1 .TH HBAL 1 2009-03-23 htools "Ganeti H-tools"
3 hbal \- Cluster balancer for Ganeti
14 .BI "[-m " cluster "]"
15 .BI "[-n " nodes-file " ]"
16 .BI "[-i " instances-file "]"
17 .BI "[--max-cpu " cpu-ratio "]"
18 .BI "[--min-disk " disk-ratio "]"
24 hbal is a cluster balancer that looks at the current state of the
25 cluster (nodes with their total and free disk, memory, etc.) and
26 instance placement and computes a series of steps designed to bring
27 the cluster into a better state.
29 The algorithm to do so is designed to be stable (i.e. it will give you
30 the same results when restarting it from the middle of the solution)
31 and reasonably fast. It is not, however, designed to be a perfect
32 algorithm - it is possible to make it go into a corner from which it
33 can find no improvement, because it only look one "step" ahead.
35 By default, the program will show the solution incrementally as it is
36 computed, in a somewhat cryptic format; for getting the actual Ganeti
37 command list, use the \fB-C\fR option.
41 The program works in independent steps; at each step, we compute the
42 best instance move that lowers the cluster score.
44 The possible move type for an instance are combinations of
45 failover/migrate and replace-disks such that we change one of the
46 instance nodes, and the other one remains (but possibly with changed
47 role, e.g. from primary it becomes secondary). The list is:
57 replace primary, a composite move (f, r, f)
60 failover and replace secondary, also composite (f, r)
63 replace secondary and failover, also composite (r, f)
66 We don't do the only remaining possibility of replacing both nodes
67 (r,f,r,f or the equivalent f,r,f,r) since these move needs an
68 exhaustive search over both candidate primary and secondary nodes, and
69 is O(n*n) in the number of nodes. Furthermore, it doesn't seems to
70 give better scores but will result in more disk replacements.
74 As said before, the algorithm tries to minimise the cluster score at
75 each step. Currently this score is computed as a sum of the following
80 coefficient of variance of the percent of free memory
83 coefficient of variance of the percent of reserved memory
86 coefficient of variance of the percent of free disk
89 percentage of nodes failing N+1 check
92 percentage of instances living (either as primary or secondary) on
96 coefficent of variance of the ratio of virtual-to-physical cpus (for
97 primary instaces of the node)
100 The free memory and free disk values help ensure that all nodes are
101 somewhat balanced in their resource usage. The reserved memory helps
102 to ensure that nodes are somewhat balanced in holding secondary
103 instances, and that no node keeps too much memory reserved for
104 N+1. And finally, the N+1 percentage helps guide the algorithm towards
105 eliminating N+1 failures, if possible.
107 Except for the N+1 failures and offline instances percentage, we use
108 the coefficient of variance since this brings the values into the same
109 unit so to speak, and with a restrict domain of values (between zero
110 and one). The percentage of N+1 failures, while also in this numeric
111 range, doesn't actually has the same meaning, but it has shown to work
114 The other alternative, using for N+1 checks the coefficient of
115 variance of (N+1 fail=1, N+1 pass=0) across nodes could hint the
116 algorithm to make more N+1 failures if most nodes are N+1 fail
117 already. Since this (making N+1 failures) is not allowed by other
118 rules of the algorithm, so the N+1 checks would simply not work
119 anymore in this case.
121 The offline instances percentage (meaning the percentage of instances
122 living on offline nodes) will cause the algorithm to actively move
123 instances away from offline nodes. This, coupled with the restriction
124 on placement given by offline nodes, will cause evacuation of such
127 On a perfectly balanced cluster (all nodes the same size, all
128 instances the same size and spread across the nodes equally), all
129 values would be zero. This doesn't happen too often in practice :)
131 .SS OFFLINE INSTANCES
133 Since current Ganeti versions do not report the memory used by offline
134 (down) instances, ignoring the run status of instances will cause
135 wrong calculations. For this reason, the algorithm subtracts the
136 memory size of down instances from the free node memory of their
137 primary node, in effect simulating the startup of such instances.
139 .SS OTHER POSSIBLE METRICS
141 It would be desirable to add more metrics to the algorithm, especially
142 dynamically-computed metrics, such as:
146 CPU usage of instances
156 The options that can be passed to the program are as follows:
158 .B -C, --print-commands
159 Print the command list at the end of the run. Without this, the
160 program will only show a shorter, but cryptic output.
163 Prints the before and after node status, in a format designed to allow
164 the user to understand the node's most important parameters.
166 The node list will contain these informations:
170 a character denoting the status of the node, with '-' meaning an
171 offline node, '*' meaning N+1 failure and blank meaning a good node
177 the total node memory
180 the memory used by the node itself
183 the memory used by instances
186 amount memory which seems to be in use but cannot be determined why or
187 by which instance; usually this means that the hypervisor has some
188 overhead or that there are other reporting errors
194 the reserved node memory, which is the amount of free memory needed
204 the number of physical cpus on the node
207 the number of virtual cpus allocated to primary instances
210 number of primary instances
213 number of secondary instances
216 percent of free memory
222 ratio of virtual to physical cpus
227 Only shows a one-line output from the program, designed for the case
228 when one wants to look at multiple clusters at once and check their
231 The line will contain four fields:
236 initial cluster score
239 number of steps in the solution
245 improvement in the cluster score
251 This option (which can be given multiple times) will mark nodes as
252 being \fIoffline\fR. This means a couple of things:
257 instances won't be placed on these nodes, not even temporarily;
258 e.g. the \fIreplace primary\fR move is not available if the secondary
259 node is offline, since this move requires a failover.
262 these nodes will not be included in the score calculation (except for
263 the percentage of instances on offline nodes)
265 Note that hbal will also mark as offline any nodes which are reported
266 by RAPI as such, or that have "?" in file-based input in any numeric
271 .BI "-e" score ", --min-score=" score
272 This parameter denotes the minimum score we are happy with and alters
273 the computation in two ways:
278 if the cluster has the initial score lower than this value, then we
279 don't enter the algorithm at all, and exit with success
282 during the iterative process, if we reach a score lower than this
283 value, we exit the algorithm
285 The default value of the parameter is currently \fI1e-9\fR (chosen
290 .BI "-n" nodefile ", --nodes=" nodefile
291 The name of the file holding node information (if not collecting via
292 RAPI), instead of the default \fInodes\fR file (but see below how to
293 customize the default value via the environment).
296 .BI "-i" instancefile ", --instances=" instancefile
297 The name of the file holding instance information (if not collecting
298 via RAPI), instead of the default \fIinstances\fR file (but see below
299 how to customize the default value via the environment).
303 Collect data not from files but directly from the
305 given as an argument via RAPI. If the argument doesn't contain a colon
306 (:), then it is converted into a fully-built URL via prepending
307 https:// and appending the default RAPI port, otherwise it's
308 considered a fully-specified URL and is used as-is.
311 .BI "-l" N ", --max-length=" N
312 Restrict the solution to this length. This can be used for example to
313 automate the execution of the balancing.
316 .BI "--max-cpu " cpu-ratio
317 The maximum virtual-to-physical cpu ratio, as a floating point number
318 between zero and one. For example, specifying \fIcpu-ratio\fR as
319 \fB2.5\fR means that, for a 4-cpu machine, a maximum of 10 virtual
320 cpus should be allowed to be in use for primary instances. A value of
321 one doesn't make sense though, as that means no disk space can be used
325 .BI "--min-disk " disk-ratio
326 The minimum amount of free disk space remaining, as a floating point
327 number. For example, specifying \fIdisk-ratio\fR as \fB0.25\fR means
328 that at least one quarter of disk space should be left free on nodes.
332 Increase the output verbosity. Each usage of this option will increase
333 the verbosity (currently more than 2 doesn't make sense) from the
338 Decrease the output verbosity. Each usage of this option will decrease
339 the verbosity (less than zero doesn't make sense) from the default of
344 Just show the program version and exit.
348 The exist status of the command will be zero, unless for some reason
349 the algorithm fatally failed (e.g. wrong node or instance data).
353 If the variables \fBHTOOLS_NODES\fR and \fBHTOOLS_INSTANCES\fR are
354 present in the environment, they will override the default names for
355 the nodes and instances files. These will have of course no effect
360 The program does not check its input data for consistency, and aborts
361 with cryptic errors messages in this case.
363 The algorithm is not perfect.
365 The output format is not easily scriptable, and the program should
366 feed moves directly into Ganeti (either via RAPI or via a gnt-debug
371 Note that this example are not for the latest version (they don't have
376 With the default options, the program shows each individual step and
377 the improvements it brings in cluster score:
382 Loaded 20 nodes, 80 instances
383 Cluster is not N+1 happy, continuing but no guarantee that the cluster will end N+1 happy.
384 Initial score: 0.52329131
385 Trying to minimize the CV...
386 1. instance14 node1:node10 => node16:node10 0.42109120 a=f r:node16 f
387 2. instance54 node4:node15 => node16:node15 0.31904594 a=f r:node16 f
388 3. instance4 node5:node2 => node2:node16 0.26611015 a=f r:node16
389 4. instance48 node18:node20 => node2:node18 0.21361717 a=r:node2 f
390 5. instance93 node19:node18 => node16:node19 0.16166425 a=r:node16 f
391 6. instance89 node3:node20 => node2:node3 0.11005629 a=r:node2 f
392 7. instance5 node6:node2 => node16:node6 0.05841589 a=r:node16 f
393 8. instance94 node7:node20 => node20:node16 0.00658759 a=f r:node16
394 9. instance44 node20:node2 => node2:node15 0.00438740 a=f r:node15
395 10. instance62 node14:node18 => node14:node16 0.00390087 a=r:node16
396 11. instance13 node11:node14 => node11:node16 0.00361787 a=r:node16
397 12. instance19 node10:node11 => node10:node7 0.00336636 a=r:node7
398 13. instance43 node12:node13 => node12:node1 0.00305681 a=r:node1
399 14. instance1 node1:node2 => node1:node4 0.00263124 a=r:node4
400 15. instance58 node19:node20 => node19:node17 0.00252594 a=r:node17
401 Cluster score improved from 0.52329131 to 0.00252594
405 In the above output, we can see:
406 - the input data (here from files) shows a cluster with 20 nodes and
408 - the cluster is not initially N+1 compliant
409 - the initial score is 0.52329131
411 The step list follows, showing the instance, its initial
412 primary/secondary nodes, the new primary secondary, the cluster list,
413 and the actions taken in this step (with 'f' denoting failover/migrate
414 and 'r' denoting replace secondary).
416 Finally, the program shows the improvement in cluster score.
418 A more detailed output is obtained via the \fB-C\fR and \fB-p\fR options:
423 Loaded 20 nodes, 80 instances
424 Cluster is not N+1 happy, continuing but no guarantee that the cluster will end N+1 happy.
425 Initial cluster status:
426 N1 Name t_mem f_mem r_mem t_dsk f_dsk pri sec p_fmem p_fdsk
427 * node1 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
428 node2 32762 31280 12000 1861 1026 0 8 0.95476 0.55179
429 * node3 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
430 * node4 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
431 * node5 32762 1280 6000 1861 978 5 5 0.03907 0.52573
432 * node6 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
433 * node7 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
434 node8 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
435 node9 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
436 * node10 32762 7280 12000 1861 1026 4 4 0.22221 0.55179
437 node11 32762 7280 6000 1861 922 4 5 0.22221 0.49577
438 node12 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
439 node13 32762 7280 6000 1861 922 4 5 0.22221 0.49577
440 node14 32762 7280 6000 1861 922 4 5 0.22221 0.49577
441 * node15 32762 7280 12000 1861 1131 4 3 0.22221 0.60782
442 node16 32762 31280 0 1861 1860 0 0 0.95476 1.00000
443 node17 32762 7280 6000 1861 1106 5 3 0.22221 0.59479
444 * node18 32762 1280 6000 1396 561 5 3 0.03907 0.40239
445 * node19 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
446 node20 32762 13280 12000 1861 689 3 9 0.40535 0.37068
448 Initial score: 0.52329131
449 Trying to minimize the CV...
450 1. instance14 node1:node10 => node16:node10 0.42109120 a=f r:node16 f
451 2. instance54 node4:node15 => node16:node15 0.31904594 a=f r:node16 f
452 3. instance4 node5:node2 => node2:node16 0.26611015 a=f r:node16
453 4. instance48 node18:node20 => node2:node18 0.21361717 a=r:node2 f
454 5. instance93 node19:node18 => node16:node19 0.16166425 a=r:node16 f
455 6. instance89 node3:node20 => node2:node3 0.11005629 a=r:node2 f
456 7. instance5 node6:node2 => node16:node6 0.05841589 a=r:node16 f
457 8. instance94 node7:node20 => node20:node16 0.00658759 a=f r:node16
458 9. instance44 node20:node2 => node2:node15 0.00438740 a=f r:node15
459 10. instance62 node14:node18 => node14:node16 0.00390087 a=r:node16
460 11. instance13 node11:node14 => node11:node16 0.00361787 a=r:node16
461 12. instance19 node10:node11 => node10:node7 0.00336636 a=r:node7
462 13. instance43 node12:node13 => node12:node1 0.00305681 a=r:node1
463 14. instance1 node1:node2 => node1:node4 0.00263124 a=r:node4
464 15. instance58 node19:node20 => node19:node17 0.00252594 a=r:node17
465 Cluster score improved from 0.52329131 to 0.00252594
467 Commands to run to reach the above solution:
469 echo gnt-instance migrate instance14
470 echo gnt-instance replace-disks -n node16 instance14
471 echo gnt-instance migrate instance14
473 echo gnt-instance migrate instance54
474 echo gnt-instance replace-disks -n node16 instance54
475 echo gnt-instance migrate instance54
477 echo gnt-instance migrate instance4
478 echo gnt-instance replace-disks -n node16 instance4
480 echo gnt-instance replace-disks -n node2 instance48
481 echo gnt-instance migrate instance48
483 echo gnt-instance replace-disks -n node16 instance93
484 echo gnt-instance migrate instance93
486 echo gnt-instance replace-disks -n node2 instance89
487 echo gnt-instance migrate instance89
489 echo gnt-instance replace-disks -n node16 instance5
490 echo gnt-instance migrate instance5
492 echo gnt-instance migrate instance94
493 echo gnt-instance replace-disks -n node16 instance94
495 echo gnt-instance migrate instance44
496 echo gnt-instance replace-disks -n node15 instance44
498 echo gnt-instance replace-disks -n node16 instance62
500 echo gnt-instance replace-disks -n node16 instance13
502 echo gnt-instance replace-disks -n node7 instance19
504 echo gnt-instance replace-disks -n node1 instance43
506 echo gnt-instance replace-disks -n node4 instance1
508 echo gnt-instance replace-disks -n node17 instance58
510 Final cluster status:
511 N1 Name t_mem f_mem r_mem t_dsk f_dsk pri sec p_fmem p_fdsk
512 node1 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
513 node2 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
514 node3 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
515 node4 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
516 node5 32762 7280 6000 1861 1078 4 5 0.22221 0.57947
517 node6 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
518 node7 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
519 node8 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
520 node9 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
521 node10 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
522 node11 32762 7280 6000 1861 1022 4 4 0.22221 0.54951
523 node12 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
524 node13 32762 7280 6000 1861 1022 4 4 0.22221 0.54951
525 node14 32762 7280 6000 1861 1022 4 4 0.22221 0.54951
526 node15 32762 7280 6000 1861 1031 4 4 0.22221 0.55408
527 node16 32762 7280 6000 1861 1060 4 4 0.22221 0.57007
528 node17 32762 7280 6000 1861 1006 5 4 0.22221 0.54105
529 node18 32762 7280 6000 1396 761 4 2 0.22221 0.54570
530 node19 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
531 node20 32762 13280 6000 1861 1089 3 5 0.40535 0.58565
536 Here we see, beside the step list, the initial and final cluster
537 status, with the final one showing all nodes being N+1 compliant, and
538 the command list to reach the final solution. In the initial listing,
539 we see which nodes are not N+1 compliant.
541 The algorithm is stable as long as each step above is fully completed,
542 e.g. in step 8, both the migrate and the replace-disks are
543 done. Otherwise, if only the migrate is done, the input data is
544 changed in a way that the program will output a different solution
545 list (but hopefully will end in the same state).
548 .BR hspace "(1), " hscan "(1), " hail "(1), "
549 .BR ganeti "(7), " gnt-instance "(8), " gnt-node "(8)"
553 Copyright (C) 2009 Google Inc. Permission is granted to copy,
554 distribute and/or modify under the terms of the GNU General Public
555 License as published by the Free Software Foundation; either version 2
556 of the License, or (at your option) any later version.
558 On Debian systems, the complete text of the GNU General Public License
559 can be found in /usr/share/common-licenses/GPL.