1 .TH HBAL 1 2009-03-23 htools "Ganeti H-tools"
3 hbal \- Cluster balancer for Ganeti
7 .B "[backend options...]"
8 .B "[algorithm options...]"
9 .B "[reporting options...]"
16 .BI "[ -m " cluster " ]"
18 .BI "[ -L[" path "] [-X]]"
20 .BI "[ -n " nodes-file " ]"
21 .BI "[ -i " instances-file " ]"
25 .BI "[ --max-cpu " cpu-ratio " ]"
26 .BI "[ --min-disk " disk-ratio " ]"
27 .BI "[ -l " limit " ]"
28 .BI "[ -e " score " ]"
29 .BI "[ -O " name... " ]"
30 .B "[ --no-disk-moves ]"
31 .BI "[ -U " util-file " ]"
35 .BI "[ -C[" file "] ]"
36 .BI "[ -p[" fields "] ]"
37 .B "[ --print-instances ]"
43 hbal is a cluster balancer that looks at the current state of the
44 cluster (nodes with their total and free disk, memory, etc.) and
45 instance placement and computes a series of steps designed to bring
46 the cluster into a better state.
48 The algorithm used is designed to be stable (i.e. it will give you the
49 same results when restarting it from the middle of the solution) and
50 reasonably fast. It is not, however, designed to be a perfect
51 algorithm \(em it is possible to make it go into a corner from which
52 it can find no improvement, because it looks only one "step" ahead.
54 By default, the program will show the solution incrementally as it is
55 computed, in a somewhat cryptic format; for getting the actual Ganeti
56 command list, use the \fB-C\fR option.
60 The program works in independent steps; at each step, we compute the
61 best instance move that lowers the cluster score.
63 The possible move type for an instance are combinations of
64 failover/migrate and replace-disks such that we change one of the
65 instance nodes, and the other one remains (but possibly with changed
66 role, e.g. from primary it becomes secondary). The list is:
76 replace primary, a composite move (f, r, f)
79 failover and replace secondary, also composite (f, r)
82 replace secondary and failover, also composite (r, f)
85 We don't do the only remaining possibility of replacing both nodes
86 (r,f,r,f or the equivalent f,r,f,r) since these move needs an
87 exhaustive search over both candidate primary and secondary nodes, and
88 is O(n*n) in the number of nodes. Furthermore, it doesn't seems to
89 give better scores but will result in more disk replacements.
93 As said before, the algorithm tries to minimise the cluster score at
94 each step. Currently this score is computed as a sum of the following
99 coefficient of variance of the percent of free memory
102 coefficient of variance of the percent of reserved memory
105 coefficient of variance of the percent of free disk
108 percentage of nodes failing N+1 check
111 percentage of instances living (either as primary or secondary) on
115 coefficent of variance of the ratio of virtual-to-physical cpus (for
116 primary instaces of the node)
119 coefficients of variance of the dynamic load on the nodes, for cpus,
120 memory, disk and network
123 The free memory and free disk values help ensure that all nodes are
124 somewhat balanced in their resource usage. The reserved memory helps
125 to ensure that nodes are somewhat balanced in holding secondary
126 instances, and that no node keeps too much memory reserved for
127 N+1. And finally, the N+1 percentage helps guide the algorithm towards
128 eliminating N+1 failures, if possible.
130 Except for the N+1 failures and offline instances percentage, we use
131 the coefficient of variance since this brings the values into the same
132 unit so to speak, and with a restrict domain of values (between zero
133 and one). The percentage of N+1 failures, while also in this numeric
134 range, doesn't actually has the same meaning, but it has shown to work
137 The other alternative, using for N+1 checks the coefficient of
138 variance of (N+1 fail=1, N+1 pass=0) across nodes could hint the
139 algorithm to make more N+1 failures if most nodes are N+1 fail
140 already. Since this (making N+1 failures) is not allowed by other
141 rules of the algorithm, so the N+1 checks would simply not work
142 anymore in this case.
144 The offline instances percentage (meaning the percentage of instances
145 living on offline nodes) will cause the algorithm to actively move
146 instances away from offline nodes. This, coupled with the restriction
147 on placement given by offline nodes, will cause evacuation of such
150 The dynamic load values need to be read from an external file (Ganeti
151 doesn't supply them), and are computed for each node as: sum of
152 primary instance cpu load, sum of primary instance memory load, sum of
153 primary and secondary instance disk load (as DRBD generates write load
154 on secondary nodes too in normal case and in degraded scenarios also
155 read load), and sum of primary instance network load. An example of
156 how to generate these values for input to hbal would be to track "xm
157 list" for instance over a day and by computing the delta of the cpu
158 values, and feed that via the \fI-U\fR option for all instances (and
159 keep the other metrics as one). For the algorithm to work, all that is
160 needed is that the values are consistent for a metric across all
161 instances (e.g. all instances use cpu% to report cpu usage, but they
162 could represent network bandwith in Gbps). Note that it's recommended
163 to not have zero as the load value for any instance metric since then
164 secondary instances are not well balanced.
166 On a perfectly balanced cluster (all nodes the same size, all
167 instances the same size and spread across the nodes equally), the
168 values for all metrics would be zero. This doesn't happen too often in
171 .SS OFFLINE INSTANCES
173 Since current Ganeti versions do not report the memory used by offline
174 (down) instances, ignoring the run status of instances will cause
175 wrong calculations. For this reason, the algorithm subtracts the
176 memory size of down instances from the free node memory of their
177 primary node, in effect simulating the startup of such instances.
179 .SS OTHER POSSIBLE METRICS
181 It would be desirable to add more metrics to the algorithm, especially
182 dynamically\(hycomputed metrics, such as:
186 CPU usage of instances
196 The options that can be passed to the program are as follows:
198 .B -C, --print-commands
199 Print the command list at the end of the run. Without this, the
200 program will only show a shorter, but cryptic output.
202 Note that the moves list will be split into independent steps, called
203 "jobsets", but only for visual inspection, not for actually
204 parallelisation. It is not possible to parallelise these directly when
205 executed via "gnt-instance" commands, since a compound command
206 (e.g. failover and replace\-disks) must be executed serially. Parallel
207 execution is only possible when using the Luxi backend and the
210 The algorithm for splitting the moves into jobsets is by accumulating
211 moves until the next move is touching nodes already touched by the
212 current moves; this means we can't execute in parallel (due to
213 resource allocation in Ganeti) and thus we start a new jobset.
217 Prints the before and after node status, in a format designed to allow
218 the user to understand the node's most important parameters.
220 It is possible to customise the listed information by passing a
221 comma\(hyseparated list of field names to this option (the field list is
222 currently undocumented). By default, the node list will contain these
227 a character denoting the status of the node, with '\-' meaning an
228 offline node, '*' meaning N+1 failure and blank meaning a good node
234 the total node memory
237 the memory used by the node itself
240 the memory used by instances
243 amount memory which seems to be in use but cannot be determined why or
244 by which instance; usually this means that the hypervisor has some
245 overhead or that there are other reporting errors
251 the reserved node memory, which is the amount of free memory needed
261 the number of physical cpus on the node
264 the number of virtual cpus allocated to primary instances
267 number of primary instances
270 number of secondary instances
273 percent of free memory
279 ratio of virtual to physical cpus
282 the dynamic CPU load (if the information is available)
285 the dynamic memory load (if the information is available)
288 the dynamic disk load (if the information is available)
291 the dynamic net load (if the information is available)
296 Prints the before and after instance map. This is less useful as the
297 node status, but it can help in understanding instance moves.
301 Only shows a one\(hyline output from the program, designed for the case
302 when one wants to look at multiple clusters at once and check their
305 The line will contain four fields:
310 initial cluster score
313 number of steps in the solution
319 improvement in the cluster score
325 This option (which can be given multiple times) will mark nodes as
326 being \fIoffline\fR. This means a couple of things:
331 instances won't be placed on these nodes, not even temporarily;
332 e.g. the \fIreplace primary\fR move is not available if the secondary
333 node is offline, since this move requires a failover.
336 these nodes will not be included in the score calculation (except for
337 the percentage of instances on offline nodes)
339 Note that hbal will also mark as offline any nodes which are reported
340 by RAPI as such, or that have "?" in file\(hybased input in any numeric
345 .BI "-e" score ", --min-score=" score
346 This parameter denotes the minimum score we are happy with and alters
347 the computation in two ways:
352 if the cluster has the initial score lower than this value, then we
353 don't enter the algorithm at all, and exit with success
356 during the iterative process, if we reach a score lower than this
357 value, we exit the algorithm
359 The default value of the parameter is currently \fI1e-9\fR (chosen
364 .BI "--no-disk-moves"
365 This parameter prevents hbal from using disk move (i.e. "gnt\-instance
366 replace\-disks") operations. This will result in a much quicker
367 balancing, but of course the improvements are limited. It is up to the
368 user to decide when to use one or another.
372 This parameter specifies a file holding instance dynamic utilisation
373 information that will be used to tweak the balancing algorithm to
374 equalise load on the nodes (as opposed to static resource usage). The
375 file is in the format "instance_name cpu_util mem_util disk_util
376 net_util" where the "_util" parameters are interpreted as numbers and
377 the instance name must match exactly the instance as read from
378 Ganeti. In case of unknown instance names, the program will abort.
380 If not given, the default values are one for all metrics and thus
381 dynamic utilisation has only one effect on the algorithm: the
382 equalisation of the secondary instances across nodes (this is the only
383 metric that is not tracked by another, dedicated value, and thus the
384 disk load of instances will cause secondary instance
385 equalisation). Note that value of one will also influence slightly the
386 primary instance count, but that is already tracked via other metrics
387 and thus the influence of the dynamic utilisation will be practically
391 .BI "-n" nodefile ", --nodes=" nodefile
392 The name of the file holding node information (if not collecting via
393 RAPI), instead of the default \fInodes\fR file (but see below how to
394 customize the default value via the environment).
397 .BI "-i" instancefile ", --instances=" instancefile
398 The name of the file holding instance information (if not collecting
399 via RAPI), instead of the default \fIinstances\fR file (but see below
400 how to customize the default value via the environment).
404 Collect data not from files but directly from the
406 given as an argument via RAPI. If the argument doesn't contain a colon
407 (:), then it is converted into a fully\(hybuilt URL via prepending
408 https:// and appending the default RAPI port, otherwise it's
409 considered a fully\(hyspecified URL and is used as\(hyis.
413 Collect data not from files but directly from the master daemon, which
414 is to be contacted via the luxi (an internal Ganeti protocol). An
415 optional \fIpath\fR argument is interpreted as the path to the unix
416 socket on which the master daemon listens; otherwise, the default path
417 used by ganeti when installed with \fI--localstatedir=/var\fR is used.
421 When using the Luxi backend, hbal can also execute the given
422 commands. The execution method is to execute the individual jobsets
423 (see the \fI-C\fR option for details) in separate stages, aborting if
424 at any time a jobset doesn't have all jobs successful. Each step in
425 the balancing solution will be translated into exactly one Ganeti job
426 (having between one and three OpCodes), and all the steps in a jobset
427 will be executed in parallel. The jobsets themselves are executed
431 .BI "-l" N ", --max-length=" N
432 Restrict the solution to this length. This can be used for example to
433 automate the execution of the balancing.
436 .BI "--max-cpu " cpu-ratio
437 The maximum virtual\(hyto\(hyphysical cpu ratio, as a floating point
438 number between zero and one. For example, specifying \fIcpu-ratio\fR
439 as \fB2.5\fR means that, for a 4\(hycpu machine, a maximum of 10
440 virtual cpus should be allowed to be in use for primary instances. A
441 value of one doesn't make sense though, as that means no disk space
445 .BI "--min-disk " disk-ratio
446 The minimum amount of free disk space remaining, as a floating point
447 number. For example, specifying \fIdisk-ratio\fR as \fB0.25\fR means
448 that at least one quarter of disk space should be left free on nodes.
452 Increase the output verbosity. Each usage of this option will increase
453 the verbosity (currently more than 2 doesn't make sense) from the
458 Decrease the output verbosity. Each usage of this option will decrease
459 the verbosity (less than zero doesn't make sense) from the default of
464 Just show the program version and exit.
468 The exist status of the command will be zero, unless for some reason
469 the algorithm fatally failed (e.g. wrong node or instance data).
473 If the variables \fBHTOOLS_NODES\fR and \fBHTOOLS_INSTANCES\fR are
474 present in the environment, they will override the default names for
475 the nodes and instances files. These will have of course no effect
476 when the RAPI or Luxi backends are used.
480 The program does not check its input data for consistency, and aborts
481 with cryptic errors messages in this case.
483 The algorithm is not perfect.
485 The output format is not easily scriptable, and the program should
486 feed moves directly into Ganeti (either via RAPI or via a gnt\-debug
491 Note that this example are not for the latest version (they don't have
496 With the default options, the program shows each individual step and
497 the improvements it brings in cluster score:
502 Loaded 20 nodes, 80 instances
503 Cluster is not N+1 happy, continuing but no guarantee that the cluster will end N+1 happy.
504 Initial score: 0.52329131
505 Trying to minimize the CV...
506 1. instance14 node1:node10 => node16:node10 0.42109120 a=f r:node16 f
507 2. instance54 node4:node15 => node16:node15 0.31904594 a=f r:node16 f
508 3. instance4 node5:node2 => node2:node16 0.26611015 a=f r:node16
509 4. instance48 node18:node20 => node2:node18 0.21361717 a=r:node2 f
510 5. instance93 node19:node18 => node16:node19 0.16166425 a=r:node16 f
511 6. instance89 node3:node20 => node2:node3 0.11005629 a=r:node2 f
512 7. instance5 node6:node2 => node16:node6 0.05841589 a=r:node16 f
513 8. instance94 node7:node20 => node20:node16 0.00658759 a=f r:node16
514 9. instance44 node20:node2 => node2:node15 0.00438740 a=f r:node15
515 10. instance62 node14:node18 => node14:node16 0.00390087 a=r:node16
516 11. instance13 node11:node14 => node11:node16 0.00361787 a=r:node16
517 12. instance19 node10:node11 => node10:node7 0.00336636 a=r:node7
518 13. instance43 node12:node13 => node12:node1 0.00305681 a=r:node1
519 14. instance1 node1:node2 => node1:node4 0.00263124 a=r:node4
520 15. instance58 node19:node20 => node19:node17 0.00252594 a=r:node17
521 Cluster score improved from 0.52329131 to 0.00252594
525 In the above output, we can see:
526 - the input data (here from files) shows a cluster with 20 nodes and
528 - the cluster is not initially N+1 compliant
529 - the initial score is 0.52329131
531 The step list follows, showing the instance, its initial
532 primary/secondary nodes, the new primary secondary, the cluster list,
533 and the actions taken in this step (with 'f' denoting failover/migrate
534 and 'r' denoting replace secondary).
536 Finally, the program shows the improvement in cluster score.
538 A more detailed output is obtained via the \fB-C\fR and \fB-p\fR options:
543 Loaded 20 nodes, 80 instances
544 Cluster is not N+1 happy, continuing but no guarantee that the cluster will end N+1 happy.
545 Initial cluster status:
546 N1 Name t_mem f_mem r_mem t_dsk f_dsk pri sec p_fmem p_fdsk
547 * node1 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
548 node2 32762 31280 12000 1861 1026 0 8 0.95476 0.55179
549 * node3 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
550 * node4 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
551 * node5 32762 1280 6000 1861 978 5 5 0.03907 0.52573
552 * node6 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
553 * node7 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
554 node8 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
555 node9 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
556 * node10 32762 7280 12000 1861 1026 4 4 0.22221 0.55179
557 node11 32762 7280 6000 1861 922 4 5 0.22221 0.49577
558 node12 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
559 node13 32762 7280 6000 1861 922 4 5 0.22221 0.49577
560 node14 32762 7280 6000 1861 922 4 5 0.22221 0.49577
561 * node15 32762 7280 12000 1861 1131 4 3 0.22221 0.60782
562 node16 32762 31280 0 1861 1860 0 0 0.95476 1.00000
563 node17 32762 7280 6000 1861 1106 5 3 0.22221 0.59479
564 * node18 32762 1280 6000 1396 561 5 3 0.03907 0.40239
565 * node19 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
566 node20 32762 13280 12000 1861 689 3 9 0.40535 0.37068
568 Initial score: 0.52329131
569 Trying to minimize the CV...
570 1. instance14 node1:node10 => node16:node10 0.42109120 a=f r:node16 f
571 2. instance54 node4:node15 => node16:node15 0.31904594 a=f r:node16 f
572 3. instance4 node5:node2 => node2:node16 0.26611015 a=f r:node16
573 4. instance48 node18:node20 => node2:node18 0.21361717 a=r:node2 f
574 5. instance93 node19:node18 => node16:node19 0.16166425 a=r:node16 f
575 6. instance89 node3:node20 => node2:node3 0.11005629 a=r:node2 f
576 7. instance5 node6:node2 => node16:node6 0.05841589 a=r:node16 f
577 8. instance94 node7:node20 => node20:node16 0.00658759 a=f r:node16
578 9. instance44 node20:node2 => node2:node15 0.00438740 a=f r:node15
579 10. instance62 node14:node18 => node14:node16 0.00390087 a=r:node16
580 11. instance13 node11:node14 => node11:node16 0.00361787 a=r:node16
581 12. instance19 node10:node11 => node10:node7 0.00336636 a=r:node7
582 13. instance43 node12:node13 => node12:node1 0.00305681 a=r:node1
583 14. instance1 node1:node2 => node1:node4 0.00263124 a=r:node4
584 15. instance58 node19:node20 => node19:node17 0.00252594 a=r:node17
585 Cluster score improved from 0.52329131 to 0.00252594
587 Commands to run to reach the above solution:
589 echo gnt\-instance migrate instance14
590 echo gnt\-instance replace\-disks \-n node16 instance14
591 echo gnt\-instance migrate instance14
593 echo gnt\-instance migrate instance54
594 echo gnt\-instance replace\-disks \-n node16 instance54
595 echo gnt\-instance migrate instance54
597 echo gnt\-instance migrate instance4
598 echo gnt\-instance replace\-disks \-n node16 instance4
600 echo gnt\-instance replace\-disks \-n node2 instance48
601 echo gnt\-instance migrate instance48
603 echo gnt\-instance replace\-disks \-n node16 instance93
604 echo gnt\-instance migrate instance93
606 echo gnt\-instance replace\-disks \-n node2 instance89
607 echo gnt\-instance migrate instance89
609 echo gnt\-instance replace\-disks \-n node16 instance5
610 echo gnt\-instance migrate instance5
612 echo gnt\-instance migrate instance94
613 echo gnt\-instance replace\-disks \-n node16 instance94
615 echo gnt\-instance migrate instance44
616 echo gnt\-instance replace\-disks \-n node15 instance44
618 echo gnt\-instance replace\-disks \-n node16 instance62
620 echo gnt\-instance replace\-disks \-n node16 instance13
622 echo gnt\-instance replace\-disks \-n node7 instance19
624 echo gnt\-instance replace\-disks \-n node1 instance43
626 echo gnt\-instance replace\-disks \-n node4 instance1
628 echo gnt\-instance replace\-disks \-n node17 instance58
630 Final cluster status:
631 N1 Name t_mem f_mem r_mem t_dsk f_dsk pri sec p_fmem p_fdsk
632 node1 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
633 node2 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
634 node3 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
635 node4 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
636 node5 32762 7280 6000 1861 1078 4 5 0.22221 0.57947
637 node6 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
638 node7 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
639 node8 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
640 node9 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
641 node10 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
642 node11 32762 7280 6000 1861 1022 4 4 0.22221 0.54951
643 node12 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
644 node13 32762 7280 6000 1861 1022 4 4 0.22221 0.54951
645 node14 32762 7280 6000 1861 1022 4 4 0.22221 0.54951
646 node15 32762 7280 6000 1861 1031 4 4 0.22221 0.55408
647 node16 32762 7280 6000 1861 1060 4 4 0.22221 0.57007
648 node17 32762 7280 6000 1861 1006 5 4 0.22221 0.54105
649 node18 32762 7280 6000 1396 761 4 2 0.22221 0.54570
650 node19 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
651 node20 32762 13280 6000 1861 1089 3 5 0.40535 0.58565
656 Here we see, beside the step list, the initial and final cluster
657 status, with the final one showing all nodes being N+1 compliant, and
658 the command list to reach the final solution. In the initial listing,
659 we see which nodes are not N+1 compliant.
661 The algorithm is stable as long as each step above is fully completed,
662 e.g. in step 8, both the migrate and the replace\-disks are
663 done. Otherwise, if only the migrate is done, the input data is
664 changed in a way that the program will output a different solution
665 list (but hopefully will end in the same state).
668 .BR hspace "(1), " hscan "(1), " hail "(1), "
669 .BR ganeti "(7), " gnt-instance "(8), " gnt-node "(8)"
673 Copyright (C) 2009 Google Inc. Permission is granted to copy,
674 distribute and/or modify under the terms of the GNU General Public
675 License as published by the Free Software Foundation; either version 2
676 of the License, or (at your option) any later version.
678 On Debian systems, the complete text of the GNU General Public License
679 can be found in /usr/share/common-licenses/GPL.