root / README @ 61d67afe
History | View | Annotate | Download (4.4 kB)
1 |
Ganeti Cluster tools (ganeti-htools) |
---|---|
2 |
==================================== |
3 |
|
4 |
These are some simple cluster tools for fixing common allocation |
5 |
problems on Ganeti 2.0 clusters. |
6 |
|
7 |
Note that these tools are most useful for bigger cluster sizes |
8 |
(e.g. more than five or ten machines); at lower sizes, the |
9 |
computations they do can also be done manually. |
10 |
|
11 |
Most of the tools revolve around the concept of keeping the cluster |
12 |
N+1 compliant: this means that in case of failure of any node, the |
13 |
instances affected can be failed over (via ``gnt-node failover`` or |
14 |
``gnt-instance failover``) to their secondary node, and there is |
15 |
enough memory reserved for this operation without needing to shutdown |
16 |
other instances or rebalance the cluster. |
17 |
|
18 |
**Quick start** (see the installation section for more details): |
19 |
|
20 |
- (have the ghc compiler and the prerequisite libraries installed) |
21 |
- make |
22 |
- ./hbal -m $cluster -C -p |
23 |
- look at the original and final cluster layout, and if acceptable, |
24 |
execute the given commands |
25 |
|
26 |
|
27 |
Available tools |
28 |
--------------- |
29 |
|
30 |
Cluster rebalancer |
31 |
~~~~~~~~~~~~~~~~~~ |
32 |
|
33 |
The rebalancer uses a simple algorithm to try to get the nodes of the |
34 |
cluster as equal as possible in their resource usage. It tries to |
35 |
repeatedly move each instance one step, so that the cluster score |
36 |
becomes better. We stop when no further move can improve the score. |
37 |
|
38 |
For algorithm details and usage, see the man page hbal(1). |
39 |
|
40 |
Cluster N+1 solver |
41 |
~~~~~~~~~~~~~~~~~~ |
42 |
|
43 |
This program runs a very simple brute force algorithm over the instance |
44 |
placement space in order to determine the shortest number of replace-disks |
45 |
needed to fix the cluster. Note this means we won't get a balanced cluster, |
46 |
just one that passes N+1 checks. |
47 |
|
48 |
For algorithm details and usage, see the man page hn1(1). |
49 |
|
50 |
.. note:: This program is deprecated, hbal should be used instead. |
51 |
|
52 |
IAllocator plugin |
53 |
~~~~~~~~~~~~~~~~~ |
54 |
|
55 |
The ``hail`` iallocator plugin can be used for allocations of mirrored |
56 |
and non-mirrored instances and for relocations of mirrored |
57 |
instances. It needs to be installed in Ganeti's iallocator search |
58 |
path—usually ``/usr/lib/ganeti/iallocators`` or |
59 |
``/usr/local/lib/ganeti/iallocators``, and after that it can be used |
60 |
via ganeti's ``--iallocator`` option (in various gnt-node/gnt-instance |
61 |
commands). See the man page hail(1) for more details. |
62 |
|
63 |
Cluster capacity estimator |
64 |
~~~~~~~~~~~~~~~~~~~~~~~~~~ |
65 |
|
66 |
The ``hspace`` program will, given an input instance specification, |
67 |
estimate how many instances of those type can be place on the cluster |
68 |
before it will become full (as in any new allocation would fail N+1 |
69 |
checks). For more details, see the man page hspace(1). |
70 |
|
71 |
|
72 |
Integration with Ganeti |
73 |
----------------------- |
74 |
|
75 |
The ``hbal``, ``hspace`` and ``hn1`` programs can either get their |
76 |
input from text files, or online from a cluster via RAPI. For online |
77 |
collection via RAPI, the "-m" argument to both hn1 and hbal should |
78 |
specify the cluster or master node name. ``hail`` uses the standard |
79 |
iallocator API and thus doesn't need any special setup (just needs to |
80 |
be installed in the right directory). |
81 |
|
82 |
For generating the text files, a separate tool (``hscan``) is provided |
83 |
to automate their gathering if RAPI is available, which is better |
84 |
since it can extract more precise information. In case RAPI is not |
85 |
usable for whatever reason, the following two commands should be run:: |
86 |
|
87 |
gnt-node list -oname,mtotal,mnode,mfree,dtotal,dfree,offline \ |
88 |
--separator '|' --no-headers > nodes |
89 |
gnt-instance list -oname,be/memory,sda_size,status,pnode,snodes \ |
90 |
--separator '|' --no-head > instances |
91 |
|
92 |
These two files should be saved under the names of *nodes* and |
93 |
*instances*. |
94 |
|
95 |
The ``hail`` program gets its data automatically from Ganeti when used |
96 |
as described in its section. |
97 |
|
98 |
Installation |
99 |
------------ |
100 |
|
101 |
If installing from source, you need a working ghc compiler (6.8 at |
102 |
least) and some extra Haskell libraries which usually need to be |
103 |
installed manually: |
104 |
|
105 |
- json (http://hackage.haskell.org/cgi-bin/hackage-scripts/package/json) |
106 |
- curl (http://hackage.haskell.org/cgi-bin/hackage-scripts/package/curl) |
107 |
|
108 |
Once these are installed, just typing *make* in the top-level |
109 |
directory should be enough. |
110 |
|
111 |
Only the ``hail`` program needs to be installed in a specific place, |
112 |
the other tools are not location-dependent. |
113 |
|
114 |
For running the (admittedly small) unittest suite (via *make check*), |
115 |
the QuickCheck version 1 library is needed. |
116 |
|
117 |
Internal (implementation) documentation is available in the ``apidoc`` |
118 |
directory. |
119 |
|
120 |
.. Local Variables: |
121 |
.. mode: rst |
122 |
.. End: |