1 .TH HBAL 1 2009-03-22 htools "Ganeti H-tools"
3 hbal \- Cluster balancer for Ganeti
12 .BI "[-m " cluster "]"
13 .BI "[-n " nodes-file " ]"
14 .BI "[-i " instances-file "]"
20 hbal is a cluster balancer that looks at the current state of the
21 cluster (nodes with their total and free disk, memory, etc.) and
22 instance placement and computes a series of steps designed to bring
23 the cluster into a better state.
25 The algorithm to do so is designed to be stable (i.e. it will give you
26 the same results when restarting it from the middle of the solution)
27 and reasonably fast. It is not, however, designed to be a perfect
28 algorithm - it is possible to make it go into a corner from which it
29 can find no improvement, because it only look one "step" ahead.
31 By default, the program will show the solution incrementally as it is
32 computed, in a somewhat cryptic format; for getting the actual Ganeti
33 command list, use the \fB-C\fR option.
37 The program works in independent steps; at each step, we compute the
38 best instance move that lowers the cluster score.
40 The possible move type for an instance are combinations of
41 failover/migrate and replace-disks such that we change one of the
42 instance nodes, and the other one remains (but possibly with changed
43 role, e.g. from primary it becomes secondary). The list is:
45 - replace secondary (r)
46 - replace primary, a composite move (f, r, f)
47 - failover and replace secondary, also composite (f, r)
48 - replace secondary and failover, also composite (r, f)
50 We don't do the only remaining possibility of replacing both nodes
51 (r,f,r,f or the equivalent f,r,f,r) since these move needs an
52 exhaustive search over both candidate primary and secondary nodes, and
53 is O(n*n) in the number of nodes. Furthermore, it doesn't seems to
54 give better scores but will result in more disk replacements.
58 As said before, the algorithm tries to minimise the cluster score at
59 each step. Currently this score is computed as a sum of the following
61 - coefficient of variance of the percent of free memory
62 - coefficient of variance of the percent of reserved memory
63 - coefficient of variance of the percent of free disk
64 - percentage of nodes failing N+1 check
65 - percentage of instances living (either as primary or secondary) on
68 The free memory and free disk values help ensure that all nodes are
69 somewhat balanced in their resource usage. The reserved memory helps
70 to ensure that nodes are somewhat balanced in holding secondary
71 instances, and that no node keeps too much memory reserved for
72 N+1. And finally, the N+1 percentage helps guide the algorithm towards
73 eliminating N+1 failures, if possible.
75 Except for the N+1 failures and offline instances percentage, we use
76 the coefficient of variance since this brings the values into the same
77 unit so to speak, and with a restrict domain of values (between zero
78 and one). The percentage of N+1 failures, while also in this numeric
79 range, doesn't actually has the same meaning, but it has shown to work
82 The other alternative, using for N+1 checks the coefficient of
83 variance of (N+1 fail=1, N+1 pass=0) across nodes could hint the
84 algorithm to make more N+1 failures if most nodes are N+1 fail
85 already. Since this (making N+1 failures) is not allowed by other
86 rules of the algorithm, so the N+1 checks would simply not work
89 The offline instances percentage (meaning the percentage of instances
90 living on offline nodes) will cause the algorithm to actively move
91 instances away from offline nodes. This, coupled with the restriction
92 on placement given by offline nodes, will cause evacuation of such
95 On a perfectly balanced cluster (all nodes the same size, all
96 instances the same size and spread across the nodes equally), all
97 values would be zero. This doesn't happen too often in practice :)
99 .SS OTHER POSSIBLE METRICS
101 It would be desirable to add more metrics to the algorithm, especially
102 dynamically-computed metrics, such as:
103 - CPU usage of instances, combined with VCPU versus PCPU count
108 The options that can be passed to the program are as follows:
110 .B -C, --print-commands
111 Print the command list at the end of the run. Without this, the
112 program will only show a shorter, but cryptic output.
115 Prints the before and after node status, in a format designed to allow
116 the user to understand the node's most important parameters.
118 The node list will contain these informations:
122 a character denoting the status of the node, with '-' meaning an
123 offline node, '*' meaning N+1 failure and blank meaning a good node
129 the total node memory
132 the memory used by the node itself
135 the memory used by instances
138 amount memory which seems to be in use but cannot be determined why or
139 by which instance; usually this means that the hypervisor has some
140 overhead or that there are other reporting errors
146 the reserved node memory, which is the amount of free memory needed
156 number of primary instances
159 number of secondary instances
162 percent of free memory
170 Only shows a one-line output from the program, designed for the case
171 when one wants to look at multiple clusters at once and check their
174 The line will contain four fields:
175 - initial cluster score
176 - number of steps in the solution
177 - final cluster score
178 - improvement in the cluster score
182 This option (which can be given multiple times) will mark nodes as
183 being \fIoffline\fR. This means a couple of things:
187 instances won't be placed on these nodes, not even temporarily;
188 e.g. the \fIreplace primary\fR move is not available if the secondary
189 node is offline, since this move requires a failover.
192 these nodes will not be included in the score calculation (except for
193 the percentage of instances on offline nodes)
197 .BI "-n" nodefile ", --nodes=" nodefile
198 The name of the file holding node information (if not collecting via
199 RAPI), instead of the default
204 .BI "-i" instancefile ", --instances=" instancefile
205 The name of the file holding instance information (if not collecting
206 via RAPI), instead of the default
212 Collect data not from files but directly from the
214 given as an argument via RAPI. This work for both Ganeti 1.2 and
218 .BI "-l" N ", --max-length=" N
219 Restrict the solution to this length. This can be used for example to
220 automate the execution of the balancing.
224 Increase the output verbosity. Each usage of this option will increase
225 the verbosity (currently more than 2 doesn't make sense) from the
230 Just show the program version and exit.
234 The exist status of the command will be zero, unless for some reason
235 the algorithm fatally failed (e.g. wrong node or instance data).
239 The program does not check its input data for consistency, and aborts
240 with cryptic errors messages in this case.
242 The algorithm is not perfect.
244 The output format is not easily scriptable, and the program should
245 feed moves directly into Ganeti (either via RAPI or via a gnt-debug
250 Note that this example are not for the latest version (they don't have
255 With the default options, the program shows each individual step and
256 the improvements it brings in cluster score:
261 Loaded 20 nodes, 80 instances
262 Cluster is not N+1 happy, continuing but no guarantee that the cluster will end N+1 happy.
263 Initial score: 0.52329131
264 Trying to minimize the CV...
265 1. instance14 node1:node10 => node16:node10 0.42109120 a=f r:node16 f
266 2. instance54 node4:node15 => node16:node15 0.31904594 a=f r:node16 f
267 3. instance4 node5:node2 => node2:node16 0.26611015 a=f r:node16
268 4. instance48 node18:node20 => node2:node18 0.21361717 a=r:node2 f
269 5. instance93 node19:node18 => node16:node19 0.16166425 a=r:node16 f
270 6. instance89 node3:node20 => node2:node3 0.11005629 a=r:node2 f
271 7. instance5 node6:node2 => node16:node6 0.05841589 a=r:node16 f
272 8. instance94 node7:node20 => node20:node16 0.00658759 a=f r:node16
273 9. instance44 node20:node2 => node2:node15 0.00438740 a=f r:node15
274 10. instance62 node14:node18 => node14:node16 0.00390087 a=r:node16
275 11. instance13 node11:node14 => node11:node16 0.00361787 a=r:node16
276 12. instance19 node10:node11 => node10:node7 0.00336636 a=r:node7
277 13. instance43 node12:node13 => node12:node1 0.00305681 a=r:node1
278 14. instance1 node1:node2 => node1:node4 0.00263124 a=r:node4
279 15. instance58 node19:node20 => node19:node17 0.00252594 a=r:node17
280 Cluster score improved from 0.52329131 to 0.00252594
284 In the above output, we can see:
285 - the input data (here from files) shows a cluster with 20 nodes and
287 - the cluster is not initially N+1 compliant
288 - the initial score is 0.52329131
290 The step list follows, showing the instance, its initial
291 primary/secondary nodes, the new primary secondary, the cluster list,
292 and the actions taken in this step (with 'f' denoting failover/migrate
293 and 'r' denoting replace secondary).
295 Finally, the program shows the improvement in cluster score.
297 A more detailed output is obtained via the \fB-C\fR and \fB-p\fR options:
302 Loaded 20 nodes, 80 instances
303 Cluster is not N+1 happy, continuing but no guarantee that the cluster will end N+1 happy.
304 Initial cluster status:
305 N1 Name t_mem f_mem r_mem t_dsk f_dsk pri sec p_fmem p_fdsk
306 * node1 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
307 node2 32762 31280 12000 1861 1026 0 8 0.95476 0.55179
308 * node3 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
309 * node4 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
310 * node5 32762 1280 6000 1861 978 5 5 0.03907 0.52573
311 * node6 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
312 * node7 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
313 node8 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
314 node9 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
315 * node10 32762 7280 12000 1861 1026 4 4 0.22221 0.55179
316 node11 32762 7280 6000 1861 922 4 5 0.22221 0.49577
317 node12 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
318 node13 32762 7280 6000 1861 922 4 5 0.22221 0.49577
319 node14 32762 7280 6000 1861 922 4 5 0.22221 0.49577
320 * node15 32762 7280 12000 1861 1131 4 3 0.22221 0.60782
321 node16 32762 31280 0 1861 1860 0 0 0.95476 1.00000
322 node17 32762 7280 6000 1861 1106 5 3 0.22221 0.59479
323 * node18 32762 1280 6000 1396 561 5 3 0.03907 0.40239
324 * node19 32762 1280 6000 1861 1026 5 3 0.03907 0.55179
325 node20 32762 13280 12000 1861 689 3 9 0.40535 0.37068
327 Initial score: 0.52329131
328 Trying to minimize the CV...
329 1. instance14 node1:node10 => node16:node10 0.42109120 a=f r:node16 f
330 2. instance54 node4:node15 => node16:node15 0.31904594 a=f r:node16 f
331 3. instance4 node5:node2 => node2:node16 0.26611015 a=f r:node16
332 4. instance48 node18:node20 => node2:node18 0.21361717 a=r:node2 f
333 5. instance93 node19:node18 => node16:node19 0.16166425 a=r:node16 f
334 6. instance89 node3:node20 => node2:node3 0.11005629 a=r:node2 f
335 7. instance5 node6:node2 => node16:node6 0.05841589 a=r:node16 f
336 8. instance94 node7:node20 => node20:node16 0.00658759 a=f r:node16
337 9. instance44 node20:node2 => node2:node15 0.00438740 a=f r:node15
338 10. instance62 node14:node18 => node14:node16 0.00390087 a=r:node16
339 11. instance13 node11:node14 => node11:node16 0.00361787 a=r:node16
340 12. instance19 node10:node11 => node10:node7 0.00336636 a=r:node7
341 13. instance43 node12:node13 => node12:node1 0.00305681 a=r:node1
342 14. instance1 node1:node2 => node1:node4 0.00263124 a=r:node4
343 15. instance58 node19:node20 => node19:node17 0.00252594 a=r:node17
344 Cluster score improved from 0.52329131 to 0.00252594
346 Commands to run to reach the above solution:
348 echo gnt-instance migrate instance14
349 echo gnt-instance replace-disks -n node16 instance14
350 echo gnt-instance migrate instance14
352 echo gnt-instance migrate instance54
353 echo gnt-instance replace-disks -n node16 instance54
354 echo gnt-instance migrate instance54
356 echo gnt-instance migrate instance4
357 echo gnt-instance replace-disks -n node16 instance4
359 echo gnt-instance replace-disks -n node2 instance48
360 echo gnt-instance migrate instance48
362 echo gnt-instance replace-disks -n node16 instance93
363 echo gnt-instance migrate instance93
365 echo gnt-instance replace-disks -n node2 instance89
366 echo gnt-instance migrate instance89
368 echo gnt-instance replace-disks -n node16 instance5
369 echo gnt-instance migrate instance5
371 echo gnt-instance migrate instance94
372 echo gnt-instance replace-disks -n node16 instance94
374 echo gnt-instance migrate instance44
375 echo gnt-instance replace-disks -n node15 instance44
377 echo gnt-instance replace-disks -n node16 instance62
379 echo gnt-instance replace-disks -n node16 instance13
381 echo gnt-instance replace-disks -n node7 instance19
383 echo gnt-instance replace-disks -n node1 instance43
385 echo gnt-instance replace-disks -n node4 instance1
387 echo gnt-instance replace-disks -n node17 instance58
389 Final cluster status:
390 N1 Name t_mem f_mem r_mem t_dsk f_dsk pri sec p_fmem p_fdsk
391 node1 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
392 node2 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
393 node3 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
394 node4 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
395 node5 32762 7280 6000 1861 1078 4 5 0.22221 0.57947
396 node6 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
397 node7 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
398 node8 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
399 node9 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
400 node10 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
401 node11 32762 7280 6000 1861 1022 4 4 0.22221 0.54951
402 node12 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
403 node13 32762 7280 6000 1861 1022 4 4 0.22221 0.54951
404 node14 32762 7280 6000 1861 1022 4 4 0.22221 0.54951
405 node15 32762 7280 6000 1861 1031 4 4 0.22221 0.55408
406 node16 32762 7280 6000 1861 1060 4 4 0.22221 0.57007
407 node17 32762 7280 6000 1861 1006 5 4 0.22221 0.54105
408 node18 32762 7280 6000 1396 761 4 2 0.22221 0.54570
409 node19 32762 7280 6000 1861 1026 4 4 0.22221 0.55179
410 node20 32762 13280 6000 1861 1089 3 5 0.40535 0.58565
415 Here we see, beside the step list, the initial and final cluster
416 status, with the final one showing all nodes being N+1 compliant, and
417 the command list to reach the final solution. In the initial listing,
418 we see which nodes are not N+1 compliant.
420 The algorithm is stable as long as each step above is fully completed,
421 e.g. in step 8, both the migrate and the replace-disks are
422 done. Otherwise, if only the migrate is done, the input data is
423 changed in a way that the program will output a different solution
424 list (but hopefully will end in the same state).
427 .BR hn1 "(1), " hscan "(1), " ganeti "(7), " gnt-instance "(8), "