root / hbal.1 @ 5f715404
History | View | Annotate | Download (26.4 kB)
1 |
.TH HBAL 1 2009-03-23 htools "Ganeti H-tools" |
---|---|
2 |
.SH NAME |
3 |
hbal \- Cluster balancer for Ganeti |
4 |
|
5 |
.SH SYNOPSIS |
6 |
.B hbal |
7 |
.B "[backend options...]" |
8 |
.B "[algorithm options...]" |
9 |
.B "[reporting options...]" |
10 |
|
11 |
.B hbal |
12 |
.B --version |
13 |
|
14 |
.TP |
15 |
Backend options: |
16 |
.BI "[ -m " cluster " ]" |
17 |
| |
18 |
.BI "[ -L[" path "] [-X]]" |
19 |
| |
20 |
.BI "[ -t " data-file " ]" |
21 |
|
22 |
.TP |
23 |
Algorithm options: |
24 |
.BI "[ --max-cpu " cpu-ratio " ]" |
25 |
.BI "[ --min-disk " disk-ratio " ]" |
26 |
.BI "[ -l " limit " ]" |
27 |
.BI "[ -e " score " ]" |
28 |
.BI "[ -g " delta " ] [ --min-gain-limit " threshold " ]" |
29 |
.BI "[ -O " name... " ]" |
30 |
.B "[ --no-disk-moves ]" |
31 |
.BI "[ -U " util-file " ]" |
32 |
.B "[ --evac-mode ]" |
33 |
.BI "[ --exclude-instances " inst... " ]" |
34 |
|
35 |
.TP |
36 |
Reporting options: |
37 |
.BI "[ -C[" file "] ]" |
38 |
.BI "[ -p[" fields "] ]" |
39 |
.B "[ --print-instances ]" |
40 |
.B "[ -o ]" |
41 |
.B "[ -v... | -q ]" |
42 |
|
43 |
|
44 |
.SH DESCRIPTION |
45 |
hbal is a cluster balancer that looks at the current state of the |
46 |
cluster (nodes with their total and free disk, memory, etc.) and |
47 |
instance placement and computes a series of steps designed to bring |
48 |
the cluster into a better state. |
49 |
|
50 |
The algorithm used is designed to be stable (i.e. it will give you the |
51 |
same results when restarting it from the middle of the solution) and |
52 |
reasonably fast. It is not, however, designed to be a perfect |
53 |
algorithm \(em it is possible to make it go into a corner from which |
54 |
it can find no improvement, because it looks only one "step" ahead. |
55 |
|
56 |
By default, the program will show the solution incrementally as it is |
57 |
computed, in a somewhat cryptic format; for getting the actual Ganeti |
58 |
command list, use the \fB-C\fR option. |
59 |
|
60 |
.SS ALGORITHM |
61 |
|
62 |
The program works in independent steps; at each step, we compute the |
63 |
best instance move that lowers the cluster score. |
64 |
|
65 |
The possible move type for an instance are combinations of |
66 |
failover/migrate and replace-disks such that we change one of the |
67 |
instance nodes, and the other one remains (but possibly with changed |
68 |
role, e.g. from primary it becomes secondary). The list is: |
69 |
.RS 4 |
70 |
.TP 3 |
71 |
\(em |
72 |
failover (f) |
73 |
.TP |
74 |
\(em |
75 |
replace secondary (r) |
76 |
.TP |
77 |
\(em |
78 |
replace primary, a composite move (f, r, f) |
79 |
.TP |
80 |
\(em |
81 |
failover and replace secondary, also composite (f, r) |
82 |
.TP |
83 |
\(em |
84 |
replace secondary and failover, also composite (r, f) |
85 |
.RE |
86 |
|
87 |
We don't do the only remaining possibility of replacing both nodes |
88 |
(r,f,r,f or the equivalent f,r,f,r) since these move needs an |
89 |
exhaustive search over both candidate primary and secondary nodes, and |
90 |
is O(n*n) in the number of nodes. Furthermore, it doesn't seems to |
91 |
give better scores but will result in more disk replacements. |
92 |
|
93 |
.SS PLACEMENT RESTRICTIONS |
94 |
|
95 |
At each step, we prevent an instance move if it would cause: |
96 |
|
97 |
.RS 4 |
98 |
.TP 3 |
99 |
\(em |
100 |
a node to go into N+1 failure state |
101 |
.TP |
102 |
\(em |
103 |
an instance to move onto an offline node (offline nodes are either |
104 |
read from the cluster or declared with \fI-O\fR) |
105 |
.TP |
106 |
\(em |
107 |
an exclusion-tag based conflict (exclusion tags are read from the |
108 |
cluster and/or defined via the \fI--exclusion-tags\fR option) |
109 |
.TP |
110 |
\(em |
111 |
a max vcpu/pcpu ratio to be exceeded (configured via \fI--max-cpu\fR) |
112 |
.TP |
113 |
\(em |
114 |
min disk free percentage to go below the configured limit (configured |
115 |
via \fI--min-disk\fR) |
116 |
|
117 |
.SS CLUSTER SCORING |
118 |
|
119 |
As said before, the algorithm tries to minimise the cluster score at |
120 |
each step. Currently this score is computed as a sum of the following |
121 |
components: |
122 |
.RS 4 |
123 |
.TP 3 |
124 |
\(em |
125 |
standard deviation of the percent of free memory |
126 |
.TP |
127 |
\(em |
128 |
standard deviation of the percent of reserved memory |
129 |
.TP |
130 |
\(em |
131 |
standard deviation of the percent of free disk |
132 |
.TP |
133 |
\(em |
134 |
count of nodes failing N+1 check |
135 |
.TP |
136 |
\(em |
137 |
count of instances living (either as primary or secondary) on |
138 |
offline nodes |
139 |
.TP |
140 |
\(em |
141 |
count of instances living (as primary) on offline nodes; this differs |
142 |
from the above metric by helping failover of such instances in 2-node |
143 |
clusters |
144 |
.TP |
145 |
\(em |
146 |
standard deviation of the ratio of virtual-to-physical cpus (for |
147 |
primary instances of the node) |
148 |
.TP |
149 |
\(em |
150 |
standard deviation of the dynamic load on the nodes, for cpus, |
151 |
memory, disk and network |
152 |
.RE |
153 |
|
154 |
The free memory and free disk values help ensure that all nodes are |
155 |
somewhat balanced in their resource usage. The reserved memory helps |
156 |
to ensure that nodes are somewhat balanced in holding secondary |
157 |
instances, and that no node keeps too much memory reserved for |
158 |
N+1. And finally, the N+1 percentage helps guide the algorithm towards |
159 |
eliminating N+1 failures, if possible. |
160 |
|
161 |
Except for the N+1 failures and offline instances counts, we use the |
162 |
standard deviation since when used with values within a fixed range |
163 |
(we use percents expressed as values between zero and one) it gives |
164 |
consistent results across all metrics (there are some small issues |
165 |
related to different means, but it works generally well). The 'count' |
166 |
type values will have higher score and thus will matter more for |
167 |
balancing; thus these are better for hard constraints (like evacuating |
168 |
nodes and fixing N+1 failures). For example, the offline instances |
169 |
count (i.e. the number of instances living on offline nodes) will |
170 |
cause the algorithm to actively move instances away from offline |
171 |
nodes. This, coupled with the restriction on placement given by |
172 |
offline nodes, will cause evacuation of such nodes. |
173 |
|
174 |
The dynamic load values need to be read from an external file (Ganeti |
175 |
doesn't supply them), and are computed for each node as: sum of |
176 |
primary instance cpu load, sum of primary instance memory load, sum of |
177 |
primary and secondary instance disk load (as DRBD generates write load |
178 |
on secondary nodes too in normal case and in degraded scenarios also |
179 |
read load), and sum of primary instance network load. An example of |
180 |
how to generate these values for input to hbal would be to track "xm |
181 |
list" for instance over a day and by computing the delta of the cpu |
182 |
values, and feed that via the \fI-U\fR option for all instances (and |
183 |
keep the other metrics as one). For the algorithm to work, all that is |
184 |
needed is that the values are consistent for a metric across all |
185 |
instances (e.g. all instances use cpu% to report cpu usage, and not |
186 |
something related to number of CPU seconds used if the CPUs are |
187 |
different), and that they are normalised to between zero and one. Note |
188 |
that it's recommended to not have zero as the load value for any |
189 |
instance metric since then secondary instances are not well balanced. |
190 |
|
191 |
On a perfectly balanced cluster (all nodes the same size, all |
192 |
instances the same size and spread across the nodes equally), the |
193 |
values for all metrics would be zero. This doesn't happen too often in |
194 |
practice :) |
195 |
|
196 |
.SS OFFLINE INSTANCES |
197 |
|
198 |
Since current Ganeti versions do not report the memory used by offline |
199 |
(down) instances, ignoring the run status of instances will cause |
200 |
wrong calculations. For this reason, the algorithm subtracts the |
201 |
memory size of down instances from the free node memory of their |
202 |
primary node, in effect simulating the startup of such instances. |
203 |
|
204 |
.SS EXCLUSION TAGS |
205 |
|
206 |
The exclusion tags mechanism is designed to prevent instances which |
207 |
run the same workload (e.g. two DNS servers) to land on the same node, |
208 |
which would make the respective node a SPOF for the given service. |
209 |
|
210 |
It works by tagging instances with certain tags and then building |
211 |
exclusion maps based on these. Which tags are actually used is |
212 |
configured either via the command line (option \fI--exclusion-tags\fR) |
213 |
or via adding them to the cluster tags: |
214 |
|
215 |
.TP |
216 |
.B --exclusion-tags=a,b |
217 |
This will make all instance tags of the form \fIa:*\fR, \fIb:*\fR be |
218 |
considered for the exclusion map |
219 |
|
220 |
.TP |
221 |
cluster tags \fBhtools:iextags:a\fR, \fBhtools:iextags:b\fR |
222 |
This will make instance tags \fIa:*\fR, \fIb:*\fR be considered for |
223 |
the exclusion map. More precisely, the suffix of cluster tags starting |
224 |
with \fBhtools:iextags:\fR will become the prefix of the exclusion |
225 |
tags. |
226 |
|
227 |
.P |
228 |
Both the above forms mean that two instances both having (e.g.) the |
229 |
tag \fIa:foo\fR or \fIb:bar\fR won't end on the same node. |
230 |
|
231 |
.SH OPTIONS |
232 |
The options that can be passed to the program are as follows: |
233 |
.TP |
234 |
.B -C, --print-commands |
235 |
Print the command list at the end of the run. Without this, the |
236 |
program will only show a shorter, but cryptic output. |
237 |
|
238 |
Note that the moves list will be split into independent steps, called |
239 |
"jobsets", but only for visual inspection, not for actually |
240 |
parallelisation. It is not possible to parallelise these directly when |
241 |
executed via "gnt-instance" commands, since a compound command |
242 |
(e.g. failover and replace\-disks) must be executed serially. Parallel |
243 |
execution is only possible when using the Luxi backend and the |
244 |
\fI-L\fR option. |
245 |
|
246 |
The algorithm for splitting the moves into jobsets is by accumulating |
247 |
moves until the next move is touching nodes already touched by the |
248 |
current moves; this means we can't execute in parallel (due to |
249 |
resource allocation in Ganeti) and thus we start a new jobset. |
250 |
|
251 |
.TP |
252 |
.B -p, --print-nodes |
253 |
Prints the before and after node status, in a format designed to allow |
254 |
the user to understand the node's most important parameters. |
255 |
|
256 |
It is possible to customise the listed information by passing a |
257 |
comma\(hyseparated list of field names to this option (the field list |
258 |
is currently undocumented), or to extend the default field list by |
259 |
prefixing the additional field list with a plus sign. By default, the |
260 |
node list will contain the following information: |
261 |
.RS |
262 |
.TP |
263 |
.B F |
264 |
a character denoting the status of the node, with '\-' meaning an |
265 |
offline node, '*' meaning N+1 failure and blank meaning a good node |
266 |
.TP |
267 |
.B Name |
268 |
the node name |
269 |
.TP |
270 |
.B t_mem |
271 |
the total node memory |
272 |
.TP |
273 |
.B n_mem |
274 |
the memory used by the node itself |
275 |
.TP |
276 |
.B i_mem |
277 |
the memory used by instances |
278 |
.TP |
279 |
.B x_mem |
280 |
amount memory which seems to be in use but cannot be determined why or |
281 |
by which instance; usually this means that the hypervisor has some |
282 |
overhead or that there are other reporting errors |
283 |
.TP |
284 |
.B f_mem |
285 |
the free node memory |
286 |
.TP |
287 |
.B r_mem |
288 |
the reserved node memory, which is the amount of free memory needed |
289 |
for N+1 compliance |
290 |
.TP |
291 |
.B t_dsk |
292 |
total disk |
293 |
.TP |
294 |
.B f_dsk |
295 |
free disk |
296 |
.TP |
297 |
.B pcpu |
298 |
the number of physical cpus on the node |
299 |
.TP |
300 |
.B vcpu |
301 |
the number of virtual cpus allocated to primary instances |
302 |
.TP |
303 |
.B pcnt |
304 |
number of primary instances |
305 |
.TP |
306 |
.B scnt |
307 |
number of secondary instances |
308 |
.TP |
309 |
.B p_fmem |
310 |
percent of free memory |
311 |
.TP |
312 |
.B p_fdsk |
313 |
percent of free disk |
314 |
.TP |
315 |
.B r_cpu |
316 |
ratio of virtual to physical cpus |
317 |
.TP |
318 |
.B lCpu |
319 |
the dynamic CPU load (if the information is available) |
320 |
.TP |
321 |
.B lMem |
322 |
the dynamic memory load (if the information is available) |
323 |
.TP |
324 |
.B lDsk |
325 |
the dynamic disk load (if the information is available) |
326 |
.TP |
327 |
.B lNet |
328 |
the dynamic net load (if the information is available) |
329 |
.RE |
330 |
|
331 |
.TP |
332 |
.B --print-instances |
333 |
Prints the before and after instance map. This is less useful as the |
334 |
node status, but it can help in understanding instance moves. |
335 |
|
336 |
.TP |
337 |
.B -o, --oneline |
338 |
Only shows a one\(hyline output from the program, designed for the case |
339 |
when one wants to look at multiple clusters at once and check their |
340 |
status. |
341 |
|
342 |
The line will contain four fields: |
343 |
.RS |
344 |
.RS 4 |
345 |
.TP 3 |
346 |
\(em |
347 |
initial cluster score |
348 |
.TP |
349 |
\(em |
350 |
number of steps in the solution |
351 |
.TP |
352 |
\(em |
353 |
final cluster score |
354 |
.TP |
355 |
\(em |
356 |
improvement in the cluster score |
357 |
.RE |
358 |
.RE |
359 |
|
360 |
.TP |
361 |
.BI "-O " name |
362 |
This option (which can be given multiple times) will mark nodes as |
363 |
being \fIoffline\fR. This means a couple of things: |
364 |
.RS |
365 |
.RS 4 |
366 |
.TP 3 |
367 |
\(em |
368 |
instances won't be placed on these nodes, not even temporarily; |
369 |
e.g. the \fIreplace primary\fR move is not available if the secondary |
370 |
node is offline, since this move requires a failover. |
371 |
.TP |
372 |
\(em |
373 |
these nodes will not be included in the score calculation (except for |
374 |
the percentage of instances on offline nodes) |
375 |
.RE |
376 |
Note that hbal will also mark as offline any nodes which are reported |
377 |
by RAPI as such, or that have "?" in file\(hybased input in any numeric |
378 |
fields. |
379 |
.RE |
380 |
|
381 |
.TP |
382 |
.BI "-e" score ", --min-score=" score |
383 |
This parameter denotes the minimum score we are happy with and alters |
384 |
the computation in two ways: |
385 |
.RS |
386 |
.RS 4 |
387 |
.TP 3 |
388 |
\(em |
389 |
if the cluster has the initial score lower than this value, then we |
390 |
don't enter the algorithm at all, and exit with success |
391 |
.TP |
392 |
\(em |
393 |
during the iterative process, if we reach a score lower than this |
394 |
value, we exit the algorithm |
395 |
.RE |
396 |
The default value of the parameter is currently \fI1e-9\fR (chosen |
397 |
empirically). |
398 |
.RE |
399 |
|
400 |
.TP |
401 |
.BI "-g" delta ", --min-gain=" delta |
402 |
Since the balancing algorithm can sometimes result in just very tiny |
403 |
improvements, that bring less gain that they cost in relocation time, |
404 |
this parameter (defaulting to 0.01) represents the minimum gain we |
405 |
require during a step, to continue balancing. |
406 |
|
407 |
.TP |
408 |
.BI "--min-gain-limit=" threshold |
409 |
The above min-gain option will only take effect if the cluster score |
410 |
is already below \fIthreshold\fR (defaults to 0.1). The rationale |
411 |
behind this setting is that at high cluster scores (badly balanced |
412 |
clusters), we don't want to abort the rebalance too quickly, as later |
413 |
gains might still be significant. However, under the threshold, the |
414 |
total gain is only the threshold value, so we can exit early. |
415 |
|
416 |
.TP |
417 |
.BI "--no-disk-moves" |
418 |
This parameter prevents hbal from using disk move (i.e. "gnt\-instance |
419 |
replace\-disks") operations. This will result in a much quicker |
420 |
balancing, but of course the improvements are limited. It is up to the |
421 |
user to decide when to use one or another. |
422 |
|
423 |
.TP |
424 |
.B "--evac-mode" |
425 |
This parameter restricts the list of instances considered for moving |
426 |
to the ones living on offline/drained nodes. It can be used as a |
427 |
(bulk) replacement for Ganeti's own \fIgnt-node evacuate\fR, with the |
428 |
note that it doesn't guarantee full evacuation. |
429 |
|
430 |
.TP |
431 |
.BI "--exclude-instances " instances |
432 |
This parameter marks the given instances (as a comma-separated list) |
433 |
from being moved during the rebalance. |
434 |
|
435 |
.TP |
436 |
.BI "-U" util-file |
437 |
This parameter specifies a file holding instance dynamic utilisation |
438 |
information that will be used to tweak the balancing algorithm to |
439 |
equalise load on the nodes (as opposed to static resource usage). The |
440 |
file is in the format "instance_name cpu_util mem_util disk_util |
441 |
net_util" where the "_util" parameters are interpreted as numbers and |
442 |
the instance name must match exactly the instance as read from |
443 |
Ganeti. In case of unknown instance names, the program will abort. |
444 |
|
445 |
If not given, the default values are one for all metrics and thus |
446 |
dynamic utilisation has only one effect on the algorithm: the |
447 |
equalisation of the secondary instances across nodes (this is the only |
448 |
metric that is not tracked by another, dedicated value, and thus the |
449 |
disk load of instances will cause secondary instance |
450 |
equalisation). Note that value of one will also influence slightly the |
451 |
primary instance count, but that is already tracked via other metrics |
452 |
and thus the influence of the dynamic utilisation will be practically |
453 |
insignificant. |
454 |
|
455 |
.TP |
456 |
.BI "-t" datafile ", --text-data=" datafile |
457 |
The name of the file holding node and instance information (if not |
458 |
collecting via RAPI or LUXI). This or one of the other backends must |
459 |
be selected. |
460 |
|
461 |
.TP |
462 |
.BI "-S" datafile ", --save-cluster=" datafile |
463 |
If given, the state of the cluster at the end of the balancing is |
464 |
saved to the given file. This allows re-feeding the cluster state to |
465 |
either hbal itself or for example hspace. |
466 |
|
467 |
.TP |
468 |
.BI "-m" cluster |
469 |
Collect data directly from the |
470 |
.I cluster |
471 |
given as an argument via RAPI. If the argument doesn't contain a colon |
472 |
(:), then it is converted into a fully\(hybuilt URL via prepending |
473 |
https:// and appending the default RAPI port, otherwise it's |
474 |
considered a fully\(hyspecified URL and is used as\(hyis. |
475 |
|
476 |
.TP |
477 |
.BI "-L[" path "]" |
478 |
Collect data directly from the master daemon, which is to be contacted |
479 |
via the luxi (an internal Ganeti protocol). An optional \fIpath\fR |
480 |
argument is interpreted as the path to the unix socket on which the |
481 |
master daemon listens; otherwise, the default path used by ganeti when |
482 |
installed with \fI--localstatedir=/var\fR is used. |
483 |
|
484 |
.TP |
485 |
.B "-X" |
486 |
When using the Luxi backend, hbal can also execute the given |
487 |
commands. The execution method is to execute the individual jobsets |
488 |
(see the \fI-C\fR option for details) in separate stages, aborting if |
489 |
at any time a jobset doesn't have all jobs successful. Each step in |
490 |
the balancing solution will be translated into exactly one Ganeti job |
491 |
(having between one and three OpCodes), and all the steps in a jobset |
492 |
will be executed in parallel. The jobsets themselves are executed |
493 |
serially. |
494 |
|
495 |
.TP |
496 |
.BI "-l" N ", --max-length=" N |
497 |
Restrict the solution to this length. This can be used for example to |
498 |
automate the execution of the balancing. |
499 |
|
500 |
.TP |
501 |
.BI "--max-cpu " cpu-ratio |
502 |
The maximum virtual\(hyto\(hyphysical cpu ratio, as a floating point |
503 |
number between zero and one. For example, specifying \fIcpu-ratio\fR |
504 |
as \fB2.5\fR means that, for a 4\(hycpu machine, a maximum of 10 |
505 |
virtual cpus should be allowed to be in use for primary instances. A |
506 |
value of one doesn't make sense though, as that means no disk space |
507 |
can be used on it. |
508 |
|
509 |
.TP |
510 |
.BI "--min-disk " disk-ratio |
511 |
The minimum amount of free disk space remaining, as a floating point |
512 |
number. For example, specifying \fIdisk-ratio\fR as \fB0.25\fR means |
513 |
that at least one quarter of disk space should be left free on nodes. |
514 |
|
515 |
.TP |
516 |
.B -v, --verbose |
517 |
Increase the output verbosity. Each usage of this option will increase |
518 |
the verbosity (currently more than 2 doesn't make sense) from the |
519 |
default of one. |
520 |
|
521 |
.TP |
522 |
.B -q, --quiet |
523 |
Decrease the output verbosity. Each usage of this option will decrease |
524 |
the verbosity (less than zero doesn't make sense) from the default of |
525 |
one. |
526 |
|
527 |
.TP |
528 |
.B -V, --version |
529 |
Just show the program version and exit. |
530 |
|
531 |
.SH EXIT STATUS |
532 |
|
533 |
The exist status of the command will be zero, unless for some reason |
534 |
the algorithm fatally failed (e.g. wrong node or instance data). |
535 |
|
536 |
.SH ENVIRONMENT |
537 |
|
538 |
If the variables \fBHTOOLS_NODES\fR and \fBHTOOLS_INSTANCES\fR are |
539 |
present in the environment, they will override the default names for |
540 |
the nodes and instances files. These will have of course no effect |
541 |
when the RAPI or Luxi backends are used. |
542 |
|
543 |
.SH BUGS |
544 |
|
545 |
The program does not check its input data for consistency, and aborts |
546 |
with cryptic errors messages in this case. |
547 |
|
548 |
The algorithm is not perfect. |
549 |
|
550 |
The output format is not easily scriptable, and the program should |
551 |
feed moves directly into Ganeti (either via RAPI or via a gnt\-debug |
552 |
input file). |
553 |
|
554 |
.SH EXAMPLE |
555 |
|
556 |
Note that this example are not for the latest version (they don't have |
557 |
full node data). |
558 |
|
559 |
.SS Default output |
560 |
|
561 |
With the default options, the program shows each individual step and |
562 |
the improvements it brings in cluster score: |
563 |
|
564 |
.in +4n |
565 |
.nf |
566 |
.RB "$" " hbal" |
567 |
Loaded 20 nodes, 80 instances |
568 |
Cluster is not N+1 happy, continuing but no guarantee that the cluster will end N+1 happy. |
569 |
Initial score: 0.52329131 |
570 |
Trying to minimize the CV... |
571 |
1. instance14 node1:node10 => node16:node10 0.42109120 a=f r:node16 f |
572 |
2. instance54 node4:node15 => node16:node15 0.31904594 a=f r:node16 f |
573 |
3. instance4 node5:node2 => node2:node16 0.26611015 a=f r:node16 |
574 |
4. instance48 node18:node20 => node2:node18 0.21361717 a=r:node2 f |
575 |
5. instance93 node19:node18 => node16:node19 0.16166425 a=r:node16 f |
576 |
6. instance89 node3:node20 => node2:node3 0.11005629 a=r:node2 f |
577 |
7. instance5 node6:node2 => node16:node6 0.05841589 a=r:node16 f |
578 |
8. instance94 node7:node20 => node20:node16 0.00658759 a=f r:node16 |
579 |
9. instance44 node20:node2 => node2:node15 0.00438740 a=f r:node15 |
580 |
10. instance62 node14:node18 => node14:node16 0.00390087 a=r:node16 |
581 |
11. instance13 node11:node14 => node11:node16 0.00361787 a=r:node16 |
582 |
12. instance19 node10:node11 => node10:node7 0.00336636 a=r:node7 |
583 |
13. instance43 node12:node13 => node12:node1 0.00305681 a=r:node1 |
584 |
14. instance1 node1:node2 => node1:node4 0.00263124 a=r:node4 |
585 |
15. instance58 node19:node20 => node19:node17 0.00252594 a=r:node17 |
586 |
Cluster score improved from 0.52329131 to 0.00252594 |
587 |
.fi |
588 |
.in |
589 |
|
590 |
In the above output, we can see: |
591 |
- the input data (here from files) shows a cluster with 20 nodes and |
592 |
80 instances |
593 |
- the cluster is not initially N+1 compliant |
594 |
- the initial score is 0.52329131 |
595 |
|
596 |
The step list follows, showing the instance, its initial |
597 |
primary/secondary nodes, the new primary secondary, the cluster list, |
598 |
and the actions taken in this step (with 'f' denoting failover/migrate |
599 |
and 'r' denoting replace secondary). |
600 |
|
601 |
Finally, the program shows the improvement in cluster score. |
602 |
|
603 |
A more detailed output is obtained via the \fB-C\fR and \fB-p\fR options: |
604 |
|
605 |
.in +4n |
606 |
.nf |
607 |
.RB "$" " hbal" |
608 |
Loaded 20 nodes, 80 instances |
609 |
Cluster is not N+1 happy, continuing but no guarantee that the cluster will end N+1 happy. |
610 |
Initial cluster status: |
611 |
N1 Name t_mem f_mem r_mem t_dsk f_dsk pri sec p_fmem p_fdsk |
612 |
* node1 32762 1280 6000 1861 1026 5 3 0.03907 0.55179 |
613 |
node2 32762 31280 12000 1861 1026 0 8 0.95476 0.55179 |
614 |
* node3 32762 1280 6000 1861 1026 5 3 0.03907 0.55179 |
615 |
* node4 32762 1280 6000 1861 1026 5 3 0.03907 0.55179 |
616 |
* node5 32762 1280 6000 1861 978 5 5 0.03907 0.52573 |
617 |
* node6 32762 1280 6000 1861 1026 5 3 0.03907 0.55179 |
618 |
* node7 32762 1280 6000 1861 1026 5 3 0.03907 0.55179 |
619 |
node8 32762 7280 6000 1861 1026 4 4 0.22221 0.55179 |
620 |
node9 32762 7280 6000 1861 1026 4 4 0.22221 0.55179 |
621 |
* node10 32762 7280 12000 1861 1026 4 4 0.22221 0.55179 |
622 |
node11 32762 7280 6000 1861 922 4 5 0.22221 0.49577 |
623 |
node12 32762 7280 6000 1861 1026 4 4 0.22221 0.55179 |
624 |
node13 32762 7280 6000 1861 922 4 5 0.22221 0.49577 |
625 |
node14 32762 7280 6000 1861 922 4 5 0.22221 0.49577 |
626 |
* node15 32762 7280 12000 1861 1131 4 3 0.22221 0.60782 |
627 |
node16 32762 31280 0 1861 1860 0 0 0.95476 1.00000 |
628 |
node17 32762 7280 6000 1861 1106 5 3 0.22221 0.59479 |
629 |
* node18 32762 1280 6000 1396 561 5 3 0.03907 0.40239 |
630 |
* node19 32762 1280 6000 1861 1026 5 3 0.03907 0.55179 |
631 |
node20 32762 13280 12000 1861 689 3 9 0.40535 0.37068 |
632 |
|
633 |
Initial score: 0.52329131 |
634 |
Trying to minimize the CV... |
635 |
1. instance14 node1:node10 => node16:node10 0.42109120 a=f r:node16 f |
636 |
2. instance54 node4:node15 => node16:node15 0.31904594 a=f r:node16 f |
637 |
3. instance4 node5:node2 => node2:node16 0.26611015 a=f r:node16 |
638 |
4. instance48 node18:node20 => node2:node18 0.21361717 a=r:node2 f |
639 |
5. instance93 node19:node18 => node16:node19 0.16166425 a=r:node16 f |
640 |
6. instance89 node3:node20 => node2:node3 0.11005629 a=r:node2 f |
641 |
7. instance5 node6:node2 => node16:node6 0.05841589 a=r:node16 f |
642 |
8. instance94 node7:node20 => node20:node16 0.00658759 a=f r:node16 |
643 |
9. instance44 node20:node2 => node2:node15 0.00438740 a=f r:node15 |
644 |
10. instance62 node14:node18 => node14:node16 0.00390087 a=r:node16 |
645 |
11. instance13 node11:node14 => node11:node16 0.00361787 a=r:node16 |
646 |
12. instance19 node10:node11 => node10:node7 0.00336636 a=r:node7 |
647 |
13. instance43 node12:node13 => node12:node1 0.00305681 a=r:node1 |
648 |
14. instance1 node1:node2 => node1:node4 0.00263124 a=r:node4 |
649 |
15. instance58 node19:node20 => node19:node17 0.00252594 a=r:node17 |
650 |
Cluster score improved from 0.52329131 to 0.00252594 |
651 |
|
652 |
Commands to run to reach the above solution: |
653 |
echo step 1 |
654 |
echo gnt\-instance migrate instance14 |
655 |
echo gnt\-instance replace\-disks \-n node16 instance14 |
656 |
echo gnt\-instance migrate instance14 |
657 |
echo step 2 |
658 |
echo gnt\-instance migrate instance54 |
659 |
echo gnt\-instance replace\-disks \-n node16 instance54 |
660 |
echo gnt\-instance migrate instance54 |
661 |
echo step 3 |
662 |
echo gnt\-instance migrate instance4 |
663 |
echo gnt\-instance replace\-disks \-n node16 instance4 |
664 |
echo step 4 |
665 |
echo gnt\-instance replace\-disks \-n node2 instance48 |
666 |
echo gnt\-instance migrate instance48 |
667 |
echo step 5 |
668 |
echo gnt\-instance replace\-disks \-n node16 instance93 |
669 |
echo gnt\-instance migrate instance93 |
670 |
echo step 6 |
671 |
echo gnt\-instance replace\-disks \-n node2 instance89 |
672 |
echo gnt\-instance migrate instance89 |
673 |
echo step 7 |
674 |
echo gnt\-instance replace\-disks \-n node16 instance5 |
675 |
echo gnt\-instance migrate instance5 |
676 |
echo step 8 |
677 |
echo gnt\-instance migrate instance94 |
678 |
echo gnt\-instance replace\-disks \-n node16 instance94 |
679 |
echo step 9 |
680 |
echo gnt\-instance migrate instance44 |
681 |
echo gnt\-instance replace\-disks \-n node15 instance44 |
682 |
echo step 10 |
683 |
echo gnt\-instance replace\-disks \-n node16 instance62 |
684 |
echo step 11 |
685 |
echo gnt\-instance replace\-disks \-n node16 instance13 |
686 |
echo step 12 |
687 |
echo gnt\-instance replace\-disks \-n node7 instance19 |
688 |
echo step 13 |
689 |
echo gnt\-instance replace\-disks \-n node1 instance43 |
690 |
echo step 14 |
691 |
echo gnt\-instance replace\-disks \-n node4 instance1 |
692 |
echo step 15 |
693 |
echo gnt\-instance replace\-disks \-n node17 instance58 |
694 |
|
695 |
Final cluster status: |
696 |
N1 Name t_mem f_mem r_mem t_dsk f_dsk pri sec p_fmem p_fdsk |
697 |
node1 32762 7280 6000 1861 1026 4 4 0.22221 0.55179 |
698 |
node2 32762 7280 6000 1861 1026 4 4 0.22221 0.55179 |
699 |
node3 32762 7280 6000 1861 1026 4 4 0.22221 0.55179 |
700 |
node4 32762 7280 6000 1861 1026 4 4 0.22221 0.55179 |
701 |
node5 32762 7280 6000 1861 1078 4 5 0.22221 0.57947 |
702 |
node6 32762 7280 6000 1861 1026 4 4 0.22221 0.55179 |
703 |
node7 32762 7280 6000 1861 1026 4 4 0.22221 0.55179 |
704 |
node8 32762 7280 6000 1861 1026 4 4 0.22221 0.55179 |
705 |
node9 32762 7280 6000 1861 1026 4 4 0.22221 0.55179 |
706 |
node10 32762 7280 6000 1861 1026 4 4 0.22221 0.55179 |
707 |
node11 32762 7280 6000 1861 1022 4 4 0.22221 0.54951 |
708 |
node12 32762 7280 6000 1861 1026 4 4 0.22221 0.55179 |
709 |
node13 32762 7280 6000 1861 1022 4 4 0.22221 0.54951 |
710 |
node14 32762 7280 6000 1861 1022 4 4 0.22221 0.54951 |
711 |
node15 32762 7280 6000 1861 1031 4 4 0.22221 0.55408 |
712 |
node16 32762 7280 6000 1861 1060 4 4 0.22221 0.57007 |
713 |
node17 32762 7280 6000 1861 1006 5 4 0.22221 0.54105 |
714 |
node18 32762 7280 6000 1396 761 4 2 0.22221 0.54570 |
715 |
node19 32762 7280 6000 1861 1026 4 4 0.22221 0.55179 |
716 |
node20 32762 13280 6000 1861 1089 3 5 0.40535 0.58565 |
717 |
|
718 |
.fi |
719 |
.in |
720 |
|
721 |
Here we see, beside the step list, the initial and final cluster |
722 |
status, with the final one showing all nodes being N+1 compliant, and |
723 |
the command list to reach the final solution. In the initial listing, |
724 |
we see which nodes are not N+1 compliant. |
725 |
|
726 |
The algorithm is stable as long as each step above is fully completed, |
727 |
e.g. in step 8, both the migrate and the replace\-disks are |
728 |
done. Otherwise, if only the migrate is done, the input data is |
729 |
changed in a way that the program will output a different solution |
730 |
list (but hopefully will end in the same state). |
731 |
|
732 |
.SH SEE ALSO |
733 |
.BR hspace "(1), " hscan "(1), " hail "(1), " |
734 |
.BR ganeti "(7), " gnt-instance "(8), " gnt-node "(8)" |
735 |
|
736 |
.SH "COPYRIGHT" |
737 |
.PP |
738 |
Copyright (C) 2009 Google Inc. Permission is granted to copy, |
739 |
distribute and/or modify under the terms of the GNU General Public |
740 |
License as published by the Free Software Foundation; either version 2 |
741 |
of the License, or (at your option) any later version. |
742 |
.PP |
743 |
On Debian systems, the complete text of the GNU General Public License |
744 |
can be found in /usr/share/common-licenses/GPL. |