root / htools / Ganeti / HTools / Program / Hcheck.hs @ 92eacdd8
History | View | Annotate | Download (9.2 kB)
1 |
{-| Cluster checker. |
---|---|
2 |
|
3 |
-} |
4 |
|
5 |
{- |
6 |
|
7 |
Copyright (C) 2012 Google Inc. |
8 |
|
9 |
This program is free software; you can redistribute it and/or modify |
10 |
it under the terms of the GNU General Public License as published by |
11 |
the Free Software Foundation; either version 2 of the License, or |
12 |
(at your option) any later version. |
13 |
|
14 |
This program is distributed in the hope that it will be useful, but |
15 |
WITHOUT ANY WARRANTY; without even the implied warranty of |
16 |
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
17 |
General Public License for more details. |
18 |
|
19 |
You should have received a copy of the GNU Gene52al Public License |
20 |
along with this program; if not, write to the Free Software |
21 |
Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA |
22 |
02110-1301, USA. |
23 |
|
24 |
-} |
25 |
|
26 |
module Ganeti.HTools.Program.Hcheck (main, options) where |
27 |
|
28 |
import Control.Monad |
29 |
import List (transpose) |
30 |
import System.Exit |
31 |
import System.IO |
32 |
import Text.Printf (printf) |
33 |
|
34 |
import qualified Ganeti.HTools.Container as Container |
35 |
import qualified Ganeti.HTools.Cluster as Cluster |
36 |
import qualified Ganeti.HTools.Node as Node |
37 |
import qualified Ganeti.HTools.Instance as Instance |
38 |
|
39 |
import qualified Ganeti.HTools.Program.Hbal as Hbal |
40 |
|
41 |
import Ganeti.HTools.CLI |
42 |
import Ganeti.HTools.ExtLoader |
43 |
import Ganeti.HTools.Loader |
44 |
import Ganeti.HTools.Types |
45 |
|
46 |
-- | Options list and functions. |
47 |
options :: [OptType] |
48 |
options = |
49 |
[ oDataFile |
50 |
, oDiskMoves |
51 |
, oDynuFile |
52 |
, oEvacMode |
53 |
, oExInst |
54 |
, oExTags |
55 |
, oIAllocSrc |
56 |
, oInstMoves |
57 |
, oLuxiSocket |
58 |
, oMachineReadable |
59 |
, oMaxCpu |
60 |
, oMaxSolLength |
61 |
, oMinDisk |
62 |
, oMinGain |
63 |
, oMinGainLim |
64 |
, oMinScore |
65 |
, oNoSimulation |
66 |
, oOfflineNode |
67 |
, oQuiet |
68 |
, oRapiMaster |
69 |
, oSelInst |
70 |
, oShowHelp |
71 |
, oShowVer |
72 |
, oVerbose |
73 |
] |
74 |
|
75 |
-- | Check phase - are we before (initial) or after rebalance. |
76 |
data Phase = Initial |
77 |
| Rebalanced |
78 |
|
79 |
-- | Level of presented statistics. |
80 |
data Level = GroupLvl |
81 |
| ClusterLvl |
82 |
|
83 |
-- | Prefix for machine readable names |
84 |
htcPrefix :: String |
85 |
htcPrefix = "HCHECK" |
86 |
|
87 |
-- | Phase-specific prefix for machine readable version. |
88 |
phasePrefix :: Phase -> String |
89 |
phasePrefix Initial = "INIT" |
90 |
phasePrefix Rebalanced = "FINAL" |
91 |
|
92 |
-- | Description of phases for human readable version. |
93 |
phaseDescription :: Phase -> String |
94 |
phaseDescription Initial = "initially" |
95 |
phaseDescription Rebalanced = "after rebalancing" |
96 |
|
97 |
-- | Level-specific prefix for machine readable version. |
98 |
levelPrefix :: Level -> String |
99 |
levelPrefix GroupLvl = "GROUP" |
100 |
levelPrefix ClusterLvl = "CLUSTER" |
101 |
|
102 |
-- | Data showed both per group and per cluster. |
103 |
commonData :: [(String, String)] |
104 |
commonData =[ ("N1_FAIL", "Nodes not N+1 happy") |
105 |
, ("CONFLICT_TAGS", "Nodes with conflicting instances") |
106 |
, ("OFFLINE_PRI", "Instances with primary on an offline node") |
107 |
, ("OFFLINE_SEC", "Instances with seondary on an offline node") |
108 |
] |
109 |
|
110 |
-- | Data showed per group. |
111 |
groupData :: [(String, String)] |
112 |
groupData = commonData ++ [("SCORE", "Group score")] |
113 |
|
114 |
-- | Data showed per cluster. |
115 |
clusterData :: [(String, String)] |
116 |
clusterData = commonData ++ |
117 |
[ ("NEED_REBALANCE", "Cluster is not healthy") |
118 |
, ("CAN_REBALANCE", "Possible to run rebalance") |
119 |
] |
120 |
|
121 |
|
122 |
-- | Format a list of key, value as a shell fragment. |
123 |
printKeysHTC :: [(String, String)] -> IO () |
124 |
printKeysHTC = printKeys htcPrefix |
125 |
|
126 |
-- | Prepare string from boolean value. |
127 |
printBool :: Bool -- ^ Whether the result should be machine readable |
128 |
-> Bool -- ^ Value to be converted to string |
129 |
-> String |
130 |
printBool True True = "1" |
131 |
printBool True False = "0" |
132 |
printBool False b = show b |
133 |
|
134 |
-- | Print all the statistics on a group level. |
135 |
printGroupStats :: Int -> Bool -> Phase -> Gdx -> [Int] -> Double -> IO () |
136 |
printGroupStats _ True phase gidx stats score = do |
137 |
let printstats = map (printf "%d") stats ++ [printf "%.8f" score] :: [String] |
138 |
printkeys = map (printf "%s_%s_%d_%s" |
139 |
(phasePrefix phase) |
140 |
(levelPrefix GroupLvl) |
141 |
gidx) |
142 |
(map fst groupData) :: [String] |
143 |
printKeysHTC (zip printkeys printstats) |
144 |
|
145 |
printGroupStats verbose False phase gidx stats score = do |
146 |
let printstats = map (printf "%d") stats ++ [printf "%.8f" score] :: [String] |
147 |
|
148 |
unless (verbose == 0) $ do |
149 |
printf "\nStatistics for group %d %s\n" |
150 |
gidx (phaseDescription phase) :: IO () |
151 |
mapM_ (\(a,b) -> printf " %s: %s\n" (snd a) b :: IO ()) |
152 |
(zip groupData printstats) |
153 |
|
154 |
-- | Print all the statistics on a cluster (global) level. |
155 |
printClusterStats :: Int -> Bool -> Phase -> [Int] -> Bool -> IO (Bool) |
156 |
printClusterStats _ True phase stats canrebal = do |
157 |
let needrebal = sum stats > 0 |
158 |
printstats = map (printf "%d") stats ++ |
159 |
map (printBool True) [needrebal, canrebal] |
160 |
printkeys = map (printf "%s_%s_%s" |
161 |
(phasePrefix phase) |
162 |
(levelPrefix ClusterLvl)) |
163 |
(map fst clusterData) :: [String] |
164 |
printKeysHTC (zip printkeys printstats) |
165 |
return needrebal |
166 |
|
167 |
printClusterStats verbose False phase stats canrebal = do |
168 |
let needrebal = sum stats > 0 |
169 |
printstats = map (printf "%d") stats ++ |
170 |
map (printBool False) [needrebal, canrebal] |
171 |
unless (verbose == 0) $ do |
172 |
printf "\nCluster statistics %s\n" (phaseDescription phase) :: IO () |
173 |
mapM_ (\(a,b) -> printf " %s: %s\n" (snd a) b :: IO ()) |
174 |
(zip clusterData printstats) |
175 |
return needrebal |
176 |
|
177 |
{- | Check group for N+1 hapiness, conflicts of primaries on nodes and |
178 |
instances residing on offline nodes. |
179 |
|
180 |
-} |
181 |
perGroupChecks :: Int -> Bool -> Phase -> (Gdx, (Node.List, Instance.List)) |
182 |
-> IO ([Int]) |
183 |
perGroupChecks verbose machineread phase (gidx, (nl, il)) = do |
184 |
let offnl = filter Node.offline (Container.elems nl) |
185 |
n1violated = length $ fst $ Cluster.computeBadItems nl il |
186 |
conflicttags = length $ filter (>0) |
187 |
(map Node.conflictingPrimaries (Container.elems nl)) |
188 |
offline_pri = sum . map length $ map Node.pList offnl |
189 |
offline_sec = length $ map Node.sList offnl |
190 |
score = Cluster.compCV nl |
191 |
groupstats = [ n1violated |
192 |
, conflicttags |
193 |
, offline_pri |
194 |
, offline_sec |
195 |
] |
196 |
printGroupStats verbose machineread phase gidx groupstats score |
197 |
return groupstats |
198 |
|
199 |
-- | Use Hbal's iterateDepth to simulate group rebalance. |
200 |
simulateRebalance :: Options -> |
201 |
(Gdx, (Node.List, Instance.List)) -> |
202 |
IO ( (Gdx, (Node.List, Instance.List)) ) |
203 |
simulateRebalance opts (gidx, (nl, il)) = do |
204 |
let ini_cv = Cluster.compCV nl |
205 |
ini_tbl = Cluster.Table nl il ini_cv [] |
206 |
min_cv = optMinScore opts |
207 |
|
208 |
|
209 |
if (ini_cv < min_cv) |
210 |
then return (gidx, (nl, il)) |
211 |
else do |
212 |
let imlen = maximum . map (length . Instance.alias) $ Container.elems il |
213 |
nmlen = maximum . map (length . Node.alias) $ Container.elems nl |
214 |
|
215 |
(fin_tbl, _) <- Hbal.iterateDepth False ini_tbl |
216 |
(optMaxLength opts) |
217 |
(optDiskMoves opts) |
218 |
(optInstMoves opts) |
219 |
nmlen imlen [] min_cv |
220 |
(optMinGainLim opts) (optMinGain opts) |
221 |
(optEvacMode opts) |
222 |
|
223 |
let (Cluster.Table fin_nl fin_il _ _) = fin_tbl |
224 |
return (gidx, (fin_nl, fin_il)) |
225 |
|
226 |
-- | Prints the final @OK@ marker in machine readable output. |
227 |
printFinalHTC :: Bool -> IO () |
228 |
printFinalHTC = printFinal htcPrefix |
229 |
|
230 |
-- | Main function. |
231 |
main :: Options -> [String] -> IO () |
232 |
main opts args = do |
233 |
unless (null args) $ do |
234 |
hPutStrLn stderr "Error: this program doesn't take any arguments." |
235 |
exitWith $ ExitFailure 1 |
236 |
|
237 |
let verbose = optVerbose opts |
238 |
machineread = optMachineReadable opts |
239 |
nosimulation = optNoSimulation opts |
240 |
|
241 |
(ClusterData _ fixed_nl ilf _ _) <- loadExternalData opts |
242 |
nlf <- setNodeStatus opts fixed_nl |
243 |
|
244 |
let splitinstances = Cluster.findSplitInstances nlf ilf |
245 |
splitcluster = Cluster.splitCluster nlf ilf |
246 |
|
247 |
groupsstats <- mapM (perGroupChecks verbose machineread Initial) splitcluster |
248 |
let clusterstats = map sum (transpose groupsstats) :: [Int] |
249 |
canrebalance = length splitinstances == 0 |
250 |
needrebalance <- printClusterStats verbose machineread Initial clusterstats canrebalance |
251 |
|
252 |
when nosimulation $ do |
253 |
unless (verbose == 0 || machineread) $ |
254 |
printf "Running in no-simulation mode. Exiting.\n" |
255 |
printFinalHTC machineread |
256 |
exitWith ExitSuccess |
257 |
|
258 |
when (length splitinstances > 0) $ do |
259 |
unless (verbose == 0 || machineread) $ |
260 |
printf "Split instances found, simulation of re-balancing not possible\n" |
261 |
exitWith $ ExitFailure 1 |
262 |
|
263 |
unless needrebalance $ do |
264 |
unless (verbose == 0 || machineread) $ |
265 |
printf "No need to rebalance cluster, no problems found. Exiting.\n" |
266 |
printFinalHTC machineread |
267 |
exitWith ExitSuccess |
268 |
|
269 |
rebalancedcluster <- mapM (simulateRebalance opts) splitcluster |
270 |
newgroupstats <- mapM (perGroupChecks verbose machineread Rebalanced) |
271 |
rebalancedcluster |
272 |
let newclusterstats = map sum (transpose newgroupstats) :: [Int] |
273 |
_ <- printClusterStats verbose machineread Rebalanced newclusterstats canrebalance |
274 |
|
275 |
printFinalHTC machineread |