1.2
350 - Add a new minimum score option (with a very low default), should help
351 with very good clusters (but is still not optimal)
352 - Add a --quiet option to hbal
353 - Add support for reading offline nodes directly from the cluster
356 Version 0.0.8 (Tue, 21 Apr 2009)
357 --------------------------------
359 - hbal: prevent mismatches in wrong node names being passed to -O, by
360 aborting in this case
361 - add the ability to write the commands (-C) to a script via (-C<file>),
362 so that it can be later executed directly; this has also changed the
363 commands to include the ncessary -f flags to skip confirmations
364 - add checks for extra argument in hbal and hn1, so that unintended
366 - raise the accepted “missing” memory limit to 512MB, to cover usual Xen
370 Version 0.0.7 (Mon, 23 Mar 2009)
371 --------------------------------
373 - added support for offline nodes, which are not used as targets for
374 instance relocation and if they hold instances the hbal algorithm will
375 attempt to relocate these away
376 - added support for offline instances, which now will no longer skew the
377 free memory estimation of nodes; the algorithm will no longer create
378 conditions for N+1 failures when such instances are later started
379 - implemented a complete model of node resources, in order to prevent an
380 unintended re-occurrence of cases like the offline instance were we
381 miscalculate some node resource; this gives warning now in case the
382 node reported free disk or free memory deviates by more than a set
383 amount from the expected value
384 - a new tool *hscan* that can generate the input text-file for the other
385 tools by collection via RAPI
386 - some small changes to the build system to make it more friendly; also
387 included the generated documentation in the source archive
390 Version 0.0.6 (Mon, 16 Mar 2009)
391 --------------------------------
393 - re-factored the hbal algorithm to make it stable in the sense that it
394 gives the same solution when restarted from the middle; barring
395 rounding of disk/memory and incomplete reporting from Ganeti (for
396 1.2), it should be now feasible to rely on its output without
397 generating moves ad infinitum
398 - the hbal algorithm now uses two more variables: the node N+1 failures
399 and the amount of reserved memory; the first of which tries to ‘fix’
400 the N+1 status, the latter tries to distribute secondaries more
402 - the hbal algorithm now uses two more moves at each step:
403 replace+failover and failover+replace (besides the original failover,
404 replace, and failover+replace+failover)
405 - slightly changed the build system to embed GIT version/tags into the
406 binaries so that we know for a binary from which tree it was done,
407 either via ‘--version’ or via “strings hbal|grep version”
408 - changed the solution list and in general the hbal output to be more
409 clear by default, and changed “gnt-instance failover” to “gnt-instance
411 - added man pages for the two binaries
414 Version 0.0.5 (Mon, 09 Mar 2009)
415 --------------------------------
417 - a few small improvements for hbal (possibly undone by later changes),
418 hbal is now quite faster
419 - fix documentation building
420 - allow hbal to work on non N+1 compliant clusters, but without
421 guarantees that the end cluster will be compliant; in any case, this
422 should give a smaller number of nodes that are not compliant if the
423 cluster state permits it
424 - strip common domain suffix from nodes and instances, so that output is
425 shorter and hopefully clearer
428 Version 0.0.4 (Sun, 15 Feb 2009)
429 --------------------------------
431 - better balancing algorithm in hbal
432 - implemented an RAPI collector, now the cluster data can be gathered
433 automatically via RAPI and doesn't need manual export of node and
437 Version 0.0.3 (Wed, 28 Jan 2009)
438 --------------------------------
440 - initial release of the hbal, a cluster rebalancing tool
441 - input data format changed due to hbal requirements
444 Version 0.0.2 (Tue, 06 Jan 2009)
445 --------------------------------
447 - fix handling of some common cases (cluster N+1 compliant from the
448 start, too big depth given, failure to compute solution)
449 - add option to print the needed command list for reaching the proposed
453 Version 0.0.1 (Tue, 06 Jan 2009)
454 --------------------------------
456 - initial release of hn1 tool
458 .. vim: set textwidth=72 :
projects / ganeti-local / blob