Statistics
| Branch: | Tag: | Revision:

root / man / hspace.rst @ 2ef8013f

History | View | Annotate | Download (14 kB)

1
HSPACE(1) Ganeti | Version @GANETI_VERSION@
2
===========================================
3

    
4
NAME
5
----
6

    
7
hspace - Cluster space analyzer for Ganeti
8

    
9
SYNOPSIS
10
--------
11

    
12
**hspace** {backend options...} [algorithm options...] [request options...]
13
[output options...] [-v... | -q]
14

    
15
**hspace** --version
16

    
17
Backend options:
18

    
19
{ **-m** *cluster* | **-L[** *path* **] [-X]** | **-t** *data-file* |
20
**--simulate** *spec* }
21

    
22

    
23
Algorithm options:
24

    
25
**[ --max-cpu *cpu-ratio* ]**
26
**[ --min-disk *disk-ratio* ]**
27
**[ -O *name...* ]**
28

    
29

    
30
Request options:
31

    
32
**[--memory** *mem* **]**
33
**[--disk** *disk* **]**
34
**[--disk-template** *template* **]**
35
**[--vcpus** *vcpus* **]**
36
**[--tiered-alloc** *spec* **]**
37

    
38
Output options:
39

    
40
**[--machine-readable**[=*CHOICE*] **]**
41
**[-p**[*fields*]**]**
42

    
43

    
44
DESCRIPTION
45
-----------
46

    
47

    
48
hspace computes how many additional instances can be fit on a cluster,
49
while maintaining N+1 status.
50

    
51
The program will try to place instances, all of the same size, on the
52
cluster, until the point where we don't have any N+1 possible
53
allocation. It uses the exact same allocation algorithm as the hail
54
iallocator plugin in *allocate* mode.
55

    
56
The output of the program is designed either for human consumption (the
57
default) or, when enabled with the ``--machine-readable`` option
58
(described further below), for machine consumption. In the latter case,
59
it is intended to interpreted as a shell fragment (or parsed as a
60
*key=value* file). Options which extend the output (e.g. -p, -v) will
61
output the additional information on stderr (such that the stdout is
62
still parseable).
63

    
64
The following keys are available in the machine-readable output of the
65
script (all prefixed with *HTS_*):
66

    
67
SPEC_MEM, SPEC_DSK, SPEC_CPU, SPEC_RQN, SPEC_DISK_TEMPLATE
68
  These represent the specifications of the instance model used for
69
  allocation (the memory, disk, cpu, requested nodes, disk template).
70

    
71
TSPEC_INI_MEM, TSPEC_INI_DSK, TSPEC_INI_CPU, ...
72
  Only defined when the tiered mode allocation is enabled, these are
73
  similar to the above specifications but show the initial starting spec
74
  for tiered allocation.
75

    
76
CLUSTER_MEM, CLUSTER_DSK, CLUSTER_CPU, CLUSTER_NODES
77
  These represent the total memory, disk, CPU count and total nodes in
78
  the cluster.
79

    
80
INI_SCORE, FIN_SCORE
81
  These are the initial (current) and final cluster score (see the hbal
82
  man page for details about the scoring algorithm).
83

    
84
INI_INST_CNT, FIN_INST_CNT
85
  The initial and final instance count.
86

    
87
INI_MEM_FREE, FIN_MEM_FREE
88
  The initial and final total free memory in the cluster (but this
89
  doesn't necessarily mean available for use).
90

    
91
INI_MEM_AVAIL, FIN_MEM_AVAIL
92
  The initial and final total available memory for allocation in the
93
  cluster. If allocating redundant instances, new instances could
94
  increase the reserved memory so it doesn't necessarily mean the
95
  entirety of this memory can be used for new instance allocations.
96

    
97
INI_MEM_RESVD, FIN_MEM_RESVD
98
  The initial and final reserved memory (for redundancy/N+1 purposes).
99

    
100
INI_MEM_INST, FIN_MEM_INST
101
  The initial and final memory used for instances (actual runtime used
102
  RAM).
103

    
104
INI_MEM_OVERHEAD, FIN_MEM_OVERHEAD
105
  The initial and final memory overhead--memory used for the node
106
  itself and unacounted memory (e.g. due to hypervisor overhead).
107

    
108
INI_MEM_EFF, HTS_INI_MEM_EFF
109
  The initial and final memory efficiency, represented as instance
110
  memory divided by total memory.
111

    
112
INI_DSK_FREE, INI_DSK_AVAIL, INI_DSK_RESVD, INI_DSK_INST, INI_DSK_EFF
113
  Initial disk stats, similar to the memory ones.
114

    
115
FIN_DSK_FREE, FIN_DSK_AVAIL, FIN_DSK_RESVD, FIN_DSK_INST, FIN_DSK_EFF
116
  Final disk stats, similar to the memory ones.
117

    
118
INI_CPU_INST, FIN_CPU_INST
119
  Initial and final number of virtual CPUs used by instances.
120

    
121
INI_CPU_EFF, FIN_CPU_EFF
122
  The initial and final CPU efficiency, represented as the count of
123
  virtual instance CPUs divided by the total physical CPU count.
124

    
125
INI_MNODE_MEM_AVAIL, FIN_MNODE_MEM_AVAIL
126
  The initial and final maximum per-node available memory. This is not
127
  very useful as a metric but can give an impression of the status of
128
  the nodes; as an example, this value restricts the maximum instance
129
  size that can be still created on the cluster.
130

    
131
INI_MNODE_DSK_AVAIL, FIN_MNODE_DSK_AVAIL
132
  Like the above but for disk.
133

    
134
TSPEC
135
  If the tiered allocation mode has been enabled, this parameter holds
136
  the pairs of specifications and counts of instances that can be
137
  created in this mode. The value of the key is a space-separated list
138
  of values; each value is of the form *memory,disk,vcpu=count* where
139
  the memory, disk and vcpu are the values for the current spec, and
140
  count is how many instances of this spec can be created. A complete
141
  value for this variable could be: **4096,102400,2=225
142
  2560,102400,2=20 512,102400,2=21**.
143

    
144
KM_USED_CPU, KM_USED_NPU, KM_USED_MEM, KM_USED_DSK
145
  These represents the metrics of used resources at the start of the
146
  computation (only for tiered allocation mode). The NPU value is
147
  "normalized" CPU count, i.e. the number of virtual CPUs divided by
148
  the maximum ratio of the virtual to physical CPUs.
149

    
150
KM_POOL_CPU, KM_POOL_NPU, KM_POOL_MEM, KM_POOL_DSK
151
  These represents the total resources allocated during the tiered
152
  allocation process. In effect, they represent how much is readily
153
  available for allocation.
154

    
155
KM_UNAV_CPU, KM_POOL_NPU, KM_UNAV_MEM, KM_UNAV_DSK
156
  These represents the resources left over (either free as in
157
  unallocable or allocable on their own) after the tiered allocation
158
  has been completed. They represent better the actual unallocable
159
  resources, because some other resource has been exhausted. For
160
  example, the cluster might still have 100GiB disk free, but with no
161
  memory left for instances, we cannot allocate another instance, so
162
  in effect the disk space is unallocable. Note that the CPUs here
163
  represent instance virtual CPUs, and in case the *--max-cpu* option
164
  hasn't been specified this will be -1.
165

    
166
ALLOC_USAGE
167
  The current usage represented as initial number of instances divided
168
  per final number of instances.
169

    
170
ALLOC_COUNT
171
  The number of instances allocated (delta between FIN_INST_CNT and
172
  INI_INST_CNT).
173

    
174
ALLOC_FAIL*_CNT
175
  For the last attemp at allocations (which would have increased
176
  FIN_INST_CNT with one, if it had succeeded), this is the count of
177
  the failure reasons per failure type; currently defined are FAILMEM,
178
  FAILDISK and FAILCPU which represent errors due to not enough
179
  memory, disk and CPUs, and FAILN1 which represents a non N+1
180
  compliant cluster on which we can't allocate instances at all.
181

    
182
ALLOC_FAIL_REASON
183
  The reason for most of the failures, being one of the above FAIL*
184
  strings.
185

    
186
OK
187
  A marker representing the successful end of the computation, and
188
  having value "1". If this key is not present in the output it means
189
  that the computation failed and any values present should not be
190
  relied upon.
191

    
192
If the tiered allocation mode is enabled, then many of the INI_/FIN_
193
metrics will be also displayed with a TRL_ prefix, and denote the
194
cluster status at the end of the tiered allocation run.
195

    
196
The human output format should be self-explanatory, so it is not
197
described further.
198

    
199
OPTIONS
200
-------
201

    
202
The options that can be passed to the program are as follows:
203

    
204
--memory *mem*
205
  The memory size of the instances to be placed (defaults to
206
  4GiB). Units can be used (see below for more details).
207

    
208
--disk *disk*
209
  The disk size of the instances to be placed (defaults to
210
  100GiB). Units can be used.
211

    
212
--disk-template *template*
213
  The disk template for the instance; one of the Ganeti disk templates
214
  (e.g. plain, drbd, so on) should be passed in.
215

    
216
--vcpus *vcpus*
217
  The number of VCPUs of the instances to be placed (defaults to 1).
218

    
219
--max-cpu=*cpu-ratio*
220
  The maximum virtual to physical cpu ratio, as a floating point number
221
  greater than or equal to one. For example, specifying *cpu-ratio* as
222
  **2.5** means that, for a 4-cpu machine, a maximum of 10 virtual cpus
223
  should be allowed to be in use for primary instances. A value of
224
  exactly one means there will be no over-subscription of CPU (except
225
  for the CPU time used by the node itself), and values below one do not
226
  make sense, as that means other resources (e.g. disk) won't be fully
227
  utilised due to CPU restrictions.
228

    
229
--min-disk=*disk-ratio*
230
  The minimum amount of free disk space remaining, as a floating point
231
  number. For example, specifying *disk-ratio* as **0.25** means that
232
  at least one quarter of disk space should be left free on nodes.
233

    
234
-p, --print-nodes
235
  Prints the before and after node status, in a format designed to allow
236
  the user to understand the node's most important parameters. See the
237
  man page **htools**(1) for more details about this option.
238

    
239
-O *name*
240
  This option (which can be given multiple times) will mark nodes as
241
  being *offline*. This means a couple of things:
242

    
243
  - instances won't be placed on these nodes, not even temporarily;
244
    e.g. the *replace primary* move is not available if the secondary
245
    node is offline, since this move requires a failover.
246
  - these nodes will not be included in the score calculation (except
247
    for the percentage of instances on offline nodes)
248

    
249
  Note that the algorithm will also mark as offline any nodes which
250
  are reported by RAPI as such, or that have "?" in file-based input
251
  in any numeric fields.
252

    
253
-t *datafile*, --text-data=*datafile*
254
  The name of the file holding node and instance information (if not
255
  collecting via RAPI or LUXI). This or one of the other backends must
256
  be selected.
257

    
258
-S *filename*, --save-cluster=*filename*
259
  If given, the state of the cluster at the end of the allocation is
260
  saved to a file named *filename.alloc*, and if tiered allocation is
261
  enabled, the state after tiered allocation will be saved to
262
  *filename.tiered*. This allows re-feeding the cluster state to
263
  either hspace itself (with different parameters) or for example
264
  hbal.
265

    
266
-m *cluster*
267
 Collect data directly from the *cluster* given as an argument via
268
 RAPI. If the argument doesn't contain a colon (:), then it is
269
 converted into a fully-built URL via prepending ``https://`` and
270
 appending the default RAPI port, otherwise it's considered a
271
 fully-specified URL and is used as-is.
272

    
273
-L [*path*]
274
  Collect data directly from the master daemon, which is to be
275
  contacted via the luxi (an internal Ganeti protocol). An optional
276
  *path* argument is interpreted as the path to the unix socket on
277
  which the master daemon listens; otherwise, the default path used by
278
  ganeti when installed with *--localstatedir=/var* is used.
279

    
280
--simulate *description*
281
  Instead of using actual data, build an empty cluster given a node
282
  description. The *description* parameter must be a comma-separated
283
  list of five elements, describing in order:
284

    
285
  - the allocation policy for this node group (*preferred*, *allocable*
286
    or *unallocable*, or alternatively the short forms *p*, *a* or *u*)
287
  - the number of nodes in the cluster
288
  - the disk size of the nodes (default in mebibytes, units can be used)
289
  - the memory size of the nodes (default in mebibytes, units can be used)
290
  - the cpu core count for the nodes
291

    
292
  An example description would be **preferred,B20,100G,16g,4**
293
  describing a 20-node cluster where each node has 100GB of disk
294
  space, 16GiB of memory and 4 CPU cores. Note that all nodes must
295
  have the same specs currently.
296

    
297
  This option can be given multiple times, and each new use defines a
298
  new node group. Hence different node groups can have different
299
  allocation policies and node count/specifications.
300

    
301
--tiered-alloc *spec*
302
  Besides the standard, fixed-size allocation, also do a tiered
303
  allocation scheme where the algorithm starts from the given
304
  specification and allocates until there is no more space; then it
305
  decreases the specification and tries the allocation again. The
306
  decrease is done on the matric that last failed during
307
  allocation. The specification given is similar to the *--simulate*
308
  option and it holds:
309

    
310
  - the disk size of the instance (units can be used)
311
  - the memory size of the instance (units can be used)
312
  - the vcpu count for the insance
313

    
314
  An example description would be *100G,4g,2* describing an initial
315
  starting specification of 100GB of disk space, 4GiB of memory and 2
316
  VCPUs.
317

    
318
  Also note that the normal allocation and the tiered allocation are
319
  independent, and both start from the initial cluster state; as such,
320
  the instance count for these two modes are not related one to
321
  another.
322

    
323
--machines-readable[=*choice*]
324
  By default, the output of the program is in "human-readable" format,
325
  i.e. text descriptions. By passing this flag you can either enable
326
  (``--machine-readable`` or ``--machine-readable=yes``) or explicitly
327
  disable (``--machine-readable=no``) the machine readable format
328
  described above.
329

    
330
-v, --verbose
331
  Increase the output verbosity. Each usage of this option will
332
  increase the verbosity (currently more than 2 doesn't make sense)
333
  from the default of one.
334

    
335
-q, --quiet
336
  Decrease the output verbosity. Each usage of this option will
337
  decrease the verbosity (less than zero doesn't make sense) from the
338
  default of one.
339

    
340
-V, --version
341
  Just show the program version and exit.
342

    
343
UNITS
344
~~~~~
345

    
346
By default, all unit-accepting options use mebibytes. Using the
347
lower-case letters of *m*, *g* and *t* (or their longer equivalents of
348
*mib*, *gib*, *tib*, for which case doesn't matter) explicit binary
349
units can be selected. Units in the SI system can be selected using the
350
upper-case letters of *M*, *G* and *T* (or their longer equivalents of
351
*MB*, *GB*, *TB*, for which case doesn't matter).
352

    
353
More details about the difference between the SI and binary systems can
354
be read in the *units(7)* man page.
355

    
356
EXIT STATUS
357
-----------
358

    
359
The exist status of the command will be zero, unless for some reason
360
the algorithm fatally failed (e.g. wrong node or instance data).
361

    
362
BUGS
363
----
364

    
365
The algorithm is highly dependent on the number of nodes; its runtime
366
grows exponentially with this number, and as such is impractical for
367
really big clusters.
368

    
369
The algorithm doesn't rebalance the cluster or try to get the optimal
370
fit; it just allocates in the best place for the current step, without
371
taking into consideration the impact on future placements.
372

    
373
.. vim: set textwidth=72 :
374
.. Local Variables:
375
.. mode: rst
376
.. fill-column: 72
377
.. End: