root / Ganeti / HTools / Cluster.hs @ 262a08a2
History | View | Annotate | Download (25.4 kB)
1 |
{-| Implementation of cluster-wide logic. |
---|---|
2 |
|
3 |
This module holds all pure cluster-logic; I\/O related functionality |
4 |
goes into the "Main" module for the individual binaries. |
5 |
|
6 |
-} |
7 |
|
8 |
module Ganeti.HTools.Cluster |
9 |
( |
10 |
-- * Types |
11 |
NameList |
12 |
, Placement |
13 |
, Solution(..) |
14 |
, Table(..) |
15 |
, Removal |
16 |
, Score |
17 |
, IMove(..) |
18 |
-- * Generic functions |
19 |
, totalResources |
20 |
-- * First phase functions |
21 |
, computeBadItems |
22 |
-- * Second phase functions |
23 |
, computeSolution |
24 |
, applySolution |
25 |
, printSolution |
26 |
, printSolutionLine |
27 |
, formatCmds |
28 |
, printNodes |
29 |
-- * Balacing functions |
30 |
, applyMove |
31 |
, checkMove |
32 |
, compCV |
33 |
, printStats |
34 |
-- * IAllocator functions |
35 |
, allocateOnSingle |
36 |
, allocateOnPair |
37 |
) where |
38 |
|
39 |
import Data.List |
40 |
import Data.Maybe (isNothing, fromJust) |
41 |
import Text.Printf (printf) |
42 |
import Data.Function |
43 |
import Control.Monad |
44 |
|
45 |
import qualified Ganeti.HTools.Container as Container |
46 |
import qualified Ganeti.HTools.Instance as Instance |
47 |
import qualified Ganeti.HTools.Node as Node |
48 |
import Ganeti.HTools.Types |
49 |
import Ganeti.HTools.Utils |
50 |
|
51 |
-- | A separate name for the cluster score type |
52 |
type Score = Double |
53 |
|
54 |
-- | The description of an instance placement. |
55 |
type Placement = (Int, Int, Int, Score) |
56 |
|
57 |
{- | A cluster solution described as the solution delta and the list |
58 |
of placements. |
59 |
|
60 |
-} |
61 |
data Solution = Solution Int [Placement] |
62 |
deriving (Eq, Ord, Show) |
63 |
|
64 |
-- | Returns the delta of a solution or -1 for Nothing |
65 |
solutionDelta :: Maybe Solution -> Int |
66 |
solutionDelta sol = case sol of |
67 |
Just (Solution d _) -> d |
68 |
_ -> -1 |
69 |
|
70 |
-- | A removal set. |
71 |
data Removal = Removal Node.List [Instance.Instance] |
72 |
|
73 |
-- | An instance move definition |
74 |
data IMove = Failover -- ^ Failover the instance (f) |
75 |
| ReplacePrimary Int -- ^ Replace primary (f, r:np, f) |
76 |
| ReplaceSecondary Int -- ^ Replace secondary (r:ns) |
77 |
| ReplaceAndFailover Int -- ^ Replace secondary, failover (r:np, f) |
78 |
| FailoverAndReplace Int -- ^ Failover, replace secondary (f, r:ns) |
79 |
deriving (Show) |
80 |
|
81 |
-- | The complete state for the balancing solution |
82 |
data Table = Table Node.List Instance.List Score [Placement] |
83 |
deriving (Show) |
84 |
|
85 |
-- General functions |
86 |
|
87 |
-- | Cap the removal list if needed. |
88 |
capRemovals :: [a] -> Int -> [a] |
89 |
capRemovals removals max_removals = |
90 |
if max_removals > 0 then |
91 |
take max_removals removals |
92 |
else |
93 |
removals |
94 |
|
95 |
-- | Check if the given node list fails the N+1 check. |
96 |
verifyN1Check :: [Node.Node] -> Bool |
97 |
verifyN1Check nl = any Node.failN1 nl |
98 |
|
99 |
-- | Verifies the N+1 status and return the affected nodes. |
100 |
verifyN1 :: [Node.Node] -> [Node.Node] |
101 |
verifyN1 nl = filter Node.failN1 nl |
102 |
|
103 |
{-| Add an instance and return the new node and instance maps. -} |
104 |
addInstance :: Node.List -> Instance.Instance -> |
105 |
Node.Node -> Node.Node -> Maybe Node.List |
106 |
addInstance nl idata pri sec = |
107 |
let pdx = Node.idx pri |
108 |
sdx = Node.idx sec |
109 |
in do |
110 |
pnode <- Node.addPri pri idata |
111 |
snode <- Node.addSec sec idata pdx |
112 |
new_nl <- return $ Container.addTwo sdx snode |
113 |
pdx pnode nl |
114 |
return new_nl |
115 |
|
116 |
-- | Remove an instance and return the new node and instance maps. |
117 |
removeInstance :: Node.List -> Instance.Instance -> Node.List |
118 |
removeInstance nl idata = |
119 |
let pnode = Instance.pnode idata |
120 |
snode = Instance.snode idata |
121 |
pn = Container.find pnode nl |
122 |
sn = Container.find snode nl |
123 |
new_nl = Container.addTwo |
124 |
pnode (Node.removePri pn idata) |
125 |
snode (Node.removeSec sn idata) nl in |
126 |
new_nl |
127 |
|
128 |
-- | Remove an instance and return the new node map. |
129 |
removeInstances :: Node.List -> [Instance.Instance] -> Node.List |
130 |
removeInstances = foldl' removeInstance |
131 |
|
132 |
-- | Compute the total free disk and memory in the cluster. |
133 |
totalResources :: Container.Container Node.Node -> (Int, Int) |
134 |
totalResources nl = |
135 |
foldl' |
136 |
(\ (mem, dsk) node -> (mem + (Node.f_mem node), |
137 |
dsk + (Node.f_dsk node))) |
138 |
(0, 0) (Container.elems nl) |
139 |
|
140 |
{- | Compute a new version of a cluster given a solution. |
141 |
|
142 |
This is not used for computing the solutions, but for applying a |
143 |
(known-good) solution to the original cluster for final display. |
144 |
|
145 |
It first removes the relocated instances after which it places them on |
146 |
their new nodes. |
147 |
|
148 |
-} |
149 |
applySolution :: Node.List -> Instance.List -> [Placement] -> Node.List |
150 |
applySolution nl il sol = |
151 |
let odxes = map (\ (a, b, c, _) -> (Container.find a il, |
152 |
Node.idx (Container.find b nl), |
153 |
Node.idx (Container.find c nl)) |
154 |
) sol |
155 |
idxes = (\ (x, _, _) -> x) (unzip3 odxes) |
156 |
nc = removeInstances nl idxes |
157 |
in |
158 |
foldl' (\ nz (a, b, c) -> |
159 |
let new_p = Container.find b nz |
160 |
new_s = Container.find c nz in |
161 |
fromJust (addInstance nz a new_p new_s) |
162 |
) nc odxes |
163 |
|
164 |
|
165 |
-- First phase functions |
166 |
|
167 |
{- | Given a list 1,2,3..n build a list of pairs [(1, [2..n]), (2, |
168 |
[3..n]), ...] |
169 |
|
170 |
-} |
171 |
genParts :: [a] -> Int -> [(a, [a])] |
172 |
genParts l count = |
173 |
case l of |
174 |
[] -> [] |
175 |
x:xs -> |
176 |
if length l < count then |
177 |
[] |
178 |
else |
179 |
(x, xs) : (genParts xs count) |
180 |
|
181 |
-- | Generates combinations of count items from the names list. |
182 |
genNames :: Int -> [b] -> [[b]] |
183 |
genNames count1 names1 = |
184 |
let aux_fn count names current = |
185 |
case count of |
186 |
0 -> [current] |
187 |
_ -> |
188 |
concatMap |
189 |
(\ (x, xs) -> aux_fn (count - 1) xs (x:current)) |
190 |
(genParts names count) |
191 |
in |
192 |
aux_fn count1 names1 [] |
193 |
|
194 |
{- | Computes the pair of bad nodes and instances. |
195 |
|
196 |
The bad node list is computed via a simple 'verifyN1' check, and the |
197 |
bad instance list is the list of primary and secondary instances of |
198 |
those nodes. |
199 |
|
200 |
-} |
201 |
computeBadItems :: Node.List -> Instance.List -> |
202 |
([Node.Node], [Instance.Instance]) |
203 |
computeBadItems nl il = |
204 |
let bad_nodes = verifyN1 $ filter (not . Node.offline) $ Container.elems nl |
205 |
bad_instances = map (\idx -> Container.find idx il) $ |
206 |
sort $ nub $ concat $ |
207 |
map (\ n -> (Node.slist n) ++ (Node.plist n)) bad_nodes |
208 |
in |
209 |
(bad_nodes, bad_instances) |
210 |
|
211 |
|
212 |
{- | Checks if removal of instances results in N+1 pass. |
213 |
|
214 |
Note: the check removal cannot optimize by scanning only the affected |
215 |
nodes, since the cluster is known to be not healthy; only the check |
216 |
placement can make this shortcut. |
217 |
|
218 |
-} |
219 |
checkRemoval :: Node.List -> [Instance.Instance] -> Maybe Removal |
220 |
checkRemoval nl victims = |
221 |
let nx = removeInstances nl victims |
222 |
failN1 = verifyN1Check (Container.elems nx) |
223 |
in |
224 |
if failN1 then |
225 |
Nothing |
226 |
else |
227 |
Just $ Removal nx victims |
228 |
|
229 |
|
230 |
-- | Computes the removals list for a given depth |
231 |
computeRemovals :: Node.List |
232 |
-> [Instance.Instance] |
233 |
-> Int |
234 |
-> [Maybe Removal] |
235 |
computeRemovals nl bad_instances depth = |
236 |
map (checkRemoval nl) $ genNames depth bad_instances |
237 |
|
238 |
-- Second phase functions |
239 |
|
240 |
-- | Single-node relocation cost |
241 |
nodeDelta :: Int -> Int -> Int -> Int |
242 |
nodeDelta i p s = |
243 |
if i == p || i == s then |
244 |
0 |
245 |
else |
246 |
1 |
247 |
|
248 |
{-| Compute best solution. |
249 |
|
250 |
This function compares two solutions, choosing the minimum valid |
251 |
solution. |
252 |
-} |
253 |
compareSolutions :: Maybe Solution -> Maybe Solution -> Maybe Solution |
254 |
compareSolutions a b = case (a, b) of |
255 |
(Nothing, x) -> x |
256 |
(x, Nothing) -> x |
257 |
(x, y) -> min x y |
258 |
|
259 |
-- | Compute best table. Note that the ordering of the arguments is important. |
260 |
compareTables :: Table -> Table -> Table |
261 |
compareTables a@(Table _ _ a_cv _) b@(Table _ _ b_cv _ ) = |
262 |
if a_cv > b_cv then b else a |
263 |
|
264 |
-- | Check if a given delta is worse then an existing solution. |
265 |
tooHighDelta :: Maybe Solution -> Int -> Int -> Bool |
266 |
tooHighDelta sol new_delta max_delta = |
267 |
if new_delta > max_delta && max_delta >=0 then |
268 |
True |
269 |
else |
270 |
case sol of |
271 |
Nothing -> False |
272 |
Just (Solution old_delta _) -> old_delta <= new_delta |
273 |
|
274 |
{-| Check if placement of instances still keeps the cluster N+1 compliant. |
275 |
|
276 |
This is the workhorse of the allocation algorithm: given the |
277 |
current node and instance maps, the list of instances to be |
278 |
placed, and the current solution, this will return all possible |
279 |
solution by recursing until all target instances are placed. |
280 |
|
281 |
-} |
282 |
checkPlacement :: Node.List -- ^ The current node list |
283 |
-> [Instance.Instance] -- ^ List of instances still to place |
284 |
-> [Placement] -- ^ Partial solution until now |
285 |
-> Int -- ^ The delta of the partial solution |
286 |
-> Maybe Solution -- ^ The previous solution |
287 |
-> Int -- ^ Abort if the we go above this delta |
288 |
-> Maybe Solution -- ^ The new solution |
289 |
checkPlacement nl victims current current_delta prev_sol max_delta = |
290 |
let target = head victims |
291 |
opdx = Instance.pnode target |
292 |
osdx = Instance.snode target |
293 |
vtail = tail victims |
294 |
have_tail = (length vtail) > 0 |
295 |
nodes = Container.elems nl |
296 |
iidx = Instance.idx target |
297 |
in |
298 |
foldl' |
299 |
(\ accu_p pri -> |
300 |
let |
301 |
pri_idx = Node.idx pri |
302 |
upri_delta = current_delta + nodeDelta pri_idx opdx osdx |
303 |
new_pri = Node.addPri pri target |
304 |
fail_delta1 = tooHighDelta accu_p upri_delta max_delta |
305 |
in |
306 |
if fail_delta1 || isNothing(new_pri) then accu_p |
307 |
else let pri_nl = Container.add pri_idx (fromJust new_pri) nl in |
308 |
foldl' |
309 |
(\ accu sec -> |
310 |
let |
311 |
sec_idx = Node.idx sec |
312 |
upd_delta = upri_delta + |
313 |
nodeDelta sec_idx opdx osdx |
314 |
fail_delta2 = tooHighDelta accu upd_delta max_delta |
315 |
new_sec = Node.addSec sec target pri_idx |
316 |
in |
317 |
if sec_idx == pri_idx || fail_delta2 || |
318 |
isNothing new_sec then accu |
319 |
else let |
320 |
nx = Container.add sec_idx (fromJust new_sec) pri_nl |
321 |
upd_cv = compCV nx |
322 |
plc = (iidx, pri_idx, sec_idx, upd_cv) |
323 |
c2 = plc:current |
324 |
result = |
325 |
if have_tail then |
326 |
checkPlacement nx vtail c2 upd_delta |
327 |
accu max_delta |
328 |
else |
329 |
Just (Solution upd_delta c2) |
330 |
in compareSolutions accu result |
331 |
) accu_p nodes |
332 |
) prev_sol nodes |
333 |
|
334 |
-- | Apply a move |
335 |
applyMove :: Node.List -> Instance.Instance |
336 |
-> IMove -> (Maybe Node.List, Instance.Instance, Int, Int) |
337 |
-- Failover (f) |
338 |
applyMove nl inst Failover = |
339 |
let old_pdx = Instance.pnode inst |
340 |
old_sdx = Instance.snode inst |
341 |
old_p = Container.find old_pdx nl |
342 |
old_s = Container.find old_sdx nl |
343 |
int_p = Node.removePri old_p inst |
344 |
int_s = Node.removeSec old_s inst |
345 |
new_nl = do -- Maybe monad |
346 |
new_p <- Node.addPri int_s inst |
347 |
new_s <- Node.addSec int_p inst old_sdx |
348 |
return $ Container.addTwo old_pdx new_s old_sdx new_p nl |
349 |
in (new_nl, Instance.setBoth inst old_sdx old_pdx, old_sdx, old_pdx) |
350 |
|
351 |
-- Replace the primary (f:, r:np, f) |
352 |
applyMove nl inst (ReplacePrimary new_pdx) = |
353 |
let old_pdx = Instance.pnode inst |
354 |
old_sdx = Instance.snode inst |
355 |
old_p = Container.find old_pdx nl |
356 |
old_s = Container.find old_sdx nl |
357 |
tgt_n = Container.find new_pdx nl |
358 |
int_p = Node.removePri old_p inst |
359 |
int_s = Node.removeSec old_s inst |
360 |
new_nl = do -- Maybe monad |
361 |
new_p <- Node.addPri tgt_n inst |
362 |
new_s <- Node.addSec int_s inst new_pdx |
363 |
return $ Container.add new_pdx new_p $ |
364 |
Container.addTwo old_pdx int_p old_sdx new_s nl |
365 |
in (new_nl, Instance.setPri inst new_pdx, new_pdx, old_sdx) |
366 |
|
367 |
-- Replace the secondary (r:ns) |
368 |
applyMove nl inst (ReplaceSecondary new_sdx) = |
369 |
let old_pdx = Instance.pnode inst |
370 |
old_sdx = Instance.snode inst |
371 |
old_s = Container.find old_sdx nl |
372 |
tgt_n = Container.find new_sdx nl |
373 |
int_s = Node.removeSec old_s inst |
374 |
new_nl = Node.addSec tgt_n inst old_pdx >>= |
375 |
\new_s -> return $ Container.addTwo new_sdx |
376 |
new_s old_sdx int_s nl |
377 |
in (new_nl, Instance.setSec inst new_sdx, old_pdx, new_sdx) |
378 |
|
379 |
-- Replace the secondary and failover (r:np, f) |
380 |
applyMove nl inst (ReplaceAndFailover new_pdx) = |
381 |
let old_pdx = Instance.pnode inst |
382 |
old_sdx = Instance.snode inst |
383 |
old_p = Container.find old_pdx nl |
384 |
old_s = Container.find old_sdx nl |
385 |
tgt_n = Container.find new_pdx nl |
386 |
int_p = Node.removePri old_p inst |
387 |
int_s = Node.removeSec old_s inst |
388 |
new_nl = do -- Maybe monad |
389 |
new_p <- Node.addPri tgt_n inst |
390 |
new_s <- Node.addSec int_p inst new_pdx |
391 |
return $ Container.add new_pdx new_p $ |
392 |
Container.addTwo old_pdx new_s old_sdx int_s nl |
393 |
in (new_nl, Instance.setBoth inst new_pdx old_pdx, new_pdx, old_pdx) |
394 |
|
395 |
-- Failver and replace the secondary (f, r:ns) |
396 |
applyMove nl inst (FailoverAndReplace new_sdx) = |
397 |
let old_pdx = Instance.pnode inst |
398 |
old_sdx = Instance.snode inst |
399 |
old_p = Container.find old_pdx nl |
400 |
old_s = Container.find old_sdx nl |
401 |
tgt_n = Container.find new_sdx nl |
402 |
int_p = Node.removePri old_p inst |
403 |
int_s = Node.removeSec old_s inst |
404 |
new_nl = do -- Maybe monad |
405 |
new_p <- Node.addPri int_s inst |
406 |
new_s <- Node.addSec tgt_n inst old_sdx |
407 |
return $ Container.add new_sdx new_s $ |
408 |
Container.addTwo old_sdx new_p old_pdx int_p nl |
409 |
in (new_nl, Instance.setBoth inst old_sdx new_sdx, old_sdx, new_sdx) |
410 |
|
411 |
allocateOnSingle :: Node.List -> Instance.Instance -> Node.Node |
412 |
-> (Maybe Node.List, Instance.Instance) |
413 |
allocateOnSingle nl inst p = |
414 |
let new_pdx = Node.idx p |
415 |
new_nl = Node.addPri p inst >>= \new_p -> |
416 |
return $ Container.add new_pdx new_p nl |
417 |
in (new_nl, Instance.setBoth inst new_pdx Node.noSecondary) |
418 |
|
419 |
allocateOnPair :: Node.List -> Instance.Instance -> Node.Node -> Node.Node |
420 |
-> (Maybe Node.List, Instance.Instance) |
421 |
allocateOnPair nl inst tgt_p tgt_s = |
422 |
let new_pdx = Node.idx tgt_p |
423 |
new_sdx = Node.idx tgt_s |
424 |
new_nl = do -- Maybe monad |
425 |
new_p <- Node.addPri tgt_p inst |
426 |
new_s <- Node.addSec tgt_s inst new_pdx |
427 |
return $ Container.addTwo new_pdx new_p new_sdx new_s nl |
428 |
in (new_nl, Instance.setBoth inst new_pdx new_sdx) |
429 |
|
430 |
checkSingleStep :: Table -- ^ The original table |
431 |
-> Instance.Instance -- ^ The instance to move |
432 |
-> Table -- ^ The current best table |
433 |
-> IMove -- ^ The move to apply |
434 |
-> Table -- ^ The final best table |
435 |
checkSingleStep ini_tbl target cur_tbl move = |
436 |
let |
437 |
Table ini_nl ini_il _ ini_plc = ini_tbl |
438 |
(tmp_nl, new_inst, pri_idx, sec_idx) = applyMove ini_nl target move |
439 |
in |
440 |
if isNothing tmp_nl then cur_tbl |
441 |
else |
442 |
let tgt_idx = Instance.idx target |
443 |
upd_nl = fromJust tmp_nl |
444 |
upd_cvar = compCV upd_nl |
445 |
upd_il = Container.add tgt_idx new_inst ini_il |
446 |
upd_plc = (tgt_idx, pri_idx, sec_idx, upd_cvar):ini_plc |
447 |
upd_tbl = Table upd_nl upd_il upd_cvar upd_plc |
448 |
in |
449 |
compareTables cur_tbl upd_tbl |
450 |
|
451 |
-- | Given the status of the current secondary as a valid new node |
452 |
-- and the current candidate target node, |
453 |
-- generate the possible moves for a instance. |
454 |
possibleMoves :: Bool -> Int -> [IMove] |
455 |
possibleMoves True tdx = |
456 |
[ReplaceSecondary tdx, |
457 |
ReplaceAndFailover tdx, |
458 |
ReplacePrimary tdx, |
459 |
FailoverAndReplace tdx] |
460 |
|
461 |
possibleMoves False tdx = |
462 |
[ReplaceSecondary tdx, |
463 |
ReplaceAndFailover tdx] |
464 |
|
465 |
-- | Compute the best move for a given instance. |
466 |
checkInstanceMove :: [Int] -- Allowed target node indices |
467 |
-> Table -- Original table |
468 |
-> Instance.Instance -- Instance to move |
469 |
-> Table -- Best new table for this instance |
470 |
checkInstanceMove nodes_idx ini_tbl target = |
471 |
let |
472 |
opdx = Instance.pnode target |
473 |
osdx = Instance.snode target |
474 |
nodes = filter (\idx -> idx /= opdx && idx /= osdx) nodes_idx |
475 |
use_secondary = elem osdx nodes_idx |
476 |
aft_failover = if use_secondary -- if allowed to failover |
477 |
then checkSingleStep ini_tbl target ini_tbl Failover |
478 |
else ini_tbl |
479 |
all_moves = concatMap (possibleMoves use_secondary) nodes |
480 |
in |
481 |
-- iterate over the possible nodes for this instance |
482 |
foldl' (checkSingleStep ini_tbl target) aft_failover all_moves |
483 |
|
484 |
-- | Compute the best next move. |
485 |
checkMove :: [Int] -- ^ Allowed target node indices |
486 |
-> Table -- ^ The current solution |
487 |
-> [Instance.Instance] -- ^ List of instances still to move |
488 |
-> Table -- ^ The new solution |
489 |
checkMove nodes_idx ini_tbl victims = |
490 |
let Table _ _ _ ini_plc = ini_tbl |
491 |
-- iterate over all instances, computing the best move |
492 |
best_tbl = |
493 |
foldl' |
494 |
(\ step_tbl elem -> |
495 |
if Instance.snode elem == Node.noSecondary then step_tbl |
496 |
else compareTables step_tbl $ |
497 |
checkInstanceMove nodes_idx ini_tbl elem) |
498 |
ini_tbl victims |
499 |
Table _ _ _ best_plc = best_tbl |
500 |
in |
501 |
if length best_plc == length ini_plc then -- no advancement |
502 |
ini_tbl |
503 |
else |
504 |
best_tbl |
505 |
|
506 |
{- | Auxiliary function for solution computation. |
507 |
|
508 |
We write this in an explicit recursive fashion in order to control |
509 |
early-abort in case we have met the min delta. We can't use foldr |
510 |
instead of explicit recursion since we need the accumulator for the |
511 |
abort decision. |
512 |
|
513 |
-} |
514 |
advanceSolution :: [Maybe Removal] -- ^ The removal to process |
515 |
-> Int -- ^ Minimum delta parameter |
516 |
-> Int -- ^ Maximum delta parameter |
517 |
-> Maybe Solution -- ^ Current best solution |
518 |
-> Maybe Solution -- ^ New best solution |
519 |
advanceSolution [] _ _ sol = sol |
520 |
advanceSolution (Nothing:xs) m n sol = advanceSolution xs m n sol |
521 |
advanceSolution ((Just (Removal nx removed)):xs) min_d max_d prev_sol = |
522 |
let new_sol = checkPlacement nx removed [] 0 prev_sol max_d |
523 |
new_delta = solutionDelta $! new_sol |
524 |
in |
525 |
if new_delta >= 0 && new_delta <= min_d then |
526 |
new_sol |
527 |
else |
528 |
advanceSolution xs min_d max_d new_sol |
529 |
|
530 |
-- | Computes the placement solution. |
531 |
solutionFromRemovals :: [Maybe Removal] -- ^ The list of (possible) removals |
532 |
-> Int -- ^ Minimum delta parameter |
533 |
-> Int -- ^ Maximum delta parameter |
534 |
-> Maybe Solution -- ^ The best solution found |
535 |
solutionFromRemovals removals min_delta max_delta = |
536 |
advanceSolution removals min_delta max_delta Nothing |
537 |
|
538 |
{- | Computes the solution at the given depth. |
539 |
|
540 |
This is a wrapper over both computeRemovals and |
541 |
solutionFromRemovals. In case we have no solution, we return Nothing. |
542 |
|
543 |
-} |
544 |
computeSolution :: Node.List -- ^ The original node data |
545 |
-> [Instance.Instance] -- ^ The list of /bad/ instances |
546 |
-> Int -- ^ The /depth/ of removals |
547 |
-> Int -- ^ Maximum number of removals to process |
548 |
-> Int -- ^ Minimum delta parameter |
549 |
-> Int -- ^ Maximum delta parameter |
550 |
-> Maybe Solution -- ^ The best solution found (or Nothing) |
551 |
computeSolution nl bad_instances depth max_removals min_delta max_delta = |
552 |
let |
553 |
removals = computeRemovals nl bad_instances depth |
554 |
removals' = capRemovals removals max_removals |
555 |
in |
556 |
solutionFromRemovals removals' min_delta max_delta |
557 |
|
558 |
-- Solution display functions (pure) |
559 |
|
560 |
-- | Given the original and final nodes, computes the relocation description. |
561 |
computeMoves :: String -- ^ The instance name |
562 |
-> String -- ^ Original primary |
563 |
-> String -- ^ Original secondary |
564 |
-> String -- ^ New primary |
565 |
-> String -- ^ New secondary |
566 |
-> (String, [String]) |
567 |
-- ^ Tuple of moves and commands list; moves is containing |
568 |
-- either @/f/@ for failover or @/r:name/@ for replace |
569 |
-- secondary, while the command list holds gnt-instance |
570 |
-- commands (without that prefix), e.g \"@failover instance1@\" |
571 |
computeMoves i a b c d = |
572 |
if c == a then {- Same primary -} |
573 |
if d == b then {- Same sec??! -} |
574 |
("-", []) |
575 |
else {- Change of secondary -} |
576 |
(printf "r:%s" d, |
577 |
[printf "replace-disks -n %s %s" d i]) |
578 |
else |
579 |
if c == b then {- Failover and ... -} |
580 |
if d == a then {- that's all -} |
581 |
("f", [printf "migrate -f %s" i]) |
582 |
else |
583 |
(printf "f r:%s" d, |
584 |
[printf "migrate -f %s" i, |
585 |
printf "replace-disks -n %s %s" d i]) |
586 |
else |
587 |
if d == a then {- ... and keep primary as secondary -} |
588 |
(printf "r:%s f" c, |
589 |
[printf "replace-disks -n %s %s" c i, |
590 |
printf "migrate -f %s" i]) |
591 |
else |
592 |
if d == b then {- ... keep same secondary -} |
593 |
(printf "f r:%s f" c, |
594 |
[printf "migrate -f %s" i, |
595 |
printf "replace-disks -n %s %s" c i, |
596 |
printf "migrate -f %s" i]) |
597 |
|
598 |
else {- Nothing in common -} |
599 |
(printf "r:%s f r:%s" c d, |
600 |
[printf "replace-disks -n %s %s" c i, |
601 |
printf "migrate -f %s" i, |
602 |
printf "replace-disks -n %s %s" d i]) |
603 |
|
604 |
{-| Converts a placement to string format -} |
605 |
printSolutionLine :: Node.List |
606 |
-> Instance.List |
607 |
-> Int |
608 |
-> Int |
609 |
-> Placement |
610 |
-> Int |
611 |
-> (String, [String]) |
612 |
printSolutionLine nl il nmlen imlen plc pos = |
613 |
let |
614 |
pmlen = (2*nmlen + 1) |
615 |
(i, p, s, c) = plc |
616 |
inst = Container.find i il |
617 |
inam = Instance.name inst |
618 |
npri = Container.nameOf nl p |
619 |
nsec = Container.nameOf nl s |
620 |
opri = Container.nameOf nl $ Instance.pnode inst |
621 |
osec = Container.nameOf nl $ Instance.snode inst |
622 |
(moves, cmds) = computeMoves inam opri osec npri nsec |
623 |
ostr = (printf "%s:%s" opri osec)::String |
624 |
nstr = (printf "%s:%s" npri nsec)::String |
625 |
in |
626 |
(printf " %3d. %-*s %-*s => %-*s %.8f a=%s" |
627 |
pos imlen inam pmlen ostr |
628 |
pmlen nstr c moves, |
629 |
cmds) |
630 |
|
631 |
formatCmds :: [[String]] -> String |
632 |
formatCmds cmd_strs = |
633 |
unlines $ |
634 |
concat $ map (\(a, b) -> |
635 |
(printf "echo step %d" (a::Int)): |
636 |
(printf "check"): |
637 |
(map ("gnt-instance " ++) b)) $ |
638 |
zip [1..] cmd_strs |
639 |
|
640 |
{-| Converts a solution to string format -} |
641 |
printSolution :: Node.List |
642 |
-> Instance.List |
643 |
-> [Placement] |
644 |
-> ([String], [[String]]) |
645 |
printSolution nl il sol = |
646 |
let |
647 |
nmlen = Container.maxNameLen nl |
648 |
imlen = Container.maxNameLen il |
649 |
in |
650 |
unzip $ map (uncurry $ printSolutionLine nl il nmlen imlen) $ |
651 |
zip sol [1..] |
652 |
|
653 |
-- | Print the node list. |
654 |
printNodes :: Node.List -> String |
655 |
printNodes nl = |
656 |
let snl = sortBy (compare `on` Node.idx) (Container.elems nl) |
657 |
m_name = maximum . map (length . Node.name) $ snl |
658 |
helper = Node.list m_name |
659 |
header = printf |
660 |
"%2s %-*s %5s %5s %5s %5s %5s %5s %5s %5s %3s %3s %7s %7s" |
661 |
" F" m_name "Name" |
662 |
"t_mem" "n_mem" "i_mem" "x_mem" "f_mem" "r_mem" |
663 |
"t_dsk" "f_dsk" |
664 |
"pri" "sec" "p_fmem" "p_fdsk" |
665 |
in unlines $ (header:map helper snl) |
666 |
|
667 |
-- | Compute the mem and disk covariance. |
668 |
compDetailedCV :: Node.List -> (Double, Double, Double, Double, Double) |
669 |
compDetailedCV nl = |
670 |
let |
671 |
all_nodes = Container.elems nl |
672 |
(offline, nodes) = partition Node.offline all_nodes |
673 |
mem_l = map Node.p_mem nodes |
674 |
dsk_l = map Node.p_dsk nodes |
675 |
mem_cv = varianceCoeff mem_l |
676 |
dsk_cv = varianceCoeff dsk_l |
677 |
n1_l = length $ filter Node.failN1 nodes |
678 |
n1_score = (fromIntegral n1_l) / (fromIntegral $ length nodes) |
679 |
res_l = map Node.p_rem nodes |
680 |
res_cv = varianceCoeff res_l |
681 |
offline_inst = sum . map (\n -> (length . Node.plist $ n) + |
682 |
(length . Node.slist $ n)) $ offline |
683 |
online_inst = sum . map (\n -> (length . Node.plist $ n) + |
684 |
(length . Node.slist $ n)) $ nodes |
685 |
off_score = (fromIntegral offline_inst) / |
686 |
(fromIntegral $ online_inst + offline_inst) |
687 |
in (mem_cv, dsk_cv, n1_score, res_cv, off_score) |
688 |
|
689 |
-- | Compute the 'total' variance. |
690 |
compCV :: Node.List -> Double |
691 |
compCV nl = |
692 |
let (mem_cv, dsk_cv, n1_score, res_cv, off_score) = compDetailedCV nl |
693 |
in mem_cv + dsk_cv + n1_score + res_cv + off_score |
694 |
|
695 |
printStats :: Node.List -> String |
696 |
printStats nl = |
697 |
let (mem_cv, dsk_cv, n1_score, res_cv, off_score) = compDetailedCV nl |
698 |
in printf "f_mem=%.8f, r_mem=%.8f, f_dsk=%.8f, n1=%.3f, uf=%.3f" |
699 |
mem_cv res_cv dsk_cv n1_score off_score |