root / htools / Ganeti / HTools / Program / Hcheck.hs @ 592601b3
History | View | Annotate | Download (9.7 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.Group as Group |
37 |
import qualified Ganeti.HTools.Node as Node |
38 |
import qualified Ganeti.HTools.Instance as Instance |
39 |
|
40 |
import qualified Ganeti.HTools.Program.Hbal as Hbal |
41 |
|
42 |
import Ganeti.HTools.CLI |
43 |
import Ganeti.HTools.ExtLoader |
44 |
import Ganeti.HTools.Loader |
45 |
import Ganeti.HTools.Types |
46 |
|
47 |
-- | Options list and functions. |
48 |
options :: [OptType] |
49 |
options = |
50 |
[ oDataFile |
51 |
, oDiskMoves |
52 |
, oDynuFile |
53 |
, oEvacMode |
54 |
, oExInst |
55 |
, oExTags |
56 |
, oIAllocSrc |
57 |
, oInstMoves |
58 |
, oLuxiSocket |
59 |
, oMachineReadable |
60 |
, oMaxCpu |
61 |
, oMaxSolLength |
62 |
, oMinDisk |
63 |
, oMinGain |
64 |
, oMinGainLim |
65 |
, oMinScore |
66 |
, oNoSimulation |
67 |
, oOfflineNode |
68 |
, oQuiet |
69 |
, oRapiMaster |
70 |
, oSelInst |
71 |
, oShowHelp |
72 |
, oShowVer |
73 |
, oVerbose |
74 |
] |
75 |
|
76 |
-- | Check phase - are we before (initial) or after rebalance. |
77 |
data Phase = Initial |
78 |
| Rebalanced |
79 |
|
80 |
-- | Level of presented statistics. |
81 |
data Level = GroupLvl |
82 |
| ClusterLvl |
83 |
|
84 |
-- | Prefix for machine readable names |
85 |
htcPrefix :: String |
86 |
htcPrefix = "HCHECK" |
87 |
|
88 |
-- | Phase-specific prefix for machine readable version. |
89 |
phasePrefix :: Phase -> String |
90 |
phasePrefix Initial = "INIT" |
91 |
phasePrefix Rebalanced = "FINAL" |
92 |
|
93 |
-- | Description of phases for human readable version. |
94 |
phaseDescription :: Phase -> String |
95 |
phaseDescription Initial = "initially" |
96 |
phaseDescription Rebalanced = "after rebalancing" |
97 |
|
98 |
-- | Level-specific prefix for machine readable version. |
99 |
levelPrefix :: Level -> String |
100 |
levelPrefix GroupLvl = "GROUP" |
101 |
levelPrefix ClusterLvl = "CLUSTER" |
102 |
|
103 |
-- | Data showed both per group and per cluster. |
104 |
commonData :: [(String, String)] |
105 |
commonData =[ ("N1_FAIL", "Nodes not N+1 happy") |
106 |
, ("CONFLICT_TAGS", "Nodes with conflicting instances") |
107 |
, ("OFFLINE_PRI", "Instances with primary on an offline node") |
108 |
, ("OFFLINE_SEC", "Instances with seondary on an offline node") |
109 |
] |
110 |
|
111 |
-- | Data showed per group. |
112 |
groupData :: [(String, String)] |
113 |
groupData = commonData ++ [("SCORE", "Group score")] |
114 |
|
115 |
-- | Data showed per cluster. |
116 |
clusterData :: [(String, String)] |
117 |
clusterData = commonData ++ |
118 |
[ ("NEED_REBALANCE", "Cluster is not healthy") |
119 |
, ("CAN_REBALANCE", "Possible to run rebalance") |
120 |
] |
121 |
|
122 |
|
123 |
-- | Format a list of key, value as a shell fragment. |
124 |
printKeysHTC :: [(String, String)] -> IO () |
125 |
printKeysHTC = printKeys htcPrefix |
126 |
|
127 |
-- | Prepare string from boolean value. |
128 |
printBool :: Bool -- ^ Whether the result should be machine readable |
129 |
-> Bool -- ^ Value to be converted to string |
130 |
-> String |
131 |
printBool True True = "1" |
132 |
printBool True False = "0" |
133 |
printBool False b = show b |
134 |
|
135 |
-- | Print mapping from group idx to group uuid (only in machine readable mode). |
136 |
printGroupsMappings :: Group.List -> IO () |
137 |
printGroupsMappings gl = do |
138 |
let extract_vals = \g -> (printf "GROUP_UUID_%d" $ Group.idx g :: String, |
139 |
printf "%s" $ Group.uuid g :: String) |
140 |
printpairs = map extract_vals (Container.elems gl) |
141 |
printKeysHTC printpairs |
142 |
|
143 |
-- | Print all the statistics on a group level. |
144 |
printGroupStats :: Int -> Bool -> Phase -> Group.Group -> [Int] -> Double -> IO () |
145 |
printGroupStats _ True phase grp stats score = do |
146 |
let printstats = map (printf "%d") stats ++ [printf "%.8f" score] :: [String] |
147 |
printkeys = map (printf "%s_%s_%d_%s" |
148 |
(phasePrefix phase) |
149 |
(levelPrefix GroupLvl) |
150 |
(Group.idx grp)) |
151 |
(map fst groupData) :: [String] |
152 |
printKeysHTC (zip printkeys printstats) |
153 |
|
154 |
printGroupStats verbose False phase grp stats score = do |
155 |
let printstats = map (printf "%d") stats ++ [printf "%.8f" score] :: [String] |
156 |
|
157 |
unless (verbose == 0) $ do |
158 |
printf "\nStatistics for group %s %s\n" |
159 |
(Group.name grp) (phaseDescription phase) :: IO () |
160 |
mapM_ (\(a,b) -> printf " %s: %s\n" (snd a) b :: IO ()) |
161 |
(zip groupData printstats) |
162 |
|
163 |
-- | Print all the statistics on a cluster (global) level. |
164 |
printClusterStats :: Int -> Bool -> Phase -> [Int] -> Bool -> IO (Bool) |
165 |
printClusterStats _ True phase stats canrebal = do |
166 |
let needrebal = sum stats > 0 |
167 |
printstats = map (printf "%d") stats ++ |
168 |
map (printBool True) [needrebal, canrebal] |
169 |
printkeys = map (printf "%s_%s_%s" |
170 |
(phasePrefix phase) |
171 |
(levelPrefix ClusterLvl)) |
172 |
(map fst clusterData) :: [String] |
173 |
printKeysHTC (zip printkeys printstats) |
174 |
return needrebal |
175 |
|
176 |
printClusterStats verbose False phase stats canrebal = do |
177 |
let needrebal = sum stats > 0 |
178 |
printstats = map (printf "%d") stats ++ |
179 |
map (printBool False) [needrebal, canrebal] |
180 |
unless (verbose == 0) $ do |
181 |
printf "\nCluster statistics %s\n" (phaseDescription phase) :: IO () |
182 |
mapM_ (\(a,b) -> printf " %s: %s\n" (snd a) b :: IO ()) |
183 |
(zip clusterData printstats) |
184 |
return needrebal |
185 |
|
186 |
{- | Check group for N+1 hapiness, conflicts of primaries on nodes and |
187 |
instances residing on offline nodes. |
188 |
|
189 |
-} |
190 |
perGroupChecks :: Int -> Bool -> Phase -> Group.List -> |
191 |
(Gdx, (Node.List, Instance.List)) -> IO ([Int]) |
192 |
perGroupChecks verbose machineread phase gl (gidx, (nl, il)) = do |
193 |
let grp = Container.find gidx gl |
194 |
offnl = filter Node.offline (Container.elems nl) |
195 |
n1violated = length $ fst $ Cluster.computeBadItems nl il |
196 |
conflicttags = length $ filter (>0) |
197 |
(map Node.conflictingPrimaries (Container.elems nl)) |
198 |
offline_pri = sum . map length $ map Node.pList offnl |
199 |
offline_sec = length $ map Node.sList offnl |
200 |
score = Cluster.compCV nl |
201 |
groupstats = [ n1violated |
202 |
, conflicttags |
203 |
, offline_pri |
204 |
, offline_sec |
205 |
] |
206 |
printGroupStats verbose machineread phase grp groupstats score |
207 |
return groupstats |
208 |
|
209 |
-- | Use Hbal's iterateDepth to simulate group rebalance. |
210 |
simulateRebalance :: Options -> |
211 |
(Gdx, (Node.List, Instance.List)) -> |
212 |
IO ( (Gdx, (Node.List, Instance.List)) ) |
213 |
simulateRebalance opts (gidx, (nl, il)) = do |
214 |
let ini_cv = Cluster.compCV nl |
215 |
ini_tbl = Cluster.Table nl il ini_cv [] |
216 |
min_cv = optMinScore opts |
217 |
|
218 |
|
219 |
if (ini_cv < min_cv) |
220 |
then return (gidx, (nl, il)) |
221 |
else do |
222 |
let imlen = maximum . map (length . Instance.alias) $ Container.elems il |
223 |
nmlen = maximum . map (length . Node.alias) $ Container.elems nl |
224 |
|
225 |
(fin_tbl, _) <- Hbal.iterateDepth False ini_tbl |
226 |
(optMaxLength opts) |
227 |
(optDiskMoves opts) |
228 |
(optInstMoves opts) |
229 |
nmlen imlen [] min_cv |
230 |
(optMinGainLim opts) (optMinGain opts) |
231 |
(optEvacMode opts) |
232 |
|
233 |
let (Cluster.Table fin_nl fin_il _ _) = fin_tbl |
234 |
return (gidx, (fin_nl, fin_il)) |
235 |
|
236 |
-- | Prints the final @OK@ marker in machine readable output. |
237 |
printFinalHTC :: Bool -> IO () |
238 |
printFinalHTC = printFinal htcPrefix |
239 |
|
240 |
-- | Main function. |
241 |
main :: Options -> [String] -> IO () |
242 |
main opts args = do |
243 |
unless (null args) $ do |
244 |
hPutStrLn stderr "Error: this program doesn't take any arguments." |
245 |
exitWith $ ExitFailure 1 |
246 |
|
247 |
let verbose = optVerbose opts |
248 |
machineread = optMachineReadable opts |
249 |
nosimulation = optNoSimulation opts |
250 |
|
251 |
(ClusterData gl fixed_nl ilf _ _) <- loadExternalData opts |
252 |
nlf <- setNodeStatus opts fixed_nl |
253 |
|
254 |
let splitinstances = Cluster.findSplitInstances nlf ilf |
255 |
splitcluster = Cluster.splitCluster nlf ilf |
256 |
|
257 |
when machineread $ printGroupsMappings gl |
258 |
|
259 |
groupsstats <- mapM (perGroupChecks verbose machineread Initial gl) splitcluster |
260 |
let clusterstats = map sum (transpose groupsstats) :: [Int] |
261 |
canrebalance = length splitinstances == 0 |
262 |
needrebalance <- printClusterStats verbose machineread Initial clusterstats canrebalance |
263 |
|
264 |
when nosimulation $ do |
265 |
unless (verbose == 0 || machineread) $ |
266 |
printf "Running in no-simulation mode. Exiting.\n" |
267 |
printFinalHTC machineread |
268 |
exitWith ExitSuccess |
269 |
|
270 |
when (length splitinstances > 0) $ do |
271 |
unless (verbose == 0 || machineread) $ |
272 |
printf "Split instances found, simulation of re-balancing not possible\n" |
273 |
exitWith $ ExitFailure 1 |
274 |
|
275 |
unless needrebalance $ do |
276 |
unless (verbose == 0 || machineread) $ |
277 |
printf "No need to rebalance cluster, no problems found. Exiting.\n" |
278 |
printFinalHTC machineread |
279 |
exitWith ExitSuccess |
280 |
|
281 |
rebalancedcluster <- mapM (simulateRebalance opts) splitcluster |
282 |
newgroupstats <- mapM (perGroupChecks verbose machineread Rebalanced gl) |
283 |
rebalancedcluster |
284 |
let newclusterstats = map sum (transpose newgroupstats) :: [Int] |
285 |
_ <- printClusterStats verbose machineread Rebalanced newclusterstats canrebalance |
286 |
|
287 |
printFinalHTC machineread |