Revision f30d0610 man/hbal.rst

b/man/hbal.rst
58 58
it is possible to make it go into a corner from which it can find no
59 59
improvement, because it looks only one "step" ahead.
60 60

  
61
The program accesses the cluster state via Rapi or Luxi. It also
62
requests data over the network from all MonDs with the --mond option.
63
Currently it uses only data produced by CPUload collector.
64

  
61 65
By default, the program will show the solution incrementally as it is
62 66
computed, in a somewhat cryptic format; for getting the actual Ganeti
63 67
command list, use the **-C** option.
......
121 125
  primary instances of the node)
122 126
- standard deviation of the dynamic load on the nodes, for cpus,
123 127
  memory, disk and network
128
- standard deviation of the CPU load provided by MonD
124 129

  
125 130
The free memory and free disk values help ensure that all nodes are
126 131
somewhat balanced in their resource usage. The reserved memory helps
......
159 164
that it's recommended to not have zero as the load value for any
160 165
instance metric since then secondary instances are not well balanced.
161 166

  
167
The CPUload from MonD's data collector will be used only if all MonDs
168
are running, otherwise it won't affect the cluster score. Since we can't
169
find the CPU load of each instance, we can assume that the CPU load of
170
an instance is proportional to the number of its vcpus. With this
171
heuristic, instances from nodes with high CPU load will tend to move to
172
nodes with less CPU load.
173

  
162 174
On a perfectly balanced cluster (all nodes the same size, all
163 175
instances the same size and spread across the nodes equally), the
164 176
values for all metrics would be zero. This doesn't happen too often in
......
320 332
\--ignore-dynu
321 333
  If given, all dynamic utilisation information will be ignored by
322 334
  assuming it to be 0. This option will take precedence over any data
323
  passed by the ``-U`` option.
335
  passed by the ``-U`` option or by the MonDs with the ``--mond`` and
336
  the ``--mond-data`` option.
324 337

  
325 338
-S *filename*, \--save-cluster=*filename*
326 339
  If given, the state of the cluster before the balancing is saved to
......
336 349
  other backends must be selected. The option is described in the man
337 350
  page **htools**\(1).
338 351

  
352
\--mond
353
  If given the program will query all MonDs to fetch data from the
354
  supported data collectors over the network.
355

  
356
\--mond-data *datafile*
357
  The name of the file holding the data provided by MonD, to override
358
  quering MonDs over the network. This is mostly used for debugging. The
359
  file must be in JSON format and present an array of JSON objects ,
360
  one for every node, with two members. The first member named ``node``
361
  is the name of the node and the second member named ``reports`` is an
362
  array of report objects. The report objects must be in the same format
363
  as produced by the monitoring agent.
364

  
339 365
-m *cluster*
340 366
  Backend specification: collect data directly from the *cluster* given
341 367
  as an argument via RAPI. The option is described in the man page

Also available in: Unified diff