root / htools / Ganeti / HTools / Cluster.hs @ b003b8c0
History | View | Annotate | Download (62.1 kB)
1 | e4f08c46 | Iustin Pop | {-| Implementation of cluster-wide logic. |
---|---|---|---|
2 | e4f08c46 | Iustin Pop | |
3 | e4f08c46 | Iustin Pop | This module holds all pure cluster-logic; I\/O related functionality |
4 | 525bfb36 | Iustin Pop | goes into the /Main/ module for the individual binaries. |
5 | e4f08c46 | Iustin Pop | |
6 | e4f08c46 | Iustin Pop | -} |
7 | e4f08c46 | Iustin Pop | |
8 | e2fa2baf | Iustin Pop | {- |
9 | e2fa2baf | Iustin Pop | |
10 | aa5b2f07 | Iustin Pop | Copyright (C) 2009, 2010, 2011, 2012 Google Inc. |
11 | e2fa2baf | Iustin Pop | |
12 | e2fa2baf | Iustin Pop | This program is free software; you can redistribute it and/or modify |
13 | e2fa2baf | Iustin Pop | it under the terms of the GNU General Public License as published by |
14 | e2fa2baf | Iustin Pop | the Free Software Foundation; either version 2 of the License, or |
15 | e2fa2baf | Iustin Pop | (at your option) any later version. |
16 | e2fa2baf | Iustin Pop | |
17 | e2fa2baf | Iustin Pop | This program is distributed in the hope that it will be useful, but |
18 | e2fa2baf | Iustin Pop | WITHOUT ANY WARRANTY; without even the implied warranty of |
19 | e2fa2baf | Iustin Pop | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
20 | e2fa2baf | Iustin Pop | General Public License for more details. |
21 | e2fa2baf | Iustin Pop | |
22 | e2fa2baf | Iustin Pop | You should have received a copy of the GNU General Public License |
23 | e2fa2baf | Iustin Pop | along with this program; if not, write to the Free Software |
24 | e2fa2baf | Iustin Pop | Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA |
25 | e2fa2baf | Iustin Pop | 02110-1301, USA. |
26 | e2fa2baf | Iustin Pop | |
27 | e2fa2baf | Iustin Pop | -} |
28 | e2fa2baf | Iustin Pop | |
29 | 669d7e3d | Iustin Pop | module Ganeti.HTools.Cluster |
30 | f23f21c3 | Iustin Pop | ( |
31 | f23f21c3 | Iustin Pop | -- * Types |
32 | f23f21c3 | Iustin Pop | AllocSolution(..) |
33 | f23f21c3 | Iustin Pop | , EvacSolution(..) |
34 | f23f21c3 | Iustin Pop | , Table(..) |
35 | f23f21c3 | Iustin Pop | , CStats(..) |
36 | f23f21c3 | Iustin Pop | , AllocResult |
37 | f23f21c3 | Iustin Pop | , AllocMethod |
38 | f23f21c3 | Iustin Pop | -- * Generic functions |
39 | f23f21c3 | Iustin Pop | , totalResources |
40 | f23f21c3 | Iustin Pop | , computeAllocationDelta |
41 | f23f21c3 | Iustin Pop | -- * First phase functions |
42 | f23f21c3 | Iustin Pop | , computeBadItems |
43 | f23f21c3 | Iustin Pop | -- * Second phase functions |
44 | f23f21c3 | Iustin Pop | , printSolutionLine |
45 | f23f21c3 | Iustin Pop | , formatCmds |
46 | f23f21c3 | Iustin Pop | , involvedNodes |
47 | f23f21c3 | Iustin Pop | , splitJobs |
48 | f23f21c3 | Iustin Pop | -- * Display functions |
49 | f23f21c3 | Iustin Pop | , printNodes |
50 | f23f21c3 | Iustin Pop | , printInsts |
51 | f23f21c3 | Iustin Pop | -- * Balacing functions |
52 | f23f21c3 | Iustin Pop | , checkMove |
53 | f23f21c3 | Iustin Pop | , doNextBalance |
54 | f23f21c3 | Iustin Pop | , tryBalance |
55 | f23f21c3 | Iustin Pop | , compCV |
56 | f23f21c3 | Iustin Pop | , compCVNodes |
57 | f23f21c3 | Iustin Pop | , compDetailedCV |
58 | f23f21c3 | Iustin Pop | , printStats |
59 | f23f21c3 | Iustin Pop | , iMoveToJob |
60 | f23f21c3 | Iustin Pop | -- * IAllocator functions |
61 | f23f21c3 | Iustin Pop | , genAllocNodes |
62 | f23f21c3 | Iustin Pop | , tryAlloc |
63 | f23f21c3 | Iustin Pop | , tryMGAlloc |
64 | f23f21c3 | Iustin Pop | , tryNodeEvac |
65 | f23f21c3 | Iustin Pop | , tryChangeGroup |
66 | f23f21c3 | Iustin Pop | , collapseFailures |
67 | f23f21c3 | Iustin Pop | -- * Allocation functions |
68 | f23f21c3 | Iustin Pop | , iterateAlloc |
69 | f23f21c3 | Iustin Pop | , tieredAlloc |
70 | f23f21c3 | Iustin Pop | -- * Node group functions |
71 | f23f21c3 | Iustin Pop | , instanceGroup |
72 | f23f21c3 | Iustin Pop | , findSplitInstances |
73 | f23f21c3 | Iustin Pop | , splitCluster |
74 | f23f21c3 | Iustin Pop | ) where |
75 | e4f08c46 | Iustin Pop | |
76 | 63a78055 | Iustin Pop | import qualified Data.IntSet as IntSet |
77 | e4f08c46 | Iustin Pop | import Data.List |
78 | 129734d3 | Iustin Pop | import Data.Maybe (fromJust, isNothing) |
79 | 5182e970 | Iustin Pop | import Data.Ord (comparing) |
80 | e4f08c46 | Iustin Pop | import Text.Printf (printf) |
81 | e4f08c46 | Iustin Pop | |
82 | 669d7e3d | Iustin Pop | import qualified Ganeti.HTools.Container as Container |
83 | 669d7e3d | Iustin Pop | import qualified Ganeti.HTools.Instance as Instance |
84 | 669d7e3d | Iustin Pop | import qualified Ganeti.HTools.Node as Node |
85 | aec636b9 | Iustin Pop | import qualified Ganeti.HTools.Group as Group |
86 | e4c5beaf | Iustin Pop | import Ganeti.HTools.Types |
87 | 669d7e3d | Iustin Pop | import Ganeti.HTools.Utils |
88 | 1adec4be | Iustin Pop | import Ganeti.HTools.Compat |
89 | 6b20875c | Iustin Pop | import qualified Ganeti.OpCodes as OpCodes |
90 | e4f08c46 | Iustin Pop | |
91 | 9188aeef | Iustin Pop | -- * Types |
92 | 9188aeef | Iustin Pop | |
93 | 0c936d24 | Iustin Pop | -- | Allocation\/relocation solution. |
94 | 85d0ddc3 | Iustin Pop | data AllocSolution = AllocSolution |
95 | 129734d3 | Iustin Pop | { asFailures :: [FailMode] -- ^ Failure counts |
96 | 129734d3 | Iustin Pop | , asAllocs :: Int -- ^ Good allocation count |
97 | 129734d3 | Iustin Pop | , asSolution :: Maybe Node.AllocElement -- ^ The actual allocation result |
98 | 129734d3 | Iustin Pop | , asLog :: [String] -- ^ Informational messages |
99 | 85d0ddc3 | Iustin Pop | } |
100 | 85d0ddc3 | Iustin Pop | |
101 | 47eed3f4 | Iustin Pop | -- | Node evacuation/group change iallocator result type. This result |
102 | 47eed3f4 | Iustin Pop | -- type consists of actual opcodes (a restricted subset) that are |
103 | 47eed3f4 | Iustin Pop | -- transmitted back to Ganeti. |
104 | 47eed3f4 | Iustin Pop | data EvacSolution = EvacSolution |
105 | f23f21c3 | Iustin Pop | { esMoved :: [(Idx, Gdx, [Ndx])] -- ^ Instances moved successfully |
106 | f23f21c3 | Iustin Pop | , esFailed :: [(Idx, String)] -- ^ Instances which were not |
107 | f23f21c3 | Iustin Pop | -- relocated |
108 | f23f21c3 | Iustin Pop | , esOpCodes :: [[OpCodes.OpCode]] -- ^ List of jobs |
109 | 6a855aaa | Iustin Pop | } deriving (Show) |
110 | 47eed3f4 | Iustin Pop | |
111 | 40ee14bc | Iustin Pop | -- | Allocation results, as used in 'iterateAlloc' and 'tieredAlloc'. |
112 | 40ee14bc | Iustin Pop | type AllocResult = (FailStats, Node.List, Instance.List, |
113 | 40ee14bc | Iustin Pop | [Instance.Instance], [CStats]) |
114 | 40ee14bc | Iustin Pop | |
115 | 6cb1649f | Iustin Pop | -- | A type denoting the valid allocation mode/pairs. |
116 | 525bfb36 | Iustin Pop | -- |
117 | b0631f10 | Iustin Pop | -- For a one-node allocation, this will be a @Left ['Ndx']@, whereas |
118 | b0631f10 | Iustin Pop | -- for a two-node allocation, this will be a @Right [('Ndx', |
119 | b0631f10 | Iustin Pop | -- ['Ndx'])]@. In the latter case, the list is basically an |
120 | b0631f10 | Iustin Pop | -- association list, grouped by primary node and holding the potential |
121 | b0631f10 | Iustin Pop | -- secondary nodes in the sub-list. |
122 | b0631f10 | Iustin Pop | type AllocNodes = Either [Ndx] [(Ndx, [Ndx])] |
123 | 6cb1649f | Iustin Pop | |
124 | 525bfb36 | Iustin Pop | -- | The empty solution we start with when computing allocations. |
125 | 97936d51 | Iustin Pop | emptyAllocSolution :: AllocSolution |
126 | 97936d51 | Iustin Pop | emptyAllocSolution = AllocSolution { asFailures = [], asAllocs = 0 |
127 | 129734d3 | Iustin Pop | , asSolution = Nothing, asLog = [] } |
128 | 78694255 | Iustin Pop | |
129 | 47eed3f4 | Iustin Pop | -- | The empty evac solution. |
130 | 47eed3f4 | Iustin Pop | emptyEvacSolution :: EvacSolution |
131 | 47eed3f4 | Iustin Pop | emptyEvacSolution = EvacSolution { esMoved = [] |
132 | 47eed3f4 | Iustin Pop | , esFailed = [] |
133 | 47eed3f4 | Iustin Pop | , esOpCodes = [] |
134 | 47eed3f4 | Iustin Pop | } |
135 | 47eed3f4 | Iustin Pop | |
136 | 525bfb36 | Iustin Pop | -- | The complete state for the balancing solution. |
137 | 262a08a2 | Iustin Pop | data Table = Table Node.List Instance.List Score [Placement] |
138 | 6bc39970 | Iustin Pop | deriving (Show, Read) |
139 | e4f08c46 | Iustin Pop | |
140 | 179c0828 | Iustin Pop | -- | Cluster statistics data type. |
141 | 33e17565 | Iustin Pop | data CStats = CStats |
142 | 33e17565 | Iustin Pop | { csFmem :: Integer -- ^ Cluster free mem |
143 | 33e17565 | Iustin Pop | , csFdsk :: Integer -- ^ Cluster free disk |
144 | 33e17565 | Iustin Pop | , csAmem :: Integer -- ^ Cluster allocatable mem |
145 | 33e17565 | Iustin Pop | , csAdsk :: Integer -- ^ Cluster allocatable disk |
146 | 33e17565 | Iustin Pop | , csAcpu :: Integer -- ^ Cluster allocatable cpus |
147 | 33e17565 | Iustin Pop | , csMmem :: Integer -- ^ Max node allocatable mem |
148 | 33e17565 | Iustin Pop | , csMdsk :: Integer -- ^ Max node allocatable disk |
149 | 33e17565 | Iustin Pop | , csMcpu :: Integer -- ^ Max node allocatable cpu |
150 | 33e17565 | Iustin Pop | , csImem :: Integer -- ^ Instance used mem |
151 | 33e17565 | Iustin Pop | , csIdsk :: Integer -- ^ Instance used disk |
152 | 33e17565 | Iustin Pop | , csIcpu :: Integer -- ^ Instance used cpu |
153 | 33e17565 | Iustin Pop | , csTmem :: Double -- ^ Cluster total mem |
154 | 33e17565 | Iustin Pop | , csTdsk :: Double -- ^ Cluster total disk |
155 | 33e17565 | Iustin Pop | , csTcpu :: Double -- ^ Cluster total cpus |
156 | 90c2f1e8 | Iustin Pop | , csVcpu :: Integer -- ^ Cluster total virtual cpus |
157 | 90c2f1e8 | Iustin Pop | , csNcpu :: Double -- ^ Equivalent to 'csIcpu' but in terms of |
158 | 90c2f1e8 | Iustin Pop | -- physical CPUs, i.e. normalised used phys CPUs |
159 | 33e17565 | Iustin Pop | , csXmem :: Integer -- ^ Unnacounted for mem |
160 | 33e17565 | Iustin Pop | , csNmem :: Integer -- ^ Node own memory |
161 | 33e17565 | Iustin Pop | , csScore :: Score -- ^ The cluster score |
162 | 33e17565 | Iustin Pop | , csNinst :: Int -- ^ The total number of instances |
163 | 33e17565 | Iustin Pop | } deriving (Show, Read) |
164 | 1a7eff0e | Iustin Pop | |
165 | 7eda951b | Iustin Pop | -- | A simple type for allocation functions. |
166 | 7eda951b | Iustin Pop | type AllocMethod = Node.List -- ^ Node list |
167 | 7eda951b | Iustin Pop | -> Instance.List -- ^ Instance list |
168 | 7eda951b | Iustin Pop | -> Maybe Int -- ^ Optional allocation limit |
169 | 7eda951b | Iustin Pop | -> Instance.Instance -- ^ Instance spec for allocation |
170 | 7eda951b | Iustin Pop | -> AllocNodes -- ^ Which nodes we should allocate on |
171 | 7eda951b | Iustin Pop | -> [Instance.Instance] -- ^ Allocated instances |
172 | 7eda951b | Iustin Pop | -> [CStats] -- ^ Running cluster stats |
173 | 7eda951b | Iustin Pop | -> Result AllocResult -- ^ Allocation result |
174 | 7eda951b | Iustin Pop | |
175 | bebe7a73 | Iustin Pop | -- | A simple type for the running solution of evacuations. |
176 | bebe7a73 | Iustin Pop | type EvacInnerState = |
177 | bebe7a73 | Iustin Pop | Either String (Node.List, Instance.Instance, Score, Ndx) |
178 | bebe7a73 | Iustin Pop | |
179 | 9188aeef | Iustin Pop | -- * Utility functions |
180 | 9188aeef | Iustin Pop | |
181 | e4f08c46 | Iustin Pop | -- | Verifies the N+1 status and return the affected nodes. |
182 | e4f08c46 | Iustin Pop | verifyN1 :: [Node.Node] -> [Node.Node] |
183 | 9f6dcdea | Iustin Pop | verifyN1 = filter Node.failN1 |
184 | e4f08c46 | Iustin Pop | |
185 | 9188aeef | Iustin Pop | {-| Computes the pair of bad nodes and instances. |
186 | 9188aeef | Iustin Pop | |
187 | 9188aeef | Iustin Pop | The bad node list is computed via a simple 'verifyN1' check, and the |
188 | 9188aeef | Iustin Pop | bad instance list is the list of primary and secondary instances of |
189 | 9188aeef | Iustin Pop | those nodes. |
190 | 9188aeef | Iustin Pop | |
191 | 9188aeef | Iustin Pop | -} |
192 | 9188aeef | Iustin Pop | computeBadItems :: Node.List -> Instance.List -> |
193 | 9188aeef | Iustin Pop | ([Node.Node], [Instance.Instance]) |
194 | 9188aeef | Iustin Pop | computeBadItems nl il = |
195 | dbba5246 | Iustin Pop | let bad_nodes = verifyN1 $ getOnline nl |
196 | 5182e970 | Iustin Pop | bad_instances = map (`Container.find` il) . |
197 | 9f6dcdea | Iustin Pop | sort . nub $ |
198 | 2060348b | Iustin Pop | concatMap (\ n -> Node.sList n ++ Node.pList n) bad_nodes |
199 | 9188aeef | Iustin Pop | in |
200 | 9188aeef | Iustin Pop | (bad_nodes, bad_instances) |
201 | 9188aeef | Iustin Pop | |
202 | 255f55a9 | Iustin Pop | -- | Extracts the node pairs for an instance. This can fail if the |
203 | 255f55a9 | Iustin Pop | -- instance is single-homed. FIXME: this needs to be improved, |
204 | 255f55a9 | Iustin Pop | -- together with the general enhancement for handling non-DRBD moves. |
205 | 255f55a9 | Iustin Pop | instanceNodes :: Node.List -> Instance.Instance -> |
206 | 255f55a9 | Iustin Pop | (Ndx, Ndx, Node.Node, Node.Node) |
207 | 255f55a9 | Iustin Pop | instanceNodes nl inst = |
208 | 255f55a9 | Iustin Pop | let old_pdx = Instance.pNode inst |
209 | 255f55a9 | Iustin Pop | old_sdx = Instance.sNode inst |
210 | 255f55a9 | Iustin Pop | old_p = Container.find old_pdx nl |
211 | 255f55a9 | Iustin Pop | old_s = Container.find old_sdx nl |
212 | 255f55a9 | Iustin Pop | in (old_pdx, old_sdx, old_p, old_s) |
213 | 255f55a9 | Iustin Pop | |
214 | 525bfb36 | Iustin Pop | -- | Zero-initializer for the CStats type. |
215 | 1a7eff0e | Iustin Pop | emptyCStats :: CStats |
216 | 90c2f1e8 | Iustin Pop | emptyCStats = CStats 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 |
217 | 1a7eff0e | Iustin Pop | |
218 | 525bfb36 | Iustin Pop | -- | Update stats with data from a new node. |
219 | 1a7eff0e | Iustin Pop | updateCStats :: CStats -> Node.Node -> CStats |
220 | 1a7eff0e | Iustin Pop | updateCStats cs node = |
221 | f23f21c3 | Iustin Pop | let CStats { csFmem = x_fmem, csFdsk = x_fdsk, |
222 | f23f21c3 | Iustin Pop | csAmem = x_amem, csAcpu = x_acpu, csAdsk = x_adsk, |
223 | f23f21c3 | Iustin Pop | csMmem = x_mmem, csMdsk = x_mdsk, csMcpu = x_mcpu, |
224 | f23f21c3 | Iustin Pop | csImem = x_imem, csIdsk = x_idsk, csIcpu = x_icpu, |
225 | f23f21c3 | Iustin Pop | csTmem = x_tmem, csTdsk = x_tdsk, csTcpu = x_tcpu, |
226 | 90c2f1e8 | Iustin Pop | csVcpu = x_vcpu, csNcpu = x_ncpu, |
227 | f23f21c3 | Iustin Pop | csXmem = x_xmem, csNmem = x_nmem, csNinst = x_ninst |
228 | f23f21c3 | Iustin Pop | } |
229 | f23f21c3 | Iustin Pop | = cs |
230 | f23f21c3 | Iustin Pop | inc_amem = Node.fMem node - Node.rMem node |
231 | f23f21c3 | Iustin Pop | inc_amem' = if inc_amem > 0 then inc_amem else 0 |
232 | f23f21c3 | Iustin Pop | inc_adsk = Node.availDisk node |
233 | f23f21c3 | Iustin Pop | inc_imem = truncate (Node.tMem node) - Node.nMem node |
234 | f23f21c3 | Iustin Pop | - Node.xMem node - Node.fMem node |
235 | f23f21c3 | Iustin Pop | inc_icpu = Node.uCpu node |
236 | f23f21c3 | Iustin Pop | inc_idsk = truncate (Node.tDsk node) - Node.fDsk node |
237 | f23f21c3 | Iustin Pop | inc_vcpu = Node.hiCpu node |
238 | f23f21c3 | Iustin Pop | inc_acpu = Node.availCpu node |
239 | 90c2f1e8 | Iustin Pop | inc_ncpu = fromIntegral (Node.uCpu node) / |
240 | 90c2f1e8 | Iustin Pop | iPolicyVcpuRatio (Node.iPolicy node) |
241 | f23f21c3 | Iustin Pop | in cs { csFmem = x_fmem + fromIntegral (Node.fMem node) |
242 | f23f21c3 | Iustin Pop | , csFdsk = x_fdsk + fromIntegral (Node.fDsk node) |
243 | f23f21c3 | Iustin Pop | , csAmem = x_amem + fromIntegral inc_amem' |
244 | f23f21c3 | Iustin Pop | , csAdsk = x_adsk + fromIntegral inc_adsk |
245 | f23f21c3 | Iustin Pop | , csAcpu = x_acpu + fromIntegral inc_acpu |
246 | f23f21c3 | Iustin Pop | , csMmem = max x_mmem (fromIntegral inc_amem') |
247 | f23f21c3 | Iustin Pop | , csMdsk = max x_mdsk (fromIntegral inc_adsk) |
248 | f23f21c3 | Iustin Pop | , csMcpu = max x_mcpu (fromIntegral inc_acpu) |
249 | f23f21c3 | Iustin Pop | , csImem = x_imem + fromIntegral inc_imem |
250 | f23f21c3 | Iustin Pop | , csIdsk = x_idsk + fromIntegral inc_idsk |
251 | f23f21c3 | Iustin Pop | , csIcpu = x_icpu + fromIntegral inc_icpu |
252 | f23f21c3 | Iustin Pop | , csTmem = x_tmem + Node.tMem node |
253 | f23f21c3 | Iustin Pop | , csTdsk = x_tdsk + Node.tDsk node |
254 | f23f21c3 | Iustin Pop | , csTcpu = x_tcpu + Node.tCpu node |
255 | f23f21c3 | Iustin Pop | , csVcpu = x_vcpu + fromIntegral inc_vcpu |
256 | 90c2f1e8 | Iustin Pop | , csNcpu = x_ncpu + inc_ncpu |
257 | f23f21c3 | Iustin Pop | , csXmem = x_xmem + fromIntegral (Node.xMem node) |
258 | f23f21c3 | Iustin Pop | , csNmem = x_nmem + fromIntegral (Node.nMem node) |
259 | f23f21c3 | Iustin Pop | , csNinst = x_ninst + length (Node.pList node) |
260 | f23f21c3 | Iustin Pop | } |
261 | 1a7eff0e | Iustin Pop | |
262 | 9188aeef | Iustin Pop | -- | Compute the total free disk and memory in the cluster. |
263 | 1a7eff0e | Iustin Pop | totalResources :: Node.List -> CStats |
264 | de4ac2c2 | Iustin Pop | totalResources nl = |
265 | f23f21c3 | Iustin Pop | let cs = foldl' updateCStats emptyCStats . Container.elems $ nl |
266 | f23f21c3 | Iustin Pop | in cs { csScore = compCV nl } |
267 | 9188aeef | Iustin Pop | |
268 | 9b8fac3d | Iustin Pop | -- | Compute the delta between two cluster state. |
269 | 9b8fac3d | Iustin Pop | -- |
270 | 9b8fac3d | Iustin Pop | -- This is used when doing allocations, to understand better the |
271 | e2436511 | Iustin Pop | -- available cluster resources. The return value is a triple of the |
272 | e2436511 | Iustin Pop | -- current used values, the delta that was still allocated, and what |
273 | e2436511 | Iustin Pop | -- was left unallocated. |
274 | 9b8fac3d | Iustin Pop | computeAllocationDelta :: CStats -> CStats -> AllocStats |
275 | 9b8fac3d | Iustin Pop | computeAllocationDelta cini cfin = |
276 | 80d7d8a1 | Iustin Pop | let CStats {csImem = i_imem, csIdsk = i_idsk, csIcpu = i_icpu, |
277 | 80d7d8a1 | Iustin Pop | csNcpu = i_ncpu } = cini |
278 | f23f21c3 | Iustin Pop | CStats {csImem = f_imem, csIdsk = f_idsk, csIcpu = f_icpu, |
279 | 80d7d8a1 | Iustin Pop | csTmem = t_mem, csTdsk = t_dsk, csVcpu = f_vcpu, |
280 | 80d7d8a1 | Iustin Pop | csNcpu = f_ncpu, csTcpu = f_tcpu } = cfin |
281 | 80d7d8a1 | Iustin Pop | rini = AllocInfo { allocInfoVCpus = fromIntegral i_icpu |
282 | 80d7d8a1 | Iustin Pop | , allocInfoNCpus = i_ncpu |
283 | 80d7d8a1 | Iustin Pop | , allocInfoMem = fromIntegral i_imem |
284 | 80d7d8a1 | Iustin Pop | , allocInfoDisk = fromIntegral i_idsk |
285 | 80d7d8a1 | Iustin Pop | } |
286 | 80d7d8a1 | Iustin Pop | rfin = AllocInfo { allocInfoVCpus = fromIntegral (f_icpu - i_icpu) |
287 | 80d7d8a1 | Iustin Pop | , allocInfoNCpus = f_ncpu - i_ncpu |
288 | 80d7d8a1 | Iustin Pop | , allocInfoMem = fromIntegral (f_imem - i_imem) |
289 | 80d7d8a1 | Iustin Pop | , allocInfoDisk = fromIntegral (f_idsk - i_idsk) |
290 | 80d7d8a1 | Iustin Pop | } |
291 | 80d7d8a1 | Iustin Pop | runa = AllocInfo { allocInfoVCpus = fromIntegral (f_vcpu - f_icpu) |
292 | 80d7d8a1 | Iustin Pop | , allocInfoNCpus = f_tcpu - f_ncpu |
293 | 80d7d8a1 | Iustin Pop | , allocInfoMem = truncate t_mem - fromIntegral f_imem |
294 | 80d7d8a1 | Iustin Pop | , allocInfoDisk = truncate t_dsk - fromIntegral f_idsk |
295 | 80d7d8a1 | Iustin Pop | } |
296 | f23f21c3 | Iustin Pop | in (rini, rfin, runa) |
297 | 9b8fac3d | Iustin Pop | |
298 | 525bfb36 | Iustin Pop | -- | The names and weights of the individual elements in the CV list. |
299 | 8a3b30ca | Iustin Pop | detailedCVInfo :: [(Double, String)] |
300 | 8a3b30ca | Iustin Pop | detailedCVInfo = [ (1, "free_mem_cv") |
301 | 8a3b30ca | Iustin Pop | , (1, "free_disk_cv") |
302 | 8a3b30ca | Iustin Pop | , (1, "n1_cnt") |
303 | 8a3b30ca | Iustin Pop | , (1, "reserved_mem_cv") |
304 | 8a3b30ca | Iustin Pop | , (4, "offline_all_cnt") |
305 | 8a3b30ca | Iustin Pop | , (16, "offline_pri_cnt") |
306 | 8a3b30ca | Iustin Pop | , (1, "vcpu_ratio_cv") |
307 | 8a3b30ca | Iustin Pop | , (1, "cpu_load_cv") |
308 | 8a3b30ca | Iustin Pop | , (1, "mem_load_cv") |
309 | 8a3b30ca | Iustin Pop | , (1, "disk_load_cv") |
310 | 8a3b30ca | Iustin Pop | , (1, "net_load_cv") |
311 | 306cccd5 | Iustin Pop | , (2, "pri_tags_score") |
312 | 084565ac | Iustin Pop | , (1, "spindles_cv") |
313 | 8a3b30ca | Iustin Pop | ] |
314 | 8a3b30ca | Iustin Pop | |
315 | 179c0828 | Iustin Pop | -- | Holds the weights used by 'compCVNodes' for each metric. |
316 | 8a3b30ca | Iustin Pop | detailedCVWeights :: [Double] |
317 | 8a3b30ca | Iustin Pop | detailedCVWeights = map fst detailedCVInfo |
318 | fca250e9 | Iustin Pop | |
319 | 9188aeef | Iustin Pop | -- | Compute the mem and disk covariance. |
320 | 9bb5721c | Iustin Pop | compDetailedCV :: [Node.Node] -> [Double] |
321 | 9bb5721c | Iustin Pop | compDetailedCV all_nodes = |
322 | f23f21c3 | Iustin Pop | let (offline, nodes) = partition Node.offline all_nodes |
323 | f23f21c3 | Iustin Pop | mem_l = map Node.pMem nodes |
324 | f23f21c3 | Iustin Pop | dsk_l = map Node.pDsk nodes |
325 | f23f21c3 | Iustin Pop | -- metric: memory covariance |
326 | f23f21c3 | Iustin Pop | mem_cv = stdDev mem_l |
327 | f23f21c3 | Iustin Pop | -- metric: disk covariance |
328 | f23f21c3 | Iustin Pop | dsk_cv = stdDev dsk_l |
329 | f23f21c3 | Iustin Pop | -- metric: count of instances living on N1 failing nodes |
330 | f23f21c3 | Iustin Pop | n1_score = fromIntegral . sum . map (\n -> length (Node.sList n) + |
331 | f23f21c3 | Iustin Pop | length (Node.pList n)) . |
332 | f23f21c3 | Iustin Pop | filter Node.failN1 $ nodes :: Double |
333 | f23f21c3 | Iustin Pop | res_l = map Node.pRem nodes |
334 | f23f21c3 | Iustin Pop | -- metric: reserved memory covariance |
335 | f23f21c3 | Iustin Pop | res_cv = stdDev res_l |
336 | f23f21c3 | Iustin Pop | -- offline instances metrics |
337 | f23f21c3 | Iustin Pop | offline_ipri = sum . map (length . Node.pList) $ offline |
338 | f23f21c3 | Iustin Pop | offline_isec = sum . map (length . Node.sList) $ offline |
339 | f23f21c3 | Iustin Pop | -- metric: count of instances on offline nodes |
340 | f23f21c3 | Iustin Pop | off_score = fromIntegral (offline_ipri + offline_isec)::Double |
341 | f23f21c3 | Iustin Pop | -- metric: count of primary instances on offline nodes (this |
342 | f23f21c3 | Iustin Pop | -- helps with evacuation/failover of primary instances on |
343 | f23f21c3 | Iustin Pop | -- 2-node clusters with one node offline) |
344 | f23f21c3 | Iustin Pop | off_pri_score = fromIntegral offline_ipri::Double |
345 | f23f21c3 | Iustin Pop | cpu_l = map Node.pCpu nodes |
346 | f23f21c3 | Iustin Pop | -- metric: covariance of vcpu/pcpu ratio |
347 | f23f21c3 | Iustin Pop | cpu_cv = stdDev cpu_l |
348 | f23f21c3 | Iustin Pop | -- metrics: covariance of cpu, memory, disk and network load |
349 | f23f21c3 | Iustin Pop | (c_load, m_load, d_load, n_load) = |
350 | f23f21c3 | Iustin Pop | unzip4 $ map (\n -> |
351 | f23f21c3 | Iustin Pop | let DynUtil c1 m1 d1 n1 = Node.utilLoad n |
352 | f23f21c3 | Iustin Pop | DynUtil c2 m2 d2 n2 = Node.utilPool n |
353 | f23f21c3 | Iustin Pop | in (c1/c2, m1/m2, d1/d2, n1/n2)) nodes |
354 | f23f21c3 | Iustin Pop | -- metric: conflicting instance count |
355 | f23f21c3 | Iustin Pop | pri_tags_inst = sum $ map Node.conflictingPrimaries nodes |
356 | f23f21c3 | Iustin Pop | pri_tags_score = fromIntegral pri_tags_inst::Double |
357 | 084565ac | Iustin Pop | -- metric: spindles % |
358 | 084565ac | Iustin Pop | spindles_cv = map (\n -> Node.instSpindles n / Node.hiSpindles n) nodes |
359 | f23f21c3 | Iustin Pop | in [ mem_cv, dsk_cv, n1_score, res_cv, off_score, off_pri_score, cpu_cv |
360 | f23f21c3 | Iustin Pop | , stdDev c_load, stdDev m_load , stdDev d_load, stdDev n_load |
361 | 084565ac | Iustin Pop | , pri_tags_score, stdDev spindles_cv ] |
362 | 9188aeef | Iustin Pop | |
363 | 9188aeef | Iustin Pop | -- | Compute the /total/ variance. |
364 | 9bb5721c | Iustin Pop | compCVNodes :: [Node.Node] -> Double |
365 | 9bb5721c | Iustin Pop | compCVNodes = sum . zipWith (*) detailedCVWeights . compDetailedCV |
366 | 9bb5721c | Iustin Pop | |
367 | 9bb5721c | Iustin Pop | -- | Wrapper over 'compCVNodes' for callers that have a 'Node.List'. |
368 | 9188aeef | Iustin Pop | compCV :: Node.List -> Double |
369 | 9bb5721c | Iustin Pop | compCV = compCVNodes . Container.elems |
370 | 9bb5721c | Iustin Pop | |
371 | 525bfb36 | Iustin Pop | -- | Compute online nodes from a 'Node.List'. |
372 | dbba5246 | Iustin Pop | getOnline :: Node.List -> [Node.Node] |
373 | dbba5246 | Iustin Pop | getOnline = filter (not . Node.offline) . Container.elems |
374 | dbba5246 | Iustin Pop | |
375 | 525bfb36 | Iustin Pop | -- * Balancing functions |
376 | 9188aeef | Iustin Pop | |
377 | 9188aeef | Iustin Pop | -- | Compute best table. Note that the ordering of the arguments is important. |
378 | 9188aeef | Iustin Pop | compareTables :: Table -> Table -> Table |
379 | 9188aeef | Iustin Pop | compareTables a@(Table _ _ a_cv _) b@(Table _ _ b_cv _ ) = |
380 | f23f21c3 | Iustin Pop | if a_cv > b_cv then b else a |
381 | 9188aeef | Iustin Pop | |
382 | 9188aeef | Iustin Pop | -- | Applies an instance move to a given node list and instance. |
383 | 262a08a2 | Iustin Pop | applyMove :: Node.List -> Instance.Instance |
384 | 8880d889 | Iustin Pop | -> IMove -> OpResult (Node.List, Instance.Instance, Ndx, Ndx) |
385 | 00b51a14 | Iustin Pop | -- Failover (f) |
386 | e4f08c46 | Iustin Pop | applyMove nl inst Failover = |
387 | 255f55a9 | Iustin Pop | let (old_pdx, old_sdx, old_p, old_s) = instanceNodes nl inst |
388 | f23f21c3 | Iustin Pop | int_p = Node.removePri old_p inst |
389 | f23f21c3 | Iustin Pop | int_s = Node.removeSec old_s inst |
390 | f23f21c3 | Iustin Pop | new_nl = do -- Maybe monad |
391 | 3603605a | Iustin Pop | new_p <- Node.addPriEx (Node.offline old_p) int_s inst |
392 | f23f21c3 | Iustin Pop | new_s <- Node.addSec int_p inst old_sdx |
393 | f23f21c3 | Iustin Pop | let new_inst = Instance.setBoth inst old_sdx old_pdx |
394 | f23f21c3 | Iustin Pop | return (Container.addTwo old_pdx new_s old_sdx new_p nl, |
395 | f23f21c3 | Iustin Pop | new_inst, old_sdx, old_pdx) |
396 | f23f21c3 | Iustin Pop | in new_nl |
397 | e4f08c46 | Iustin Pop | |
398 | 00b51a14 | Iustin Pop | -- Replace the primary (f:, r:np, f) |
399 | e4f08c46 | Iustin Pop | applyMove nl inst (ReplacePrimary new_pdx) = |
400 | 255f55a9 | Iustin Pop | let (old_pdx, old_sdx, old_p, old_s) = instanceNodes nl inst |
401 | f23f21c3 | Iustin Pop | tgt_n = Container.find new_pdx nl |
402 | f23f21c3 | Iustin Pop | int_p = Node.removePri old_p inst |
403 | f23f21c3 | Iustin Pop | int_s = Node.removeSec old_s inst |
404 | f23f21c3 | Iustin Pop | force_p = Node.offline old_p |
405 | f23f21c3 | Iustin Pop | new_nl = do -- Maybe monad |
406 | f23f21c3 | Iustin Pop | -- check that the current secondary can host the instance |
407 | f23f21c3 | Iustin Pop | -- during the migration |
408 | f23f21c3 | Iustin Pop | tmp_s <- Node.addPriEx force_p int_s inst |
409 | f23f21c3 | Iustin Pop | let tmp_s' = Node.removePri tmp_s inst |
410 | f23f21c3 | Iustin Pop | new_p <- Node.addPriEx force_p tgt_n inst |
411 | f23f21c3 | Iustin Pop | new_s <- Node.addSecEx force_p tmp_s' inst new_pdx |
412 | f23f21c3 | Iustin Pop | let new_inst = Instance.setPri inst new_pdx |
413 | f23f21c3 | Iustin Pop | return (Container.add new_pdx new_p $ |
414 | f23f21c3 | Iustin Pop | Container.addTwo old_pdx int_p old_sdx new_s nl, |
415 | f23f21c3 | Iustin Pop | new_inst, new_pdx, old_sdx) |
416 | f23f21c3 | Iustin Pop | in new_nl |
417 | e4f08c46 | Iustin Pop | |
418 | 00b51a14 | Iustin Pop | -- Replace the secondary (r:ns) |
419 | e4f08c46 | Iustin Pop | applyMove nl inst (ReplaceSecondary new_sdx) = |
420 | f23f21c3 | Iustin Pop | let old_pdx = Instance.pNode inst |
421 | f23f21c3 | Iustin Pop | old_sdx = Instance.sNode inst |
422 | f23f21c3 | Iustin Pop | old_s = Container.find old_sdx nl |
423 | f23f21c3 | Iustin Pop | tgt_n = Container.find new_sdx nl |
424 | f23f21c3 | Iustin Pop | int_s = Node.removeSec old_s inst |
425 | f23f21c3 | Iustin Pop | force_s = Node.offline old_s |
426 | f23f21c3 | Iustin Pop | new_inst = Instance.setSec inst new_sdx |
427 | f23f21c3 | Iustin Pop | new_nl = Node.addSecEx force_s tgt_n inst old_pdx >>= |
428 | f23f21c3 | Iustin Pop | \new_s -> return (Container.addTwo new_sdx |
429 | f23f21c3 | Iustin Pop | new_s old_sdx int_s nl, |
430 | f23f21c3 | Iustin Pop | new_inst, old_pdx, new_sdx) |
431 | f23f21c3 | Iustin Pop | in new_nl |
432 | e4f08c46 | Iustin Pop | |
433 | 00b51a14 | Iustin Pop | -- Replace the secondary and failover (r:np, f) |
434 | 79ac6b6f | Iustin Pop | applyMove nl inst (ReplaceAndFailover new_pdx) = |
435 | 255f55a9 | Iustin Pop | let (old_pdx, old_sdx, old_p, old_s) = instanceNodes nl inst |
436 | f23f21c3 | Iustin Pop | tgt_n = Container.find new_pdx nl |
437 | f23f21c3 | Iustin Pop | int_p = Node.removePri old_p inst |
438 | f23f21c3 | Iustin Pop | int_s = Node.removeSec old_s inst |
439 | f23f21c3 | Iustin Pop | force_s = Node.offline old_s |
440 | f23f21c3 | Iustin Pop | new_nl = do -- Maybe monad |
441 | f23f21c3 | Iustin Pop | new_p <- Node.addPri tgt_n inst |
442 | f23f21c3 | Iustin Pop | new_s <- Node.addSecEx force_s int_p inst new_pdx |
443 | f23f21c3 | Iustin Pop | let new_inst = Instance.setBoth inst new_pdx old_pdx |
444 | f23f21c3 | Iustin Pop | return (Container.add new_pdx new_p $ |
445 | f23f21c3 | Iustin Pop | Container.addTwo old_pdx new_s old_sdx int_s nl, |
446 | f23f21c3 | Iustin Pop | new_inst, new_pdx, old_pdx) |
447 | f23f21c3 | Iustin Pop | in new_nl |
448 | 79ac6b6f | Iustin Pop | |
449 | 19493d33 | Iustin Pop | -- Failver and replace the secondary (f, r:ns) |
450 | 19493d33 | Iustin Pop | applyMove nl inst (FailoverAndReplace new_sdx) = |
451 | 255f55a9 | Iustin Pop | let (old_pdx, old_sdx, old_p, old_s) = instanceNodes nl inst |
452 | f23f21c3 | Iustin Pop | tgt_n = Container.find new_sdx nl |
453 | f23f21c3 | Iustin Pop | int_p = Node.removePri old_p inst |
454 | f23f21c3 | Iustin Pop | int_s = Node.removeSec old_s inst |
455 | f23f21c3 | Iustin Pop | force_p = Node.offline old_p |
456 | f23f21c3 | Iustin Pop | new_nl = do -- Maybe monad |
457 | f23f21c3 | Iustin Pop | new_p <- Node.addPriEx force_p int_s inst |
458 | f23f21c3 | Iustin Pop | new_s <- Node.addSecEx force_p tgt_n inst old_sdx |
459 | f23f21c3 | Iustin Pop | let new_inst = Instance.setBoth inst old_sdx new_sdx |
460 | f23f21c3 | Iustin Pop | return (Container.add new_sdx new_s $ |
461 | f23f21c3 | Iustin Pop | Container.addTwo old_sdx new_p old_pdx int_p nl, |
462 | f23f21c3 | Iustin Pop | new_inst, old_sdx, new_sdx) |
463 | f23f21c3 | Iustin Pop | in new_nl |
464 | 19493d33 | Iustin Pop | |
465 | 9188aeef | Iustin Pop | -- | Tries to allocate an instance on one given node. |
466 | 0d66ea67 | Iustin Pop | allocateOnSingle :: Node.List -> Instance.Instance -> Ndx |
467 | 1fe81531 | Iustin Pop | -> OpResult Node.AllocElement |
468 | 0d66ea67 | Iustin Pop | allocateOnSingle nl inst new_pdx = |
469 | f23f21c3 | Iustin Pop | let p = Container.find new_pdx nl |
470 | f23f21c3 | Iustin Pop | new_inst = Instance.setBoth inst new_pdx Node.noSecondary |
471 | aa5b2f07 | Iustin Pop | in do |
472 | aa5b2f07 | Iustin Pop | Instance.instMatchesPolicy inst (Node.iPolicy p) |
473 | aa5b2f07 | Iustin Pop | new_p <- Node.addPri p inst |
474 | f23f21c3 | Iustin Pop | let new_nl = Container.add new_pdx new_p nl |
475 | f23f21c3 | Iustin Pop | new_score = compCV nl |
476 | f23f21c3 | Iustin Pop | return (new_nl, new_inst, [new_p], new_score) |
477 | 5e15f460 | Iustin Pop | |
478 | 9188aeef | Iustin Pop | -- | Tries to allocate an instance on a given pair of nodes. |
479 | 0d66ea67 | Iustin Pop | allocateOnPair :: Node.List -> Instance.Instance -> Ndx -> Ndx |
480 | 1fe81531 | Iustin Pop | -> OpResult Node.AllocElement |
481 | 0d66ea67 | Iustin Pop | allocateOnPair nl inst new_pdx new_sdx = |
482 | f23f21c3 | Iustin Pop | let tgt_p = Container.find new_pdx nl |
483 | f23f21c3 | Iustin Pop | tgt_s = Container.find new_sdx nl |
484 | f23f21c3 | Iustin Pop | in do |
485 | aa5b2f07 | Iustin Pop | Instance.instMatchesPolicy inst (Node.iPolicy tgt_p) |
486 | f23f21c3 | Iustin Pop | new_p <- Node.addPri tgt_p inst |
487 | f23f21c3 | Iustin Pop | new_s <- Node.addSec tgt_s inst new_pdx |
488 | f23f21c3 | Iustin Pop | let new_inst = Instance.setBoth inst new_pdx new_sdx |
489 | f23f21c3 | Iustin Pop | new_nl = Container.addTwo new_pdx new_p new_sdx new_s nl |
490 | f23f21c3 | Iustin Pop | return (new_nl, new_inst, [new_p, new_s], compCV new_nl) |
491 | 4a340313 | Iustin Pop | |
492 | 9188aeef | Iustin Pop | -- | Tries to perform an instance move and returns the best table |
493 | 9188aeef | Iustin Pop | -- between the original one and the new one. |
494 | e4f08c46 | Iustin Pop | checkSingleStep :: Table -- ^ The original table |
495 | e4f08c46 | Iustin Pop | -> Instance.Instance -- ^ The instance to move |
496 | e4f08c46 | Iustin Pop | -> Table -- ^ The current best table |
497 | e4f08c46 | Iustin Pop | -> IMove -- ^ The move to apply |
498 | e4f08c46 | Iustin Pop | -> Table -- ^ The final best table |
499 | e4f08c46 | Iustin Pop | checkSingleStep ini_tbl target cur_tbl move = |
500 | f23f21c3 | Iustin Pop | let Table ini_nl ini_il _ ini_plc = ini_tbl |
501 | f23f21c3 | Iustin Pop | tmp_resu = applyMove ini_nl target move |
502 | f23f21c3 | Iustin Pop | in case tmp_resu of |
503 | f23f21c3 | Iustin Pop | OpFail _ -> cur_tbl |
504 | f23f21c3 | Iustin Pop | OpGood (upd_nl, new_inst, pri_idx, sec_idx) -> |
505 | f23f21c3 | Iustin Pop | let tgt_idx = Instance.idx target |
506 | f23f21c3 | Iustin Pop | upd_cvar = compCV upd_nl |
507 | f23f21c3 | Iustin Pop | upd_il = Container.add tgt_idx new_inst ini_il |
508 | f23f21c3 | Iustin Pop | upd_plc = (tgt_idx, pri_idx, sec_idx, move, upd_cvar):ini_plc |
509 | f23f21c3 | Iustin Pop | upd_tbl = Table upd_nl upd_il upd_cvar upd_plc |
510 | f23f21c3 | Iustin Pop | in compareTables cur_tbl upd_tbl |
511 | e4f08c46 | Iustin Pop | |
512 | c0501c69 | Iustin Pop | -- | Given the status of the current secondary as a valid new node and |
513 | c0501c69 | Iustin Pop | -- the current candidate target node, generate the possible moves for |
514 | c0501c69 | Iustin Pop | -- a instance. |
515 | 5f4464db | Iustin Pop | possibleMoves :: MirrorType -- ^ The mirroring type of the instance |
516 | 5f4464db | Iustin Pop | -> Bool -- ^ Whether the secondary node is a valid new node |
517 | 5f4464db | Iustin Pop | -> Bool -- ^ Whether we can change the primary node |
518 | 5f4464db | Iustin Pop | -> Ndx -- ^ Target node candidate |
519 | 5f4464db | Iustin Pop | -> [IMove] -- ^ List of valid result moves |
520 | e08424a8 | Guido Trotter | |
521 | 5f4464db | Iustin Pop | possibleMoves MirrorNone _ _ _ = [] |
522 | e08424a8 | Guido Trotter | |
523 | 5f4464db | Iustin Pop | possibleMoves MirrorExternal _ _ _ = [] |
524 | 5f4464db | Iustin Pop | |
525 | 5f4464db | Iustin Pop | possibleMoves MirrorInternal _ False tdx = |
526 | 5f4464db | Iustin Pop | [ ReplaceSecondary tdx ] |
527 | 5f4464db | Iustin Pop | |
528 | 5f4464db | Iustin Pop | possibleMoves MirrorInternal True True tdx = |
529 | f23f21c3 | Iustin Pop | [ ReplaceSecondary tdx |
530 | f23f21c3 | Iustin Pop | , ReplaceAndFailover tdx |
531 | f23f21c3 | Iustin Pop | , ReplacePrimary tdx |
532 | f23f21c3 | Iustin Pop | , FailoverAndReplace tdx |
533 | f23f21c3 | Iustin Pop | ] |
534 | 40d4eba0 | Iustin Pop | |
535 | 5f4464db | Iustin Pop | possibleMoves MirrorInternal False True tdx = |
536 | f23f21c3 | Iustin Pop | [ ReplaceSecondary tdx |
537 | f23f21c3 | Iustin Pop | , ReplaceAndFailover tdx |
538 | f23f21c3 | Iustin Pop | ] |
539 | 40d4eba0 | Iustin Pop | |
540 | 40d4eba0 | Iustin Pop | -- | Compute the best move for a given instance. |
541 | c0501c69 | Iustin Pop | checkInstanceMove :: [Ndx] -- ^ Allowed target node indices |
542 | c0501c69 | Iustin Pop | -> Bool -- ^ Whether disk moves are allowed |
543 | e08424a8 | Guido Trotter | -> Bool -- ^ Whether instance moves are allowed |
544 | c0501c69 | Iustin Pop | -> Table -- ^ Original table |
545 | c0501c69 | Iustin Pop | -> Instance.Instance -- ^ Instance to move |
546 | c0501c69 | Iustin Pop | -> Table -- ^ Best new table for this instance |
547 | e08424a8 | Guido Trotter | checkInstanceMove nodes_idx disk_moves inst_moves ini_tbl target = |
548 | f23f21c3 | Iustin Pop | let opdx = Instance.pNode target |
549 | f23f21c3 | Iustin Pop | osdx = Instance.sNode target |
550 | 3603605a | Iustin Pop | bad_nodes = [opdx, osdx] |
551 | 3603605a | Iustin Pop | nodes = filter (`notElem` bad_nodes) nodes_idx |
552 | 5f4464db | Iustin Pop | mir_type = templateMirrorType $ Instance.diskTemplate target |
553 | f23f21c3 | Iustin Pop | use_secondary = elem osdx nodes_idx && inst_moves |
554 | 5f4464db | Iustin Pop | aft_failover = if mir_type == MirrorInternal && use_secondary |
555 | 5f4464db | Iustin Pop | -- if drbd and allowed to failover |
556 | 40d4eba0 | Iustin Pop | then checkSingleStep ini_tbl target ini_tbl Failover |
557 | 40d4eba0 | Iustin Pop | else ini_tbl |
558 | 5f4464db | Iustin Pop | all_moves = |
559 | 5f4464db | Iustin Pop | if disk_moves |
560 | 5f4464db | Iustin Pop | then concatMap (possibleMoves mir_type use_secondary inst_moves) |
561 | 5f4464db | Iustin Pop | nodes |
562 | 5f4464db | Iustin Pop | else [] |
563 | 4e25d1c2 | Iustin Pop | in |
564 | 4e25d1c2 | Iustin Pop | -- iterate over the possible nodes for this instance |
565 | 9dc6023f | Iustin Pop | foldl' (checkSingleStep ini_tbl target) aft_failover all_moves |
566 | 4e25d1c2 | Iustin Pop | |
567 | e4f08c46 | Iustin Pop | -- | Compute the best next move. |
568 | 608efcce | Iustin Pop | checkMove :: [Ndx] -- ^ Allowed target node indices |
569 | c0501c69 | Iustin Pop | -> Bool -- ^ Whether disk moves are allowed |
570 | e08424a8 | Guido Trotter | -> Bool -- ^ Whether instance moves are allowed |
571 | 256810de | Iustin Pop | -> Table -- ^ The current solution |
572 | e4f08c46 | Iustin Pop | -> [Instance.Instance] -- ^ List of instances still to move |
573 | 256810de | Iustin Pop | -> Table -- ^ The new solution |
574 | e08424a8 | Guido Trotter | checkMove nodes_idx disk_moves inst_moves ini_tbl victims = |
575 | f23f21c3 | Iustin Pop | let Table _ _ _ ini_plc = ini_tbl |
576 | f23f21c3 | Iustin Pop | -- we're using rwhnf from the Control.Parallel.Strategies |
577 | f23f21c3 | Iustin Pop | -- package; we don't need to use rnf as that would force too |
578 | f23f21c3 | Iustin Pop | -- much evaluation in single-threaded cases, and in |
579 | f23f21c3 | Iustin Pop | -- multi-threaded case the weak head normal form is enough to |
580 | f23f21c3 | Iustin Pop | -- spark the evaluation |
581 | f23f21c3 | Iustin Pop | tables = parMap rwhnf (checkInstanceMove nodes_idx disk_moves |
582 | f23f21c3 | Iustin Pop | inst_moves ini_tbl) |
583 | f23f21c3 | Iustin Pop | victims |
584 | f23f21c3 | Iustin Pop | -- iterate over all instances, computing the best move |
585 | f23f21c3 | Iustin Pop | best_tbl = foldl' compareTables ini_tbl tables |
586 | f23f21c3 | Iustin Pop | Table _ _ _ best_plc = best_tbl |
587 | f23f21c3 | Iustin Pop | in if length best_plc == length ini_plc |
588 | a804261a | Iustin Pop | then ini_tbl -- no advancement |
589 | a804261a | Iustin Pop | else best_tbl |
590 | e4f08c46 | Iustin Pop | |
591 | 525bfb36 | Iustin Pop | -- | Check if we are allowed to go deeper in the balancing. |
592 | 3fea6959 | Iustin Pop | doNextBalance :: Table -- ^ The starting table |
593 | 3fea6959 | Iustin Pop | -> Int -- ^ Remaining length |
594 | 3fea6959 | Iustin Pop | -> Score -- ^ Score at which to stop |
595 | 3fea6959 | Iustin Pop | -> Bool -- ^ The resulting table and commands |
596 | 5ad86777 | Iustin Pop | doNextBalance ini_tbl max_rounds min_score = |
597 | f23f21c3 | Iustin Pop | let Table _ _ ini_cv ini_plc = ini_tbl |
598 | f23f21c3 | Iustin Pop | ini_plc_len = length ini_plc |
599 | f23f21c3 | Iustin Pop | in (max_rounds < 0 || ini_plc_len < max_rounds) && ini_cv > min_score |
600 | 5ad86777 | Iustin Pop | |
601 | 525bfb36 | Iustin Pop | -- | Run a balance move. |
602 | f25e5aac | Iustin Pop | tryBalance :: Table -- ^ The starting table |
603 | f25e5aac | Iustin Pop | -> Bool -- ^ Allow disk moves |
604 | e08424a8 | Guido Trotter | -> Bool -- ^ Allow instance moves |
605 | 2e28ac32 | Iustin Pop | -> Bool -- ^ Only evacuate moves |
606 | 848b65c9 | Iustin Pop | -> Score -- ^ Min gain threshold |
607 | 848b65c9 | Iustin Pop | -> Score -- ^ Min gain |
608 | f25e5aac | Iustin Pop | -> Maybe Table -- ^ The resulting table and commands |
609 | e08424a8 | Guido Trotter | tryBalance ini_tbl disk_moves inst_moves evac_mode mg_limit min_gain = |
610 | 5ad86777 | Iustin Pop | let Table ini_nl ini_il ini_cv _ = ini_tbl |
611 | 5ad86777 | Iustin Pop | all_inst = Container.elems ini_il |
612 | 73d12eab | Iustin Pop | all_nodes = Container.elems ini_nl |
613 | 73d12eab | Iustin Pop | (offline_nodes, online_nodes) = partition Node.offline all_nodes |
614 | 2e28ac32 | Iustin Pop | all_inst' = if evac_mode |
615 | 73d12eab | Iustin Pop | then let bad_nodes = map Node.idx offline_nodes |
616 | 73d12eab | Iustin Pop | in filter (any (`elem` bad_nodes) . |
617 | 73d12eab | Iustin Pop | Instance.allNodes) all_inst |
618 | 73d12eab | Iustin Pop | else all_inst |
619 | c424cdc8 | Iustin Pop | reloc_inst = filter Instance.movable all_inst' |
620 | 73d12eab | Iustin Pop | node_idx = map Node.idx online_nodes |
621 | e08424a8 | Guido Trotter | fin_tbl = checkMove node_idx disk_moves inst_moves ini_tbl reloc_inst |
622 | 5ad86777 | Iustin Pop | (Table _ _ fin_cv _) = fin_tbl |
623 | f25e5aac | Iustin Pop | in |
624 | 848b65c9 | Iustin Pop | if fin_cv < ini_cv && (ini_cv > mg_limit || ini_cv - fin_cv >= min_gain) |
625 | 5ad86777 | Iustin Pop | then Just fin_tbl -- this round made success, return the new table |
626 | f25e5aac | Iustin Pop | else Nothing |
627 | f25e5aac | Iustin Pop | |
628 | 478df686 | Iustin Pop | -- * Allocation functions |
629 | 478df686 | Iustin Pop | |
630 | 525bfb36 | Iustin Pop | -- | Build failure stats out of a list of failures. |
631 | 478df686 | Iustin Pop | collapseFailures :: [FailMode] -> FailStats |
632 | 478df686 | Iustin Pop | collapseFailures flst = |
633 | b4bae394 | Iustin Pop | map (\k -> (k, foldl' (\a e -> if e == k then a + 1 else a) 0 flst)) |
634 | b4bae394 | Iustin Pop | [minBound..maxBound] |
635 | 478df686 | Iustin Pop | |
636 | d7339c99 | Iustin Pop | -- | Compares two Maybe AllocElement and chooses the besst score. |
637 | d7339c99 | Iustin Pop | bestAllocElement :: Maybe Node.AllocElement |
638 | d7339c99 | Iustin Pop | -> Maybe Node.AllocElement |
639 | d7339c99 | Iustin Pop | -> Maybe Node.AllocElement |
640 | d7339c99 | Iustin Pop | bestAllocElement a Nothing = a |
641 | d7339c99 | Iustin Pop | bestAllocElement Nothing b = b |
642 | d7339c99 | Iustin Pop | bestAllocElement a@(Just (_, _, _, ascore)) b@(Just (_, _, _, bscore)) = |
643 | 9fc18384 | Iustin Pop | if ascore < bscore then a else b |
644 | d7339c99 | Iustin Pop | |
645 | 478df686 | Iustin Pop | -- | Update current Allocation solution and failure stats with new |
646 | 525bfb36 | Iustin Pop | -- elements. |
647 | 1fe81531 | Iustin Pop | concatAllocs :: AllocSolution -> OpResult Node.AllocElement -> AllocSolution |
648 | 85d0ddc3 | Iustin Pop | concatAllocs as (OpFail reason) = as { asFailures = reason : asFailures as } |
649 | 478df686 | Iustin Pop | |
650 | d7339c99 | Iustin Pop | concatAllocs as (OpGood ns) = |
651 | 9fc18384 | Iustin Pop | let -- Choose the old or new solution, based on the cluster score |
652 | 9fc18384 | Iustin Pop | cntok = asAllocs as |
653 | 9fc18384 | Iustin Pop | osols = asSolution as |
654 | 9fc18384 | Iustin Pop | nsols = bestAllocElement osols (Just ns) |
655 | 9fc18384 | Iustin Pop | nsuc = cntok + 1 |
656 | 478df686 | Iustin Pop | -- Note: we force evaluation of nsols here in order to keep the |
657 | 478df686 | Iustin Pop | -- memory profile low - we know that we will need nsols for sure |
658 | 478df686 | Iustin Pop | -- in the next cycle, so we force evaluation of nsols, since the |
659 | 478df686 | Iustin Pop | -- foldl' in the caller will only evaluate the tuple, but not the |
660 | 7d11799b | Iustin Pop | -- elements of the tuple |
661 | 9fc18384 | Iustin Pop | in nsols `seq` nsuc `seq` as { asAllocs = nsuc, asSolution = nsols } |
662 | dbba5246 | Iustin Pop | |
663 | f828f4aa | Iustin Pop | -- | Sums two 'AllocSolution' structures. |
664 | f828f4aa | Iustin Pop | sumAllocs :: AllocSolution -> AllocSolution -> AllocSolution |
665 | f828f4aa | Iustin Pop | sumAllocs (AllocSolution aFails aAllocs aSols aLog) |
666 | f828f4aa | Iustin Pop | (AllocSolution bFails bAllocs bSols bLog) = |
667 | 9fc18384 | Iustin Pop | -- note: we add b first, since usually it will be smaller; when |
668 | 9fc18384 | Iustin Pop | -- fold'ing, a will grow and grow whereas b is the per-group |
669 | 9fc18384 | Iustin Pop | -- result, hence smaller |
670 | 9fc18384 | Iustin Pop | let nFails = bFails ++ aFails |
671 | 9fc18384 | Iustin Pop | nAllocs = aAllocs + bAllocs |
672 | 9fc18384 | Iustin Pop | nSols = bestAllocElement aSols bSols |
673 | 9fc18384 | Iustin Pop | nLog = bLog ++ aLog |
674 | 9fc18384 | Iustin Pop | in AllocSolution nFails nAllocs nSols nLog |
675 | f828f4aa | Iustin Pop | |
676 | 525bfb36 | Iustin Pop | -- | Given a solution, generates a reasonable description for it. |
677 | 859fc11d | Iustin Pop | describeSolution :: AllocSolution -> String |
678 | 859fc11d | Iustin Pop | describeSolution as = |
679 | 859fc11d | Iustin Pop | let fcnt = asFailures as |
680 | 129734d3 | Iustin Pop | sols = asSolution as |
681 | 859fc11d | Iustin Pop | freasons = |
682 | 859fc11d | Iustin Pop | intercalate ", " . map (\(a, b) -> printf "%s: %d" (show a) b) . |
683 | 859fc11d | Iustin Pop | filter ((> 0) . snd) . collapseFailures $ fcnt |
684 | 129734d3 | Iustin Pop | in case sols of |
685 | 129734d3 | Iustin Pop | Nothing -> "No valid allocation solutions, failure reasons: " ++ |
686 | 129734d3 | Iustin Pop | (if null fcnt then "unknown reasons" else freasons) |
687 | 129734d3 | Iustin Pop | Just (_, _, nodes, cv) -> |
688 | 129734d3 | Iustin Pop | printf ("score: %.8f, successes %d, failures %d (%s)" ++ |
689 | 129734d3 | Iustin Pop | " for node(s) %s") cv (asAllocs as) (length fcnt) freasons |
690 | 129734d3 | Iustin Pop | (intercalate "/" . map Node.name $ nodes) |
691 | 859fc11d | Iustin Pop | |
692 | 525bfb36 | Iustin Pop | -- | Annotates a solution with the appropriate string. |
693 | 859fc11d | Iustin Pop | annotateSolution :: AllocSolution -> AllocSolution |
694 | 859fc11d | Iustin Pop | annotateSolution as = as { asLog = describeSolution as : asLog as } |
695 | 859fc11d | Iustin Pop | |
696 | 47eed3f4 | Iustin Pop | -- | Reverses an evacuation solution. |
697 | 47eed3f4 | Iustin Pop | -- |
698 | 47eed3f4 | Iustin Pop | -- Rationale: we always concat the results to the top of the lists, so |
699 | 47eed3f4 | Iustin Pop | -- for proper jobset execution, we should reverse all lists. |
700 | 47eed3f4 | Iustin Pop | reverseEvacSolution :: EvacSolution -> EvacSolution |
701 | 47eed3f4 | Iustin Pop | reverseEvacSolution (EvacSolution f m o) = |
702 | 9fc18384 | Iustin Pop | EvacSolution (reverse f) (reverse m) (reverse o) |
703 | 47eed3f4 | Iustin Pop | |
704 | 6cb1649f | Iustin Pop | -- | Generate the valid node allocation singles or pairs for a new instance. |
705 | 6d0bc5ca | Iustin Pop | genAllocNodes :: Group.List -- ^ Group list |
706 | 6d0bc5ca | Iustin Pop | -> Node.List -- ^ The node map |
707 | 6cb1649f | Iustin Pop | -> Int -- ^ The number of nodes required |
708 | 6d0bc5ca | Iustin Pop | -> Bool -- ^ Whether to drop or not |
709 | 6d0bc5ca | Iustin Pop | -- unallocable nodes |
710 | 6cb1649f | Iustin Pop | -> Result AllocNodes -- ^ The (monadic) result |
711 | 6d0bc5ca | Iustin Pop | genAllocNodes gl nl count drop_unalloc = |
712 | 9fc18384 | Iustin Pop | let filter_fn = if drop_unalloc |
713 | e4491427 | Iustin Pop | then filter (Group.isAllocable . |
714 | e4491427 | Iustin Pop | flip Container.find gl . Node.group) |
715 | 6d0bc5ca | Iustin Pop | else id |
716 | 9fc18384 | Iustin Pop | all_nodes = filter_fn $ getOnline nl |
717 | 9fc18384 | Iustin Pop | all_pairs = [(Node.idx p, |
718 | 9fc18384 | Iustin Pop | [Node.idx s | s <- all_nodes, |
719 | 9fc18384 | Iustin Pop | Node.idx p /= Node.idx s, |
720 | 9fc18384 | Iustin Pop | Node.group p == Node.group s]) | |
721 | 9fc18384 | Iustin Pop | p <- all_nodes] |
722 | 9fc18384 | Iustin Pop | in case count of |
723 | 9fc18384 | Iustin Pop | 1 -> Ok (Left (map Node.idx all_nodes)) |
724 | 9fc18384 | Iustin Pop | 2 -> Ok (Right (filter (not . null . snd) all_pairs)) |
725 | 9fc18384 | Iustin Pop | _ -> Bad "Unsupported number of nodes, only one or two supported" |
726 | 6cb1649f | Iustin Pop | |
727 | dbba5246 | Iustin Pop | -- | Try to allocate an instance on the cluster. |
728 | dbba5246 | Iustin Pop | tryAlloc :: (Monad m) => |
729 | dbba5246 | Iustin Pop | Node.List -- ^ The node list |
730 | dbba5246 | Iustin Pop | -> Instance.List -- ^ The instance list |
731 | dbba5246 | Iustin Pop | -> Instance.Instance -- ^ The instance to allocate |
732 | 6cb1649f | Iustin Pop | -> AllocNodes -- ^ The allocation targets |
733 | 78694255 | Iustin Pop | -> m AllocSolution -- ^ Possible solution list |
734 | 1bf6d813 | Iustin Pop | tryAlloc _ _ _ (Right []) = fail "Not enough online nodes" |
735 | 6cb1649f | Iustin Pop | tryAlloc nl _ inst (Right ok_pairs) = |
736 | 9fc18384 | Iustin Pop | let psols = parMap rwhnf (\(p, ss) -> |
737 | 9fc18384 | Iustin Pop | foldl' (\cstate -> |
738 | 9fc18384 | Iustin Pop | concatAllocs cstate . |
739 | 9fc18384 | Iustin Pop | allocateOnPair nl inst p) |
740 | 9fc18384 | Iustin Pop | emptyAllocSolution ss) ok_pairs |
741 | 9fc18384 | Iustin Pop | sols = foldl' sumAllocs emptyAllocSolution psols |
742 | 9fc18384 | Iustin Pop | in return $ annotateSolution sols |
743 | dbba5246 | Iustin Pop | |
744 | 1bf6d813 | Iustin Pop | tryAlloc _ _ _ (Left []) = fail "No online nodes" |
745 | 6cb1649f | Iustin Pop | tryAlloc nl _ inst (Left all_nodes) = |
746 | 9fc18384 | Iustin Pop | let sols = foldl' (\cstate -> |
747 | 9fc18384 | Iustin Pop | concatAllocs cstate . allocateOnSingle nl inst |
748 | 9fc18384 | Iustin Pop | ) emptyAllocSolution all_nodes |
749 | 9fc18384 | Iustin Pop | in return $ annotateSolution sols |
750 | dbba5246 | Iustin Pop | |
751 | 525bfb36 | Iustin Pop | -- | Given a group/result, describe it as a nice (list of) messages. |
752 | aec636b9 | Iustin Pop | solutionDescription :: Group.List -> (Gdx, Result AllocSolution) -> [String] |
753 | aec636b9 | Iustin Pop | solutionDescription gl (groupId, result) = |
754 | 9b1584fc | Iustin Pop | case result of |
755 | 73206d0a | Iustin Pop | Ok solution -> map (printf "Group %s (%s): %s" gname pol) (asLog solution) |
756 | aec636b9 | Iustin Pop | Bad message -> [printf "Group %s: error %s" gname message] |
757 | 73206d0a | Iustin Pop | where grp = Container.find groupId gl |
758 | 73206d0a | Iustin Pop | gname = Group.name grp |
759 | 5f828ce4 | Agata Murawska | pol = allocPolicyToRaw (Group.allocPolicy grp) |
760 | 9b1584fc | Iustin Pop | |
761 | 9b1584fc | Iustin Pop | -- | From a list of possibly bad and possibly empty solutions, filter |
762 | 88253d03 | Iustin Pop | -- only the groups with a valid result. Note that the result will be |
763 | 525bfb36 | Iustin Pop | -- reversed compared to the original list. |
764 | 73206d0a | Iustin Pop | filterMGResults :: Group.List |
765 | 73206d0a | Iustin Pop | -> [(Gdx, Result AllocSolution)] |
766 | 73206d0a | Iustin Pop | -> [(Gdx, AllocSolution)] |
767 | 88253d03 | Iustin Pop | filterMGResults gl = foldl' fn [] |
768 | 9fc18384 | Iustin Pop | where unallocable = not . Group.isAllocable . flip Container.find gl |
769 | 9fc18384 | Iustin Pop | fn accu (gdx, rasol) = |
770 | 9fc18384 | Iustin Pop | case rasol of |
771 | 9fc18384 | Iustin Pop | Bad _ -> accu |
772 | 9fc18384 | Iustin Pop | Ok sol | isNothing (asSolution sol) -> accu |
773 | 9fc18384 | Iustin Pop | | unallocable gdx -> accu |
774 | 9fc18384 | Iustin Pop | | otherwise -> (gdx, sol):accu |
775 | 9b1584fc | Iustin Pop | |
776 | 525bfb36 | Iustin Pop | -- | Sort multigroup results based on policy and score. |
777 | 73206d0a | Iustin Pop | sortMGResults :: Group.List |
778 | 73206d0a | Iustin Pop | -> [(Gdx, AllocSolution)] |
779 | 73206d0a | Iustin Pop | -> [(Gdx, AllocSolution)] |
780 | 73206d0a | Iustin Pop | sortMGResults gl sols = |
781 | 9fc18384 | Iustin Pop | let extractScore (_, _, _, x) = x |
782 | 9fc18384 | Iustin Pop | solScore (gdx, sol) = (Group.allocPolicy (Container.find gdx gl), |
783 | 9fc18384 | Iustin Pop | (extractScore . fromJust . asSolution) sol) |
784 | 9fc18384 | Iustin Pop | in sortBy (comparing solScore) sols |
785 | 73206d0a | Iustin Pop | |
786 | 8fd09137 | Iustin Pop | -- | Finds the best group for an instance on a multi-group cluster. |
787 | d72ff6c3 | Iustin Pop | -- |
788 | d72ff6c3 | Iustin Pop | -- Only solutions in @preferred@ and @last_resort@ groups will be |
789 | d72ff6c3 | Iustin Pop | -- accepted as valid, and additionally if the allowed groups parameter |
790 | d72ff6c3 | Iustin Pop | -- is not null then allocation will only be run for those group |
791 | d72ff6c3 | Iustin Pop | -- indices. |
792 | 8fd09137 | Iustin Pop | findBestAllocGroup :: Group.List -- ^ The group list |
793 | 8fd09137 | Iustin Pop | -> Node.List -- ^ The node list |
794 | 8fd09137 | Iustin Pop | -> Instance.List -- ^ The instance list |
795 | d72ff6c3 | Iustin Pop | -> Maybe [Gdx] -- ^ The allowed groups |
796 | 8fd09137 | Iustin Pop | -> Instance.Instance -- ^ The instance to allocate |
797 | 8fd09137 | Iustin Pop | -> Int -- ^ Required number of nodes |
798 | 8fd09137 | Iustin Pop | -> Result (Gdx, AllocSolution, [String]) |
799 | d72ff6c3 | Iustin Pop | findBestAllocGroup mggl mgnl mgil allowed_gdxs inst cnt = |
800 | 9b1584fc | Iustin Pop | let groups = splitCluster mgnl mgil |
801 | d72ff6c3 | Iustin Pop | groups' = maybe groups (\gs -> filter ((`elem` gs) . fst) groups) |
802 | d72ff6c3 | Iustin Pop | allowed_gdxs |
803 | 9b1584fc | Iustin Pop | sols = map (\(gid, (nl, il)) -> |
804 | 6d0bc5ca | Iustin Pop | (gid, genAllocNodes mggl nl cnt False >>= |
805 | 6d0bc5ca | Iustin Pop | tryAlloc nl il inst)) |
806 | d72ff6c3 | Iustin Pop | groups'::[(Gdx, Result AllocSolution)] |
807 | aec636b9 | Iustin Pop | all_msgs = concatMap (solutionDescription mggl) sols |
808 | 73206d0a | Iustin Pop | goodSols = filterMGResults mggl sols |
809 | 73206d0a | Iustin Pop | sortedSols = sortMGResults mggl goodSols |
810 | 9b1584fc | Iustin Pop | in if null sortedSols |
811 | 6a855aaa | Iustin Pop | then if null groups' |
812 | 6a855aaa | Iustin Pop | then Bad $ "no groups for evacuation: allowed groups was" ++ |
813 | 6a855aaa | Iustin Pop | show allowed_gdxs ++ ", all groups: " ++ |
814 | 6a855aaa | Iustin Pop | show (map fst groups) |
815 | 6a855aaa | Iustin Pop | else Bad $ intercalate ", " all_msgs |
816 | 9fc18384 | Iustin Pop | else let (final_group, final_sol) = head sortedSols |
817 | 9fc18384 | Iustin Pop | in return (final_group, final_sol, all_msgs) |
818 | 8fd09137 | Iustin Pop | |
819 | 8fd09137 | Iustin Pop | -- | Try to allocate an instance on a multi-group cluster. |
820 | 8fd09137 | Iustin Pop | tryMGAlloc :: Group.List -- ^ The group list |
821 | 8fd09137 | Iustin Pop | -> Node.List -- ^ The node list |
822 | 8fd09137 | Iustin Pop | -> Instance.List -- ^ The instance list |
823 | 8fd09137 | Iustin Pop | -> Instance.Instance -- ^ The instance to allocate |
824 | 8fd09137 | Iustin Pop | -> Int -- ^ Required number of nodes |
825 | 8fd09137 | Iustin Pop | -> Result AllocSolution -- ^ Possible solution list |
826 | 8fd09137 | Iustin Pop | tryMGAlloc mggl mgnl mgil inst cnt = do |
827 | 8fd09137 | Iustin Pop | (best_group, solution, all_msgs) <- |
828 | d72ff6c3 | Iustin Pop | findBestAllocGroup mggl mgnl mgil Nothing inst cnt |
829 | 8fd09137 | Iustin Pop | let group_name = Group.name $ Container.find best_group mggl |
830 | 8fd09137 | Iustin Pop | selmsg = "Selected group: " ++ group_name |
831 | 8fd09137 | Iustin Pop | return $ solution { asLog = selmsg:all_msgs } |
832 | 9b1584fc | Iustin Pop | |
833 | 47eed3f4 | Iustin Pop | -- | Function which fails if the requested mode is change secondary. |
834 | 47eed3f4 | Iustin Pop | -- |
835 | 47eed3f4 | Iustin Pop | -- This is useful since except DRBD, no other disk template can |
836 | 47eed3f4 | Iustin Pop | -- execute change secondary; thus, we can just call this function |
837 | 47eed3f4 | Iustin Pop | -- instead of always checking for secondary mode. After the call to |
838 | 47eed3f4 | Iustin Pop | -- this function, whatever mode we have is just a primary change. |
839 | 47eed3f4 | Iustin Pop | failOnSecondaryChange :: (Monad m) => EvacMode -> DiskTemplate -> m () |
840 | 47eed3f4 | Iustin Pop | failOnSecondaryChange ChangeSecondary dt = |
841 | 9fc18384 | Iustin Pop | fail $ "Instances with disk template '" ++ diskTemplateToRaw dt ++ |
842 | 47eed3f4 | Iustin Pop | "' can't execute change secondary" |
843 | 47eed3f4 | Iustin Pop | failOnSecondaryChange _ _ = return () |
844 | 47eed3f4 | Iustin Pop | |
845 | 47eed3f4 | Iustin Pop | -- | Run evacuation for a single instance. |
846 | 20b376ff | Iustin Pop | -- |
847 | 20b376ff | Iustin Pop | -- /Note:/ this function should correctly execute both intra-group |
848 | 20b376ff | Iustin Pop | -- evacuations (in all modes) and inter-group evacuations (in the |
849 | 20b376ff | Iustin Pop | -- 'ChangeAll' mode). Of course, this requires that the correct list |
850 | 20b376ff | Iustin Pop | -- of target nodes is passed. |
851 | 47eed3f4 | Iustin Pop | nodeEvacInstance :: Node.List -- ^ The node list (cluster-wide) |
852 | 47eed3f4 | Iustin Pop | -> Instance.List -- ^ Instance list (cluster-wide) |
853 | 47eed3f4 | Iustin Pop | -> EvacMode -- ^ The evacuation mode |
854 | 47eed3f4 | Iustin Pop | -> Instance.Instance -- ^ The instance to be evacuated |
855 | a86fbf36 | Iustin Pop | -> Gdx -- ^ The group we're targetting |
856 | 47eed3f4 | Iustin Pop | -> [Ndx] -- ^ The list of available nodes |
857 | 47eed3f4 | Iustin Pop | -- for allocation |
858 | 47eed3f4 | Iustin Pop | -> Result (Node.List, Instance.List, [OpCodes.OpCode]) |
859 | 47eed3f4 | Iustin Pop | nodeEvacInstance _ _ mode (Instance.Instance |
860 | a86fbf36 | Iustin Pop | {Instance.diskTemplate = dt@DTDiskless}) _ _ = |
861 | 47eed3f4 | Iustin Pop | failOnSecondaryChange mode dt >> |
862 | 47eed3f4 | Iustin Pop | fail "Diskless relocations not implemented yet" |
863 | 47eed3f4 | Iustin Pop | |
864 | 47eed3f4 | Iustin Pop | nodeEvacInstance _ _ _ (Instance.Instance |
865 | a86fbf36 | Iustin Pop | {Instance.diskTemplate = DTPlain}) _ _ = |
866 | 47eed3f4 | Iustin Pop | fail "Instances of type plain cannot be relocated" |
867 | 47eed3f4 | Iustin Pop | |
868 | 47eed3f4 | Iustin Pop | nodeEvacInstance _ _ _ (Instance.Instance |
869 | a86fbf36 | Iustin Pop | {Instance.diskTemplate = DTFile}) _ _ = |
870 | 47eed3f4 | Iustin Pop | fail "Instances of type file cannot be relocated" |
871 | 47eed3f4 | Iustin Pop | |
872 | 47eed3f4 | Iustin Pop | nodeEvacInstance _ _ mode (Instance.Instance |
873 | a86fbf36 | Iustin Pop | {Instance.diskTemplate = dt@DTSharedFile}) _ _ = |
874 | 47eed3f4 | Iustin Pop | failOnSecondaryChange mode dt >> |
875 | 47eed3f4 | Iustin Pop | fail "Shared file relocations not implemented yet" |
876 | 47eed3f4 | Iustin Pop | |
877 | 47eed3f4 | Iustin Pop | nodeEvacInstance _ _ mode (Instance.Instance |
878 | a86fbf36 | Iustin Pop | {Instance.diskTemplate = dt@DTBlock}) _ _ = |
879 | 47eed3f4 | Iustin Pop | failOnSecondaryChange mode dt >> |
880 | 47eed3f4 | Iustin Pop | fail "Block device relocations not implemented yet" |
881 | 47eed3f4 | Iustin Pop | |
882 | bdd6931c | Guido Trotter | nodeEvacInstance _ _ mode (Instance.Instance |
883 | bdd6931c | Guido Trotter | {Instance.diskTemplate = dt@DTRbd}) _ _ = |
884 | bdd6931c | Guido Trotter | failOnSecondaryChange mode dt >> |
885 | bdd6931c | Guido Trotter | fail "Rbd relocations not implemented yet" |
886 | bdd6931c | Guido Trotter | |
887 | bef83fd1 | Iustin Pop | nodeEvacInstance nl il ChangePrimary |
888 | a86fbf36 | Iustin Pop | inst@(Instance.Instance {Instance.diskTemplate = DTDrbd8}) |
889 | a86fbf36 | Iustin Pop | _ _ = |
890 | bef83fd1 | Iustin Pop | do |
891 | bef83fd1 | Iustin Pop | (nl', inst', _, _) <- opToResult $ applyMove nl inst Failover |
892 | bef83fd1 | Iustin Pop | let idx = Instance.idx inst |
893 | bef83fd1 | Iustin Pop | il' = Container.add idx inst' il |
894 | bef83fd1 | Iustin Pop | ops = iMoveToJob nl' il' idx Failover |
895 | bef83fd1 | Iustin Pop | return (nl', il', ops) |
896 | bef83fd1 | Iustin Pop | |
897 | db56cfc4 | Iustin Pop | nodeEvacInstance nl il ChangeSecondary |
898 | db56cfc4 | Iustin Pop | inst@(Instance.Instance {Instance.diskTemplate = DTDrbd8}) |
899 | a86fbf36 | Iustin Pop | gdx avail_nodes = |
900 | db56cfc4 | Iustin Pop | do |
901 | 80b27509 | Iustin Pop | let op_fn = ReplaceSecondary |
902 | db56cfc4 | Iustin Pop | (nl', inst', _, ndx) <- annotateResult "Can't find any good node" $ |
903 | db56cfc4 | Iustin Pop | eitherToResult $ |
904 | 80b27509 | Iustin Pop | foldl' (evacDrbdSecondaryInner nl inst gdx op_fn) |
905 | db56cfc4 | Iustin Pop | (Left "no nodes available") avail_nodes |
906 | db56cfc4 | Iustin Pop | let idx = Instance.idx inst |
907 | db56cfc4 | Iustin Pop | il' = Container.add idx inst' il |
908 | 80b27509 | Iustin Pop | ops = iMoveToJob nl' il' idx (op_fn ndx) |
909 | db56cfc4 | Iustin Pop | return (nl', il', ops) |
910 | db56cfc4 | Iustin Pop | |
911 | 97da6b71 | Iustin Pop | -- The algorithm for ChangeAll is as follows: |
912 | 97da6b71 | Iustin Pop | -- |
913 | 97da6b71 | Iustin Pop | -- * generate all (primary, secondary) node pairs for the target groups |
914 | 97da6b71 | Iustin Pop | -- * for each pair, execute the needed moves (r:s, f, r:s) and compute |
915 | 97da6b71 | Iustin Pop | -- the final node list state and group score |
916 | 97da6b71 | Iustin Pop | -- * select the best choice via a foldl that uses the same Either |
917 | 97da6b71 | Iustin Pop | -- String solution as the ChangeSecondary mode |
918 | d52d41de | Iustin Pop | nodeEvacInstance nl il ChangeAll |
919 | d52d41de | Iustin Pop | inst@(Instance.Instance {Instance.diskTemplate = DTDrbd8}) |
920 | a86fbf36 | Iustin Pop | gdx avail_nodes = |
921 | d52d41de | Iustin Pop | do |
922 | 97da6b71 | Iustin Pop | let no_nodes = Left "no nodes available" |
923 | 97da6b71 | Iustin Pop | node_pairs = [(p,s) | p <- avail_nodes, s <- avail_nodes, p /= s] |
924 | 97da6b71 | Iustin Pop | (nl', il', ops, _) <- |
925 | 97da6b71 | Iustin Pop | annotateResult "Can't find any good nodes for relocation" $ |
926 | d52d41de | Iustin Pop | eitherToResult $ |
927 | 97da6b71 | Iustin Pop | foldl' |
928 | 97da6b71 | Iustin Pop | (\accu nodes -> case evacDrbdAllInner nl il inst gdx nodes of |
929 | 97da6b71 | Iustin Pop | Bad msg -> |
930 | 97da6b71 | Iustin Pop | case accu of |
931 | 97da6b71 | Iustin Pop | Right _ -> accu |
932 | 97da6b71 | Iustin Pop | -- we don't need more details (which |
933 | 97da6b71 | Iustin Pop | -- nodes, etc.) as we only selected |
934 | 97da6b71 | Iustin Pop | -- this group if we can allocate on |
935 | 97da6b71 | Iustin Pop | -- it, hence failures will not |
936 | 97da6b71 | Iustin Pop | -- propagate out of this fold loop |
937 | 97da6b71 | Iustin Pop | Left _ -> Left $ "Allocation failed: " ++ msg |
938 | 97da6b71 | Iustin Pop | Ok result@(_, _, _, new_cv) -> |
939 | 97da6b71 | Iustin Pop | let new_accu = Right result in |
940 | 97da6b71 | Iustin Pop | case accu of |
941 | 97da6b71 | Iustin Pop | Left _ -> new_accu |
942 | 97da6b71 | Iustin Pop | Right (_, _, _, old_cv) -> |
943 | 97da6b71 | Iustin Pop | if old_cv < new_cv |
944 | 97da6b71 | Iustin Pop | then accu |
945 | 97da6b71 | Iustin Pop | else new_accu |
946 | 97da6b71 | Iustin Pop | ) no_nodes node_pairs |
947 | 97da6b71 | Iustin Pop | |
948 | 97da6b71 | Iustin Pop | return (nl', il', ops) |
949 | 47eed3f4 | Iustin Pop | |
950 | db56cfc4 | Iustin Pop | -- | Inner fold function for changing secondary of a DRBD instance. |
951 | db56cfc4 | Iustin Pop | -- |
952 | 97da6b71 | Iustin Pop | -- The running solution is either a @Left String@, which means we |
953 | db56cfc4 | Iustin Pop | -- don't have yet a working solution, or a @Right (...)@, which |
954 | db56cfc4 | Iustin Pop | -- represents a valid solution; it holds the modified node list, the |
955 | db56cfc4 | Iustin Pop | -- modified instance (after evacuation), the score of that solution, |
956 | db56cfc4 | Iustin Pop | -- and the new secondary node index. |
957 | db56cfc4 | Iustin Pop | evacDrbdSecondaryInner :: Node.List -- ^ Cluster node list |
958 | db56cfc4 | Iustin Pop | -> Instance.Instance -- ^ Instance being evacuated |
959 | db56cfc4 | Iustin Pop | -> Gdx -- ^ The group index of the instance |
960 | 80b27509 | Iustin Pop | -> (Ndx -> IMove) -- ^ Operation constructor |
961 | bebe7a73 | Iustin Pop | -> EvacInnerState -- ^ Current best solution |
962 | db56cfc4 | Iustin Pop | -> Ndx -- ^ Node we're evaluating as new secondary |
963 | bebe7a73 | Iustin Pop | -> EvacInnerState -- ^ New best solution |
964 | 80b27509 | Iustin Pop | evacDrbdSecondaryInner nl inst gdx op_fn accu ndx = |
965 | 80b27509 | Iustin Pop | case applyMove nl inst (op_fn ndx) of |
966 | 9fc18384 | Iustin Pop | OpFail fm -> |
967 | 9fc18384 | Iustin Pop | case accu of |
968 | 9fc18384 | Iustin Pop | Right _ -> accu |
969 | 9fc18384 | Iustin Pop | Left _ -> Left $ "Node " ++ Container.nameOf nl ndx ++ |
970 | 9fc18384 | Iustin Pop | " failed: " ++ show fm |
971 | 9fc18384 | Iustin Pop | OpGood (nl', inst', _, _) -> |
972 | 9fc18384 | Iustin Pop | let nodes = Container.elems nl' |
973 | 9fc18384 | Iustin Pop | -- The fromJust below is ugly (it can fail nastily), but |
974 | 9fc18384 | Iustin Pop | -- at this point we should have any internal mismatches, |
975 | 9fc18384 | Iustin Pop | -- and adding a monad here would be quite involved |
976 | 9fc18384 | Iustin Pop | grpnodes = fromJust (gdx `lookup` Node.computeGroups nodes) |
977 | 9fc18384 | Iustin Pop | new_cv = compCVNodes grpnodes |
978 | 9fc18384 | Iustin Pop | new_accu = Right (nl', inst', new_cv, ndx) |
979 | 9fc18384 | Iustin Pop | in case accu of |
980 | 9fc18384 | Iustin Pop | Left _ -> new_accu |
981 | 9fc18384 | Iustin Pop | Right (_, _, old_cv, _) -> |
982 | 9fc18384 | Iustin Pop | if old_cv < new_cv |
983 | 9fc18384 | Iustin Pop | then accu |
984 | 9fc18384 | Iustin Pop | else new_accu |
985 | db56cfc4 | Iustin Pop | |
986 | 97da6b71 | Iustin Pop | -- | Compute result of changing all nodes of a DRBD instance. |
987 | 97da6b71 | Iustin Pop | -- |
988 | 97da6b71 | Iustin Pop | -- Given the target primary and secondary node (which might be in a |
989 | 97da6b71 | Iustin Pop | -- different group or not), this function will 'execute' all the |
990 | 97da6b71 | Iustin Pop | -- required steps and assuming all operations succceed, will return |
991 | 97da6b71 | Iustin Pop | -- the modified node and instance lists, the opcodes needed for this |
992 | 97da6b71 | Iustin Pop | -- and the new group score. |
993 | 97da6b71 | Iustin Pop | evacDrbdAllInner :: Node.List -- ^ Cluster node list |
994 | 97da6b71 | Iustin Pop | -> Instance.List -- ^ Cluster instance list |
995 | 97da6b71 | Iustin Pop | -> Instance.Instance -- ^ The instance to be moved |
996 | 97da6b71 | Iustin Pop | -> Gdx -- ^ The target group index |
997 | 97da6b71 | Iustin Pop | -- (which can differ from the |
998 | 97da6b71 | Iustin Pop | -- current group of the |
999 | 97da6b71 | Iustin Pop | -- instance) |
1000 | 97da6b71 | Iustin Pop | -> (Ndx, Ndx) -- ^ Tuple of new |
1001 | 97da6b71 | Iustin Pop | -- primary\/secondary nodes |
1002 | 97da6b71 | Iustin Pop | -> Result (Node.List, Instance.List, [OpCodes.OpCode], Score) |
1003 | 9fc18384 | Iustin Pop | evacDrbdAllInner nl il inst gdx (t_pdx, t_sdx) = do |
1004 | 9fc18384 | Iustin Pop | let primary = Container.find (Instance.pNode inst) nl |
1005 | 9fc18384 | Iustin Pop | idx = Instance.idx inst |
1006 | 9fc18384 | Iustin Pop | -- if the primary is offline, then we first failover |
1007 | 9fc18384 | Iustin Pop | (nl1, inst1, ops1) <- |
1008 | 9fc18384 | Iustin Pop | if Node.offline primary |
1009 | 9fc18384 | Iustin Pop | then do |
1010 | 9fc18384 | Iustin Pop | (nl', inst', _, _) <- |
1011 | 9fc18384 | Iustin Pop | annotateResult "Failing over to the secondary" $ |
1012 | 9fc18384 | Iustin Pop | opToResult $ applyMove nl inst Failover |
1013 | 9fc18384 | Iustin Pop | return (nl', inst', [Failover]) |
1014 | 9fc18384 | Iustin Pop | else return (nl, inst, []) |
1015 | 9fc18384 | Iustin Pop | let (o1, o2, o3) = (ReplaceSecondary t_pdx, |
1016 | 9fc18384 | Iustin Pop | Failover, |
1017 | 9fc18384 | Iustin Pop | ReplaceSecondary t_sdx) |
1018 | 9fc18384 | Iustin Pop | -- we now need to execute a replace secondary to the future |
1019 | 9fc18384 | Iustin Pop | -- primary node |
1020 | 9fc18384 | Iustin Pop | (nl2, inst2, _, _) <- |
1021 | 9fc18384 | Iustin Pop | annotateResult "Changing secondary to new primary" $ |
1022 | 9fc18384 | Iustin Pop | opToResult $ |
1023 | 9fc18384 | Iustin Pop | applyMove nl1 inst1 o1 |
1024 | 9fc18384 | Iustin Pop | let ops2 = o1:ops1 |
1025 | 9fc18384 | Iustin Pop | -- we now execute another failover, the primary stays fixed now |
1026 | 9fc18384 | Iustin Pop | (nl3, inst3, _, _) <- annotateResult "Failing over to new primary" $ |
1027 | 9fc18384 | Iustin Pop | opToResult $ applyMove nl2 inst2 o2 |
1028 | 9fc18384 | Iustin Pop | let ops3 = o2:ops2 |
1029 | 9fc18384 | Iustin Pop | -- and finally another replace secondary, to the final secondary |
1030 | 9fc18384 | Iustin Pop | (nl4, inst4, _, _) <- |
1031 | 9fc18384 | Iustin Pop | annotateResult "Changing secondary to final secondary" $ |
1032 | 9fc18384 | Iustin Pop | opToResult $ |
1033 | 9fc18384 | Iustin Pop | applyMove nl3 inst3 o3 |
1034 | 9fc18384 | Iustin Pop | let ops4 = o3:ops3 |
1035 | 9fc18384 | Iustin Pop | il' = Container.add idx inst4 il |
1036 | 9fc18384 | Iustin Pop | ops = concatMap (iMoveToJob nl4 il' idx) $ reverse ops4 |
1037 | 9fc18384 | Iustin Pop | let nodes = Container.elems nl4 |
1038 | 9fc18384 | Iustin Pop | -- The fromJust below is ugly (it can fail nastily), but |
1039 | 9fc18384 | Iustin Pop | -- at this point we should have any internal mismatches, |
1040 | 9fc18384 | Iustin Pop | -- and adding a monad here would be quite involved |
1041 | 9fc18384 | Iustin Pop | grpnodes = fromJust (gdx `lookup` Node.computeGroups nodes) |
1042 | 9fc18384 | Iustin Pop | new_cv = compCVNodes grpnodes |
1043 | 9fc18384 | Iustin Pop | return (nl4, il', ops, new_cv) |
1044 | 97da6b71 | Iustin Pop | |
1045 | c9a9b853 | Iustin Pop | -- | Computes the nodes in a given group which are available for |
1046 | c9a9b853 | Iustin Pop | -- allocation. |
1047 | c9a9b853 | Iustin Pop | availableGroupNodes :: [(Gdx, [Ndx])] -- ^ Group index/node index assoc list |
1048 | c9a9b853 | Iustin Pop | -> IntSet.IntSet -- ^ Nodes that are excluded |
1049 | c9a9b853 | Iustin Pop | -> Gdx -- ^ The group for which we |
1050 | c9a9b853 | Iustin Pop | -- query the nodes |
1051 | c9a9b853 | Iustin Pop | -> Result [Ndx] -- ^ List of available node indices |
1052 | c9a9b853 | Iustin Pop | availableGroupNodes group_nodes excl_ndx gdx = do |
1053 | 47eed3f4 | Iustin Pop | local_nodes <- maybe (Bad $ "Can't find group with index " ++ show gdx) |
1054 | 47eed3f4 | Iustin Pop | Ok (lookup gdx group_nodes) |
1055 | 47eed3f4 | Iustin Pop | let avail_nodes = filter (not . flip IntSet.member excl_ndx) local_nodes |
1056 | 47eed3f4 | Iustin Pop | return avail_nodes |
1057 | 47eed3f4 | Iustin Pop | |
1058 | 47eed3f4 | Iustin Pop | -- | Updates the evac solution with the results of an instance |
1059 | 47eed3f4 | Iustin Pop | -- evacuation. |
1060 | 47eed3f4 | Iustin Pop | updateEvacSolution :: (Node.List, Instance.List, EvacSolution) |
1061 | 5440c877 | Iustin Pop | -> Idx |
1062 | 47eed3f4 | Iustin Pop | -> Result (Node.List, Instance.List, [OpCodes.OpCode]) |
1063 | 47eed3f4 | Iustin Pop | -> (Node.List, Instance.List, EvacSolution) |
1064 | 5440c877 | Iustin Pop | updateEvacSolution (nl, il, es) idx (Bad msg) = |
1065 | 9fc18384 | Iustin Pop | (nl, il, es { esFailed = (idx, msg):esFailed es}) |
1066 | 5440c877 | Iustin Pop | updateEvacSolution (_, _, es) idx (Ok (nl, il, opcodes)) = |
1067 | 9fc18384 | Iustin Pop | (nl, il, es { esMoved = new_elem:esMoved es |
1068 | 9fc18384 | Iustin Pop | , esOpCodes = opcodes:esOpCodes es }) |
1069 | 9fc18384 | Iustin Pop | where inst = Container.find idx il |
1070 | 9fc18384 | Iustin Pop | new_elem = (idx, |
1071 | 9fc18384 | Iustin Pop | instancePriGroup nl inst, |
1072 | 9fc18384 | Iustin Pop | Instance.allNodes inst) |
1073 | 47eed3f4 | Iustin Pop | |
1074 | 47eed3f4 | Iustin Pop | -- | Node-evacuation IAllocator mode main function. |
1075 | 47eed3f4 | Iustin Pop | tryNodeEvac :: Group.List -- ^ The cluster groups |
1076 | 47eed3f4 | Iustin Pop | -> Node.List -- ^ The node list (cluster-wide, not per group) |
1077 | 47eed3f4 | Iustin Pop | -> Instance.List -- ^ Instance list (cluster-wide) |
1078 | 47eed3f4 | Iustin Pop | -> EvacMode -- ^ The evacuation mode |
1079 | 47eed3f4 | Iustin Pop | -> [Idx] -- ^ List of instance (indices) to be evacuated |
1080 | 4036f63a | Iustin Pop | -> Result (Node.List, Instance.List, EvacSolution) |
1081 | 47eed3f4 | Iustin Pop | tryNodeEvac _ ini_nl ini_il mode idxs = |
1082 | 9fc18384 | Iustin Pop | let evac_ndx = nodesToEvacuate ini_il mode idxs |
1083 | 9fc18384 | Iustin Pop | offline = map Node.idx . filter Node.offline $ Container.elems ini_nl |
1084 | 9fc18384 | Iustin Pop | excl_ndx = foldl' (flip IntSet.insert) evac_ndx offline |
1085 | 9fc18384 | Iustin Pop | group_ndx = map (\(gdx, (nl, _)) -> (gdx, map Node.idx |
1086 | 9fc18384 | Iustin Pop | (Container.elems nl))) $ |
1087 | 9fc18384 | Iustin Pop | splitCluster ini_nl ini_il |
1088 | 9fc18384 | Iustin Pop | (fin_nl, fin_il, esol) = |
1089 | 9fc18384 | Iustin Pop | foldl' (\state@(nl, il, _) inst -> |
1090 | 9fc18384 | Iustin Pop | let gdx = instancePriGroup nl inst |
1091 | 9fc18384 | Iustin Pop | pdx = Instance.pNode inst in |
1092 | 9fc18384 | Iustin Pop | updateEvacSolution state (Instance.idx inst) $ |
1093 | 9fc18384 | Iustin Pop | availableGroupNodes group_ndx |
1094 | 9fc18384 | Iustin Pop | (IntSet.insert pdx excl_ndx) gdx >>= |
1095 | 9fc18384 | Iustin Pop | nodeEvacInstance nl il mode inst gdx |
1096 | 9fc18384 | Iustin Pop | ) |
1097 | 9fc18384 | Iustin Pop | (ini_nl, ini_il, emptyEvacSolution) |
1098 | 9fc18384 | Iustin Pop | (map (`Container.find` ini_il) idxs) |
1099 | 9fc18384 | Iustin Pop | in return (fin_nl, fin_il, reverseEvacSolution esol) |
1100 | 47eed3f4 | Iustin Pop | |
1101 | 20b376ff | Iustin Pop | -- | Change-group IAllocator mode main function. |
1102 | 20b376ff | Iustin Pop | -- |
1103 | 20b376ff | Iustin Pop | -- This is very similar to 'tryNodeEvac', the only difference is that |
1104 | 20b376ff | Iustin Pop | -- we don't choose as target group the current instance group, but |
1105 | 20b376ff | Iustin Pop | -- instead: |
1106 | 20b376ff | Iustin Pop | -- |
1107 | 20b376ff | Iustin Pop | -- 1. at the start of the function, we compute which are the target |
1108 | 20b376ff | Iustin Pop | -- groups; either no groups were passed in, in which case we choose |
1109 | 20b376ff | Iustin Pop | -- all groups out of which we don't evacuate instance, or there were |
1110 | 20b376ff | Iustin Pop | -- some groups passed, in which case we use those |
1111 | 20b376ff | Iustin Pop | -- |
1112 | 20b376ff | Iustin Pop | -- 2. for each instance, we use 'findBestAllocGroup' to choose the |
1113 | 20b376ff | Iustin Pop | -- best group to hold the instance, and then we do what |
1114 | 20b376ff | Iustin Pop | -- 'tryNodeEvac' does, except for this group instead of the current |
1115 | 20b376ff | Iustin Pop | -- instance group. |
1116 | 20b376ff | Iustin Pop | -- |
1117 | 20b376ff | Iustin Pop | -- Note that the correct behaviour of this function relies on the |
1118 | 20b376ff | Iustin Pop | -- function 'nodeEvacInstance' to be able to do correctly both |
1119 | 20b376ff | Iustin Pop | -- intra-group and inter-group moves when passed the 'ChangeAll' mode. |
1120 | 20b376ff | Iustin Pop | tryChangeGroup :: Group.List -- ^ The cluster groups |
1121 | 20b376ff | Iustin Pop | -> Node.List -- ^ The node list (cluster-wide) |
1122 | 20b376ff | Iustin Pop | -> Instance.List -- ^ Instance list (cluster-wide) |
1123 | 20b376ff | Iustin Pop | -> [Gdx] -- ^ Target groups; if empty, any |
1124 | 20b376ff | Iustin Pop | -- groups not being evacuated |
1125 | 20b376ff | Iustin Pop | -> [Idx] -- ^ List of instance (indices) to be evacuated |
1126 | 4036f63a | Iustin Pop | -> Result (Node.List, Instance.List, EvacSolution) |
1127 | 20b376ff | Iustin Pop | tryChangeGroup gl ini_nl ini_il gdxs idxs = |
1128 | 9fc18384 | Iustin Pop | let evac_gdxs = nub $ map (instancePriGroup ini_nl . |
1129 | 9fc18384 | Iustin Pop | flip Container.find ini_il) idxs |
1130 | 9fc18384 | Iustin Pop | target_gdxs = (if null gdxs |
1131 | 20b376ff | Iustin Pop | then Container.keys gl |
1132 | 20b376ff | Iustin Pop | else gdxs) \\ evac_gdxs |
1133 | 9fc18384 | Iustin Pop | offline = map Node.idx . filter Node.offline $ Container.elems ini_nl |
1134 | 9fc18384 | Iustin Pop | excl_ndx = foldl' (flip IntSet.insert) IntSet.empty offline |
1135 | 9fc18384 | Iustin Pop | group_ndx = map (\(gdx, (nl, _)) -> (gdx, map Node.idx |
1136 | 9fc18384 | Iustin Pop | (Container.elems nl))) $ |
1137 | 9fc18384 | Iustin Pop | splitCluster ini_nl ini_il |
1138 | 9fc18384 | Iustin Pop | (fin_nl, fin_il, esol) = |
1139 | 9fc18384 | Iustin Pop | foldl' (\state@(nl, il, _) inst -> |
1140 | 9fc18384 | Iustin Pop | let solution = do |
1141 | 9fc18384 | Iustin Pop | let ncnt = Instance.requiredNodes $ |
1142 | 9fc18384 | Iustin Pop | Instance.diskTemplate inst |
1143 | 9fc18384 | Iustin Pop | (gdx, _, _) <- findBestAllocGroup gl nl il |
1144 | 9fc18384 | Iustin Pop | (Just target_gdxs) inst ncnt |
1145 | 9fc18384 | Iustin Pop | av_nodes <- availableGroupNodes group_ndx |
1146 | 9fc18384 | Iustin Pop | excl_ndx gdx |
1147 | 9fc18384 | Iustin Pop | nodeEvacInstance nl il ChangeAll inst gdx av_nodes |
1148 | 9fc18384 | Iustin Pop | in updateEvacSolution state (Instance.idx inst) solution |
1149 | 9fc18384 | Iustin Pop | ) |
1150 | 9fc18384 | Iustin Pop | (ini_nl, ini_il, emptyEvacSolution) |
1151 | 9fc18384 | Iustin Pop | (map (`Container.find` ini_il) idxs) |
1152 | 9fc18384 | Iustin Pop | in return (fin_nl, fin_il, reverseEvacSolution esol) |
1153 | 20b376ff | Iustin Pop | |
1154 | 7eda951b | Iustin Pop | -- | Standard-sized allocation method. |
1155 | 7eda951b | Iustin Pop | -- |
1156 | 7eda951b | Iustin Pop | -- This places instances of the same size on the cluster until we're |
1157 | 7eda951b | Iustin Pop | -- out of space. The result will be a list of identically-sized |
1158 | 7eda951b | Iustin Pop | -- instances. |
1159 | 7eda951b | Iustin Pop | iterateAlloc :: AllocMethod |
1160 | 8f48f67d | Iustin Pop | iterateAlloc nl il limit newinst allocnodes ixes cstats = |
1161 | 9fc18384 | Iustin Pop | let depth = length ixes |
1162 | 9fc18384 | Iustin Pop | newname = printf "new-%d" depth::String |
1163 | dce9bbb3 | Iustin Pop | newidx = Container.size il |
1164 | 9fc18384 | Iustin Pop | newi2 = Instance.setIdx (Instance.setName newinst newname) newidx |
1165 | 9fc18384 | Iustin Pop | newlimit = fmap (flip (-) 1) limit |
1166 | 9fc18384 | Iustin Pop | in case tryAlloc nl il newi2 allocnodes of |
1167 | 9fc18384 | Iustin Pop | Bad s -> Bad s |
1168 | 9fc18384 | Iustin Pop | Ok (AllocSolution { asFailures = errs, asSolution = sols3 }) -> |
1169 | 9fc18384 | Iustin Pop | let newsol = Ok (collapseFailures errs, nl, il, ixes, cstats) in |
1170 | 9fc18384 | Iustin Pop | case sols3 of |
1171 | 9fc18384 | Iustin Pop | Nothing -> newsol |
1172 | 9fc18384 | Iustin Pop | Just (xnl, xi, _, _) -> |
1173 | 9fc18384 | Iustin Pop | if limit == Just 0 |
1174 | 9fc18384 | Iustin Pop | then newsol |
1175 | 9fc18384 | Iustin Pop | else iterateAlloc xnl (Container.add newidx xi il) |
1176 | 9fc18384 | Iustin Pop | newlimit newinst allocnodes (xi:ixes) |
1177 | 9fc18384 | Iustin Pop | (totalResources xnl:cstats) |
1178 | 3ce8009a | Iustin Pop | |
1179 | 7eda951b | Iustin Pop | -- | Tiered allocation method. |
1180 | 7eda951b | Iustin Pop | -- |
1181 | 7eda951b | Iustin Pop | -- This places instances on the cluster, and decreases the spec until |
1182 | 7eda951b | Iustin Pop | -- we can allocate again. The result will be a list of decreasing |
1183 | 7eda951b | Iustin Pop | -- instance specs. |
1184 | 7eda951b | Iustin Pop | tieredAlloc :: AllocMethod |
1185 | 8f48f67d | Iustin Pop | tieredAlloc nl il limit newinst allocnodes ixes cstats = |
1186 | 9fc18384 | Iustin Pop | case iterateAlloc nl il limit newinst allocnodes ixes cstats of |
1187 | 9fc18384 | Iustin Pop | Bad s -> Bad s |
1188 | 9fc18384 | Iustin Pop | Ok (errs, nl', il', ixes', cstats') -> |
1189 | 9fc18384 | Iustin Pop | let newsol = Ok (errs, nl', il', ixes', cstats') |
1190 | 9fc18384 | Iustin Pop | ixes_cnt = length ixes' |
1191 | 9fc18384 | Iustin Pop | (stop, newlimit) = case limit of |
1192 | 9fc18384 | Iustin Pop | Nothing -> (False, Nothing) |
1193 | 9fc18384 | Iustin Pop | Just n -> (n <= ixes_cnt, |
1194 | 9fc18384 | Iustin Pop | Just (n - ixes_cnt)) in |
1195 | 9fc18384 | Iustin Pop | if stop then newsol else |
1196 | 3ce8009a | Iustin Pop | case Instance.shrinkByType newinst . fst . last $ |
1197 | 3ce8009a | Iustin Pop | sortBy (comparing snd) errs of |
1198 | 8f48f67d | Iustin Pop | Bad _ -> newsol |
1199 | 8f48f67d | Iustin Pop | Ok newinst' -> tieredAlloc nl' il' newlimit |
1200 | 8f48f67d | Iustin Pop | newinst' allocnodes ixes' cstats' |
1201 | 3ce8009a | Iustin Pop | |
1202 | 9188aeef | Iustin Pop | -- * Formatting functions |
1203 | e4f08c46 | Iustin Pop | |
1204 | e4f08c46 | Iustin Pop | -- | Given the original and final nodes, computes the relocation description. |
1205 | c9926b22 | Iustin Pop | computeMoves :: Instance.Instance -- ^ The instance to be moved |
1206 | c9926b22 | Iustin Pop | -> String -- ^ The instance name |
1207 | 668c03b3 | Iustin Pop | -> IMove -- ^ The move being performed |
1208 | e4f08c46 | Iustin Pop | -> String -- ^ New primary |
1209 | e4f08c46 | Iustin Pop | -> String -- ^ New secondary |
1210 | e4f08c46 | Iustin Pop | -> (String, [String]) |
1211 | e4f08c46 | Iustin Pop | -- ^ Tuple of moves and commands list; moves is containing |
1212 | e4f08c46 | Iustin Pop | -- either @/f/@ for failover or @/r:name/@ for replace |
1213 | e4f08c46 | Iustin Pop | -- secondary, while the command list holds gnt-instance |
1214 | e4f08c46 | Iustin Pop | -- commands (without that prefix), e.g \"@failover instance1@\" |
1215 | 668c03b3 | Iustin Pop | computeMoves i inam mv c d = |
1216 | 9fc18384 | Iustin Pop | case mv of |
1217 | 9fc18384 | Iustin Pop | Failover -> ("f", [mig]) |
1218 | 9fc18384 | Iustin Pop | FailoverAndReplace _ -> (printf "f r:%s" d, [mig, rep d]) |
1219 | 9fc18384 | Iustin Pop | ReplaceSecondary _ -> (printf "r:%s" d, [rep d]) |
1220 | 9fc18384 | Iustin Pop | ReplaceAndFailover _ -> (printf "r:%s f" c, [rep c, mig]) |
1221 | 9fc18384 | Iustin Pop | ReplacePrimary _ -> (printf "f r:%s f" c, [mig, rep c, mig]) |
1222 | 7959cbb9 | Iustin Pop | where morf = if Instance.isRunning i then "migrate" else "failover" |
1223 | 9fc18384 | Iustin Pop | mig = printf "%s -f %s" morf inam::String |
1224 | 9fc18384 | Iustin Pop | rep n = printf "replace-disks -n %s %s" n inam |
1225 | e4f08c46 | Iustin Pop | |
1226 | 9188aeef | Iustin Pop | -- | Converts a placement to string format. |
1227 | 9188aeef | Iustin Pop | printSolutionLine :: Node.List -- ^ The node list |
1228 | 9188aeef | Iustin Pop | -> Instance.List -- ^ The instance list |
1229 | 9188aeef | Iustin Pop | -> Int -- ^ Maximum node name length |
1230 | 9188aeef | Iustin Pop | -> Int -- ^ Maximum instance name length |
1231 | 9188aeef | Iustin Pop | -> Placement -- ^ The current placement |
1232 | 9188aeef | Iustin Pop | -> Int -- ^ The index of the placement in |
1233 | 9188aeef | Iustin Pop | -- the solution |
1234 | db1bcfe8 | Iustin Pop | -> (String, [String]) |
1235 | db1bcfe8 | Iustin Pop | printSolutionLine nl il nmlen imlen plc pos = |
1236 | 9fc18384 | Iustin Pop | let pmlen = (2*nmlen + 1) |
1237 | 9fc18384 | Iustin Pop | (i, p, s, mv, c) = plc |
1238 | e85444d0 | Iustin Pop | old_sec = Instance.sNode inst |
1239 | 9fc18384 | Iustin Pop | inst = Container.find i il |
1240 | 9fc18384 | Iustin Pop | inam = Instance.alias inst |
1241 | 9fc18384 | Iustin Pop | npri = Node.alias $ Container.find p nl |
1242 | 9fc18384 | Iustin Pop | nsec = Node.alias $ Container.find s nl |
1243 | 9fc18384 | Iustin Pop | opri = Node.alias $ Container.find (Instance.pNode inst) nl |
1244 | e85444d0 | Iustin Pop | osec = Node.alias $ Container.find old_sec nl |
1245 | 9fc18384 | Iustin Pop | (moves, cmds) = computeMoves inst inam mv npri nsec |
1246 | e85444d0 | Iustin Pop | -- FIXME: this should check instead/also the disk template |
1247 | e85444d0 | Iustin Pop | ostr = if old_sec == Node.noSecondary |
1248 | e85444d0 | Iustin Pop | then printf "%s" opri |
1249 | e85444d0 | Iustin Pop | else printf "%s:%s" opri osec |
1250 | e85444d0 | Iustin Pop | nstr = if s == Node.noSecondary |
1251 | e85444d0 | Iustin Pop | then printf "%s" npri |
1252 | e85444d0 | Iustin Pop | else printf "%s:%s" npri nsec |
1253 | 255d140d | Iustin Pop | in (printf " %3d. %-*s %-*s => %-*s %12.8f a=%s" |
1254 | e85444d0 | Iustin Pop | pos imlen inam pmlen (ostr::String) |
1255 | e85444d0 | Iustin Pop | pmlen (nstr::String) c moves, |
1256 | 9fc18384 | Iustin Pop | cmds) |
1257 | ca8258d9 | Iustin Pop | |
1258 | 0e8ae201 | Iustin Pop | -- | Return the instance and involved nodes in an instance move. |
1259 | 77ecfa82 | Iustin Pop | -- |
1260 | 77ecfa82 | Iustin Pop | -- Note that the output list length can vary, and is not required nor |
1261 | 77ecfa82 | Iustin Pop | -- guaranteed to be of any specific length. |
1262 | 77ecfa82 | Iustin Pop | involvedNodes :: Instance.List -- ^ Instance list, used for retrieving |
1263 | 77ecfa82 | Iustin Pop | -- the instance from its index; note |
1264 | 77ecfa82 | Iustin Pop | -- that this /must/ be the original |
1265 | 77ecfa82 | Iustin Pop | -- instance list, so that we can |
1266 | 77ecfa82 | Iustin Pop | -- retrieve the old nodes |
1267 | 77ecfa82 | Iustin Pop | -> Placement -- ^ The placement we're investigating, |
1268 | 77ecfa82 | Iustin Pop | -- containing the new nodes and |
1269 | 77ecfa82 | Iustin Pop | -- instance index |
1270 | 77ecfa82 | Iustin Pop | -> [Ndx] -- ^ Resulting list of node indices |
1271 | 0e8ae201 | Iustin Pop | involvedNodes il plc = |
1272 | 9fc18384 | Iustin Pop | let (i, np, ns, _, _) = plc |
1273 | 9fc18384 | Iustin Pop | inst = Container.find i il |
1274 | 9fc18384 | Iustin Pop | in nub $ [np, ns] ++ Instance.allNodes inst |
1275 | 0e8ae201 | Iustin Pop | |
1276 | 0e8ae201 | Iustin Pop | -- | Inner function for splitJobs, that either appends the next job to |
1277 | 0e8ae201 | Iustin Pop | -- the current jobset, or starts a new jobset. |
1278 | 0e8ae201 | Iustin Pop | mergeJobs :: ([JobSet], [Ndx]) -> MoveJob -> ([JobSet], [Ndx]) |
1279 | 924f9c16 | Iustin Pop | mergeJobs ([], _) n@(ndx, _, _, _) = ([[n]], ndx) |
1280 | 924f9c16 | Iustin Pop | mergeJobs (cjs@(j:js), nbuf) n@(ndx, _, _, _) |
1281 | 9fc18384 | Iustin Pop | | null (ndx `intersect` nbuf) = ((n:j):js, ndx ++ nbuf) |
1282 | 9fc18384 | Iustin Pop | | otherwise = ([n]:cjs, ndx) |
1283 | 0e8ae201 | Iustin Pop | |
1284 | 0e8ae201 | Iustin Pop | -- | Break a list of moves into independent groups. Note that this |
1285 | 0e8ae201 | Iustin Pop | -- will reverse the order of jobs. |
1286 | 0e8ae201 | Iustin Pop | splitJobs :: [MoveJob] -> [JobSet] |
1287 | 0e8ae201 | Iustin Pop | splitJobs = fst . foldl mergeJobs ([], []) |
1288 | 0e8ae201 | Iustin Pop | |
1289 | 0e8ae201 | Iustin Pop | -- | Given a list of commands, prefix them with @gnt-instance@ and |
1290 | 0e8ae201 | Iustin Pop | -- also beautify the display a little. |
1291 | 0e8ae201 | Iustin Pop | formatJob :: Int -> Int -> (Int, MoveJob) -> [String] |
1292 | 924f9c16 | Iustin Pop | formatJob jsn jsl (sn, (_, _, _, cmds)) = |
1293 | 9fc18384 | Iustin Pop | let out = |
1294 | 9fc18384 | Iustin Pop | printf " echo job %d/%d" jsn sn: |
1295 | 9fc18384 | Iustin Pop | printf " check": |
1296 | 9fc18384 | Iustin Pop | map (" gnt-instance " ++) cmds |
1297 | 9fc18384 | Iustin Pop | in if sn == 1 |
1298 | 0e8ae201 | Iustin Pop | then ["", printf "echo jobset %d, %d jobs" jsn jsl] ++ out |
1299 | 0e8ae201 | Iustin Pop | else out |
1300 | 0e8ae201 | Iustin Pop | |
1301 | 9188aeef | Iustin Pop | -- | Given a list of commands, prefix them with @gnt-instance@ and |
1302 | 9188aeef | Iustin Pop | -- also beautify the display a little. |
1303 | 0e8ae201 | Iustin Pop | formatCmds :: [JobSet] -> String |
1304 | 9f6dcdea | Iustin Pop | formatCmds = |
1305 | 9fc18384 | Iustin Pop | unlines . |
1306 | 9fc18384 | Iustin Pop | concatMap (\(jsn, js) -> concatMap (formatJob jsn (length js)) |
1307 | 9fc18384 | Iustin Pop | (zip [1..] js)) . |
1308 | 9fc18384 | Iustin Pop | zip [1..] |
1309 | 142538ff | Iustin Pop | |
1310 | e4f08c46 | Iustin Pop | -- | Print the node list. |
1311 | e98fb766 | Iustin Pop | printNodes :: Node.List -> [String] -> String |
1312 | e98fb766 | Iustin Pop | printNodes nl fs = |
1313 | 9fc18384 | Iustin Pop | let fields = case fs of |
1314 | 9fc18384 | Iustin Pop | [] -> Node.defaultFields |
1315 | 9fc18384 | Iustin Pop | "+":rest -> Node.defaultFields ++ rest |
1316 | 9fc18384 | Iustin Pop | _ -> fs |
1317 | 9fc18384 | Iustin Pop | snl = sortBy (comparing Node.idx) (Container.elems nl) |
1318 | 9fc18384 | Iustin Pop | (header, isnum) = unzip $ map Node.showHeader fields |
1319 | c3024b7e | René Nussbaumer | in printTable "" header (map (Node.list fields) snl) isnum |
1320 | e4f08c46 | Iustin Pop | |
1321 | 507fda3f | Iustin Pop | -- | Print the instance list. |
1322 | 507fda3f | Iustin Pop | printInsts :: Node.List -> Instance.List -> String |
1323 | 507fda3f | Iustin Pop | printInsts nl il = |
1324 | 9fc18384 | Iustin Pop | let sil = sortBy (comparing Instance.idx) (Container.elems il) |
1325 | 7959cbb9 | Iustin Pop | helper inst = [ if Instance.isRunning inst then "R" else " " |
1326 | 9fc18384 | Iustin Pop | , Instance.name inst |
1327 | 9fc18384 | Iustin Pop | , Container.nameOf nl (Instance.pNode inst) |
1328 | 9fc18384 | Iustin Pop | , let sdx = Instance.sNode inst |
1329 | 9fc18384 | Iustin Pop | in if sdx == Node.noSecondary |
1330 | 5182e970 | Iustin Pop | then "" |
1331 | 5182e970 | Iustin Pop | else Container.nameOf nl sdx |
1332 | 9fc18384 | Iustin Pop | , if Instance.autoBalance inst then "Y" else "N" |
1333 | 9fc18384 | Iustin Pop | , printf "%3d" $ Instance.vcpus inst |
1334 | 9fc18384 | Iustin Pop | , printf "%5d" $ Instance.mem inst |
1335 | 9fc18384 | Iustin Pop | , printf "%5d" $ Instance.dsk inst `div` 1024 |
1336 | 9fc18384 | Iustin Pop | , printf "%5.3f" lC |
1337 | 9fc18384 | Iustin Pop | , printf "%5.3f" lM |
1338 | 9fc18384 | Iustin Pop | , printf "%5.3f" lD |
1339 | 9fc18384 | Iustin Pop | , printf "%5.3f" lN |
1340 | 9fc18384 | Iustin Pop | ] |
1341 | 9fc18384 | Iustin Pop | where DynUtil lC lM lD lN = Instance.util inst |
1342 | 9fc18384 | Iustin Pop | header = [ "F", "Name", "Pri_node", "Sec_node", "Auto_bal" |
1343 | 9fc18384 | Iustin Pop | , "vcpu", "mem" , "dsk", "lCpu", "lMem", "lDsk", "lNet" ] |
1344 | 9fc18384 | Iustin Pop | isnum = False:False:False:False:False:repeat True |
1345 | c3024b7e | René Nussbaumer | in printTable "" header (map helper sil) isnum |
1346 | 507fda3f | Iustin Pop | |
1347 | 9188aeef | Iustin Pop | -- | Shows statistics for a given node list. |
1348 | 2922d2c5 | René Nussbaumer | printStats :: String -> Node.List -> String |
1349 | 2922d2c5 | René Nussbaumer | printStats lp nl = |
1350 | 9fc18384 | Iustin Pop | let dcvs = compDetailedCV $ Container.elems nl |
1351 | 9fc18384 | Iustin Pop | (weights, names) = unzip detailedCVInfo |
1352 | 9fc18384 | Iustin Pop | hd = zip3 (weights ++ repeat 1) (names ++ repeat "unknown") dcvs |
1353 | 2922d2c5 | René Nussbaumer | header = [ "Field", "Value", "Weight" ] |
1354 | 2922d2c5 | René Nussbaumer | formatted = map (\(w, h, val) -> |
1355 | 2922d2c5 | René Nussbaumer | [ h |
1356 | 2922d2c5 | René Nussbaumer | , printf "%.8f" val |
1357 | 2922d2c5 | René Nussbaumer | , printf "x%.2f" w |
1358 | 2922d2c5 | René Nussbaumer | ]) hd |
1359 | c3024b7e | René Nussbaumer | in printTable lp header formatted $ False:repeat True |
1360 | 6b20875c | Iustin Pop | |
1361 | 6b20875c | Iustin Pop | -- | Convert a placement into a list of OpCodes (basically a job). |
1362 | 179c0828 | Iustin Pop | iMoveToJob :: Node.List -- ^ The node list; only used for node |
1363 | 179c0828 | Iustin Pop | -- names, so any version is good |
1364 | 179c0828 | Iustin Pop | -- (before or after the operation) |
1365 | 179c0828 | Iustin Pop | -> Instance.List -- ^ The instance list; also used for |
1366 | 179c0828 | Iustin Pop | -- names only |
1367 | 179c0828 | Iustin Pop | -> Idx -- ^ The index of the instance being |
1368 | 179c0828 | Iustin Pop | -- moved |
1369 | 179c0828 | Iustin Pop | -> IMove -- ^ The actual move to be described |
1370 | 179c0828 | Iustin Pop | -> [OpCodes.OpCode] -- ^ The list of opcodes equivalent to |
1371 | 179c0828 | Iustin Pop | -- the given move |
1372 | 3e4480e0 | Iustin Pop | iMoveToJob nl il idx move = |
1373 | 9fc18384 | Iustin Pop | let inst = Container.find idx il |
1374 | 9fc18384 | Iustin Pop | iname = Instance.name inst |
1375 | 9fc18384 | Iustin Pop | lookNode = Just . Container.nameOf nl |
1376 | 9fc18384 | Iustin Pop | opF = OpCodes.OpInstanceMigrate iname True False True Nothing |
1377 | 9fc18384 | Iustin Pop | opR n = OpCodes.OpInstanceReplaceDisks iname (lookNode n) |
1378 | 9fc18384 | Iustin Pop | OpCodes.ReplaceNewSecondary [] Nothing |
1379 | 9fc18384 | Iustin Pop | in case move of |
1380 | 9fc18384 | Iustin Pop | Failover -> [ opF ] |
1381 | 9fc18384 | Iustin Pop | ReplacePrimary np -> [ opF, opR np, opF ] |
1382 | 9fc18384 | Iustin Pop | ReplaceSecondary ns -> [ opR ns ] |
1383 | 9fc18384 | Iustin Pop | ReplaceAndFailover np -> [ opR np, opF ] |
1384 | 9fc18384 | Iustin Pop | FailoverAndReplace ns -> [ opF, opR ns ] |
1385 | 32b8d9c0 | Iustin Pop | |
1386 | 949397c8 | Iustin Pop | -- * Node group functions |
1387 | 949397c8 | Iustin Pop | |
1388 | 525bfb36 | Iustin Pop | -- | Computes the group of an instance. |
1389 | 10ef6b4e | Iustin Pop | instanceGroup :: Node.List -> Instance.Instance -> Result Gdx |
1390 | 32b8d9c0 | Iustin Pop | instanceGroup nl i = |
1391 | 32b8d9c0 | Iustin Pop | let sidx = Instance.sNode i |
1392 | 32b8d9c0 | Iustin Pop | pnode = Container.find (Instance.pNode i) nl |
1393 | 32b8d9c0 | Iustin Pop | snode = if sidx == Node.noSecondary |
1394 | 32b8d9c0 | Iustin Pop | then pnode |
1395 | 32b8d9c0 | Iustin Pop | else Container.find sidx nl |
1396 | 10ef6b4e | Iustin Pop | pgroup = Node.group pnode |
1397 | 10ef6b4e | Iustin Pop | sgroup = Node.group snode |
1398 | 10ef6b4e | Iustin Pop | in if pgroup /= sgroup |
1399 | 9fc18384 | Iustin Pop | then fail ("Instance placed accross two node groups, primary " ++ |
1400 | 9fc18384 | Iustin Pop | show pgroup ++ ", secondary " ++ show sgroup) |
1401 | 9fc18384 | Iustin Pop | else return pgroup |
1402 | 32b8d9c0 | Iustin Pop | |
1403 | 525bfb36 | Iustin Pop | -- | Computes the group of an instance per the primary node. |
1404 | 4bc33d60 | Iustin Pop | instancePriGroup :: Node.List -> Instance.Instance -> Gdx |
1405 | 4bc33d60 | Iustin Pop | instancePriGroup nl i = |
1406 | 4bc33d60 | Iustin Pop | let pnode = Container.find (Instance.pNode i) nl |
1407 | 4bc33d60 | Iustin Pop | in Node.group pnode |
1408 | 4bc33d60 | Iustin Pop | |
1409 | 32b8d9c0 | Iustin Pop | -- | Compute the list of badly allocated instances (split across node |
1410 | 525bfb36 | Iustin Pop | -- groups). |
1411 | 32b8d9c0 | Iustin Pop | findSplitInstances :: Node.List -> Instance.List -> [Instance.Instance] |
1412 | 2a8e2dc9 | Iustin Pop | findSplitInstances nl = |
1413 | 2a8e2dc9 | Iustin Pop | filter (not . isOk . instanceGroup nl) . Container.elems |
1414 | f4161783 | Iustin Pop | |
1415 | 525bfb36 | Iustin Pop | -- | Splits a cluster into the component node groups. |
1416 | f4161783 | Iustin Pop | splitCluster :: Node.List -> Instance.List -> |
1417 | 10ef6b4e | Iustin Pop | [(Gdx, (Node.List, Instance.List))] |
1418 | f4161783 | Iustin Pop | splitCluster nl il = |
1419 | f4161783 | Iustin Pop | let ngroups = Node.computeGroups (Container.elems nl) |
1420 | f4161783 | Iustin Pop | in map (\(guuid, nodes) -> |
1421 | f4161783 | Iustin Pop | let nidxs = map Node.idx nodes |
1422 | f4161783 | Iustin Pop | nodes' = zip nidxs nodes |
1423 | f4161783 | Iustin Pop | instances = Container.filter ((`elem` nidxs) . Instance.pNode) il |
1424 | cb0c77ff | Iustin Pop | in (guuid, (Container.fromList nodes', instances))) ngroups |
1425 | 1f4ae205 | Iustin Pop | |
1426 | 63a78055 | Iustin Pop | -- | Compute the list of nodes that are to be evacuated, given a list |
1427 | 63a78055 | Iustin Pop | -- of instances and an evacuation mode. |
1428 | 63a78055 | Iustin Pop | nodesToEvacuate :: Instance.List -- ^ The cluster-wide instance list |
1429 | 63a78055 | Iustin Pop | -> EvacMode -- ^ The evacuation mode we're using |
1430 | 63a78055 | Iustin Pop | -> [Idx] -- ^ List of instance indices being evacuated |
1431 | 63a78055 | Iustin Pop | -> IntSet.IntSet -- ^ Set of node indices |
1432 | 63a78055 | Iustin Pop | nodesToEvacuate il mode = |
1433 | 9fc18384 | Iustin Pop | IntSet.delete Node.noSecondary . |
1434 | 9fc18384 | Iustin Pop | foldl' (\ns idx -> |
1435 | 9fc18384 | Iustin Pop | let i = Container.find idx il |
1436 | 9fc18384 | Iustin Pop | pdx = Instance.pNode i |
1437 | 9fc18384 | Iustin Pop | sdx = Instance.sNode i |
1438 | 9fc18384 | Iustin Pop | dt = Instance.diskTemplate i |
1439 | 9fc18384 | Iustin Pop | withSecondary = case dt of |
1440 | 9fc18384 | Iustin Pop | DTDrbd8 -> IntSet.insert sdx ns |
1441 | 9fc18384 | Iustin Pop | _ -> ns |
1442 | 9fc18384 | Iustin Pop | in case mode of |
1443 | 9fc18384 | Iustin Pop | ChangePrimary -> IntSet.insert pdx ns |
1444 | 9fc18384 | Iustin Pop | ChangeSecondary -> withSecondary |
1445 | 9fc18384 | Iustin Pop | ChangeAll -> IntSet.insert pdx withSecondary |
1446 | 9fc18384 | Iustin Pop | ) IntSet.empty |