root / Ganeti / HTools / Cluster.hs @ 5d1baf63
History | View | Annotate | Download (29.7 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 |
NodeList |
12 |
, InstanceList |
13 |
, NameList |
14 |
, Placement |
15 |
, Solution(..) |
16 |
, Table(..) |
17 |
, Removal |
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 |
, checkMove |
31 |
, compCV |
32 |
, printStats |
33 |
-- * Loading functions |
34 |
, loadData |
35 |
, checkData |
36 |
) where |
37 |
|
38 |
import Data.List |
39 |
import Data.Maybe (isNothing, fromJust) |
40 |
import Text.Printf (printf) |
41 |
import Data.Function |
42 |
|
43 |
import qualified Ganeti.HTools.Container as Container |
44 |
import qualified Ganeti.HTools.Instance as Instance |
45 |
import qualified Ganeti.HTools.Node as Node |
46 |
import Ganeti.HTools.Utils |
47 |
|
48 |
type NodeList = Container.Container Node.Node |
49 |
type InstanceList = Container.Container Instance.Instance |
50 |
-- | The type used to hold idx-to-name mappings |
51 |
type NameList = [(Int, String)] |
52 |
-- | A separate name for the cluster score type |
53 |
type Score = Double |
54 |
|
55 |
-- | The description of an instance placement. |
56 |
type Placement = (Int, Int, Int, Score) |
57 |
|
58 |
{- | A cluster solution described as the solution delta and the list |
59 |
of placements. |
60 |
|
61 |
-} |
62 |
data Solution = Solution Int [Placement] |
63 |
deriving (Eq, Ord, Show) |
64 |
|
65 |
-- | Returns the delta of a solution or -1 for Nothing |
66 |
solutionDelta :: Maybe Solution -> Int |
67 |
solutionDelta sol = case sol of |
68 |
Just (Solution d _) -> d |
69 |
_ -> -1 |
70 |
|
71 |
-- | A removal set. |
72 |
data Removal = Removal NodeList [Instance.Instance] |
73 |
|
74 |
-- | An instance move definition |
75 |
data IMove = Failover -- ^ Failover the instance (f) |
76 |
| ReplacePrimary Int -- ^ Replace primary (f, r:np, f) |
77 |
| ReplaceSecondary Int -- ^ Replace secondary (r:ns) |
78 |
| ReplaceAndFailover Int -- ^ Replace secondary, failover (r:np, f) |
79 |
| FailoverAndReplace Int -- ^ Failover, replace secondary (f, r:ns) |
80 |
deriving (Show) |
81 |
|
82 |
-- | The complete state for the balancing solution |
83 |
data Table = Table NodeList InstanceList Score [Placement] |
84 |
deriving (Show) |
85 |
|
86 |
-- General functions |
87 |
|
88 |
-- | Cap the removal list if needed. |
89 |
capRemovals :: [a] -> Int -> [a] |
90 |
capRemovals removals max_removals = |
91 |
if max_removals > 0 then |
92 |
take max_removals removals |
93 |
else |
94 |
removals |
95 |
|
96 |
-- | Check if the given node list fails the N+1 check. |
97 |
verifyN1Check :: [Node.Node] -> Bool |
98 |
verifyN1Check nl = any Node.failN1 nl |
99 |
|
100 |
-- | Verifies the N+1 status and return the affected nodes. |
101 |
verifyN1 :: [Node.Node] -> [Node.Node] |
102 |
verifyN1 nl = filter Node.failN1 nl |
103 |
|
104 |
{-| Add an instance and return the new node and instance maps. -} |
105 |
addInstance :: NodeList -> Instance.Instance -> |
106 |
Node.Node -> Node.Node -> Maybe NodeList |
107 |
addInstance nl idata pri sec = |
108 |
let pdx = Node.idx pri |
109 |
sdx = Node.idx sec |
110 |
in do |
111 |
pnode <- Node.addPri pri idata |
112 |
snode <- Node.addSec sec idata pdx |
113 |
new_nl <- return $ Container.addTwo sdx snode |
114 |
pdx pnode nl |
115 |
return new_nl |
116 |
|
117 |
-- | Remove an instance and return the new node and instance maps. |
118 |
removeInstance :: NodeList -> Instance.Instance -> NodeList |
119 |
removeInstance nl idata = |
120 |
let pnode = Instance.pnode idata |
121 |
snode = Instance.snode idata |
122 |
pn = Container.find pnode nl |
123 |
sn = Container.find snode nl |
124 |
new_nl = Container.addTwo |
125 |
pnode (Node.removePri pn idata) |
126 |
snode (Node.removeSec sn idata) nl in |
127 |
new_nl |
128 |
|
129 |
-- | Remove an instance and return the new node map. |
130 |
removeInstances :: NodeList -> [Instance.Instance] -> NodeList |
131 |
removeInstances = foldl' removeInstance |
132 |
|
133 |
-- | Compute the total free disk and memory in the cluster. |
134 |
totalResources :: Container.Container Node.Node -> (Int, Int) |
135 |
totalResources nl = |
136 |
foldl' |
137 |
(\ (mem, dsk) node -> (mem + (Node.f_mem node), |
138 |
dsk + (Node.f_dsk node))) |
139 |
(0, 0) (Container.elems nl) |
140 |
|
141 |
{- | Compute a new version of a cluster given a solution. |
142 |
|
143 |
This is not used for computing the solutions, but for applying a |
144 |
(known-good) solution to the original cluster for final display. |
145 |
|
146 |
It first removes the relocated instances after which it places them on |
147 |
their new nodes. |
148 |
|
149 |
-} |
150 |
applySolution :: NodeList -> InstanceList -> [Placement] -> NodeList |
151 |
applySolution nl il sol = |
152 |
let odxes = map (\ (a, b, c, _) -> (Container.find a il, |
153 |
Node.idx (Container.find b nl), |
154 |
Node.idx (Container.find c nl)) |
155 |
) sol |
156 |
idxes = (\ (x, _, _) -> x) (unzip3 odxes) |
157 |
nc = removeInstances nl idxes |
158 |
in |
159 |
foldl' (\ nz (a, b, c) -> |
160 |
let new_p = Container.find b nz |
161 |
new_s = Container.find c nz in |
162 |
fromJust (addInstance nz a new_p new_s) |
163 |
) nc odxes |
164 |
|
165 |
|
166 |
-- First phase functions |
167 |
|
168 |
{- | Given a list 1,2,3..n build a list of pairs [(1, [2..n]), (2, |
169 |
[3..n]), ...] |
170 |
|
171 |
-} |
172 |
genParts :: [a] -> Int -> [(a, [a])] |
173 |
genParts l count = |
174 |
case l of |
175 |
[] -> [] |
176 |
x:xs -> |
177 |
if length l < count then |
178 |
[] |
179 |
else |
180 |
(x, xs) : (genParts xs count) |
181 |
|
182 |
-- | Generates combinations of count items from the names list. |
183 |
genNames :: Int -> [b] -> [[b]] |
184 |
genNames count1 names1 = |
185 |
let aux_fn count names current = |
186 |
case count of |
187 |
0 -> [current] |
188 |
_ -> |
189 |
concatMap |
190 |
(\ (x, xs) -> aux_fn (count - 1) xs (x:current)) |
191 |
(genParts names count) |
192 |
in |
193 |
aux_fn count1 names1 [] |
194 |
|
195 |
{- | Computes the pair of bad nodes and instances. |
196 |
|
197 |
The bad node list is computed via a simple 'verifyN1' check, and the |
198 |
bad instance list is the list of primary and secondary instances of |
199 |
those nodes. |
200 |
|
201 |
-} |
202 |
computeBadItems :: NodeList -> InstanceList -> |
203 |
([Node.Node], [Instance.Instance]) |
204 |
computeBadItems nl il = |
205 |
let bad_nodes = verifyN1 $ Container.elems nl |
206 |
bad_instances = map (\idx -> Container.find idx il) $ |
207 |
sort $ nub $ concat $ |
208 |
map (\ n -> (Node.slist n) ++ (Node.plist n)) bad_nodes |
209 |
in |
210 |
(bad_nodes, bad_instances) |
211 |
|
212 |
|
213 |
{- | Checks if removal of instances results in N+1 pass. |
214 |
|
215 |
Note: the check removal cannot optimize by scanning only the affected |
216 |
nodes, since the cluster is known to be not healthy; only the check |
217 |
placement can make this shortcut. |
218 |
|
219 |
-} |
220 |
checkRemoval :: NodeList -> [Instance.Instance] -> Maybe Removal |
221 |
checkRemoval nl victims = |
222 |
let nx = removeInstances nl victims |
223 |
failN1 = verifyN1Check (Container.elems nx) |
224 |
in |
225 |
if failN1 then |
226 |
Nothing |
227 |
else |
228 |
Just $ Removal nx victims |
229 |
|
230 |
|
231 |
-- | Computes the removals list for a given depth |
232 |
computeRemovals :: NodeList |
233 |
-> [Instance.Instance] |
234 |
-> Int |
235 |
-> [Maybe Removal] |
236 |
computeRemovals nl bad_instances depth = |
237 |
map (checkRemoval nl) $ genNames depth bad_instances |
238 |
|
239 |
-- Second phase functions |
240 |
|
241 |
-- | Single-node relocation cost |
242 |
nodeDelta :: Int -> Int -> Int -> Int |
243 |
nodeDelta i p s = |
244 |
if i == p || i == s then |
245 |
0 |
246 |
else |
247 |
1 |
248 |
|
249 |
{-| Compute best solution. |
250 |
|
251 |
This function compares two solutions, choosing the minimum valid |
252 |
solution. |
253 |
-} |
254 |
compareSolutions :: Maybe Solution -> Maybe Solution -> Maybe Solution |
255 |
compareSolutions a b = case (a, b) of |
256 |
(Nothing, x) -> x |
257 |
(x, Nothing) -> x |
258 |
(x, y) -> min x y |
259 |
|
260 |
-- | Compute best table. Note that the ordering of the arguments is important. |
261 |
compareTables :: Table -> Table -> Table |
262 |
compareTables a@(Table _ _ a_cv _) b@(Table _ _ b_cv _ ) = |
263 |
if a_cv > b_cv then b else a |
264 |
|
265 |
-- | Check if a given delta is worse then an existing solution. |
266 |
tooHighDelta :: Maybe Solution -> Int -> Int -> Bool |
267 |
tooHighDelta sol new_delta max_delta = |
268 |
if new_delta > max_delta && max_delta >=0 then |
269 |
True |
270 |
else |
271 |
case sol of |
272 |
Nothing -> False |
273 |
Just (Solution old_delta _) -> old_delta <= new_delta |
274 |
|
275 |
{-| Check if placement of instances still keeps the cluster N+1 compliant. |
276 |
|
277 |
This is the workhorse of the allocation algorithm: given the |
278 |
current node and instance maps, the list of instances to be |
279 |
placed, and the current solution, this will return all possible |
280 |
solution by recursing until all target instances are placed. |
281 |
|
282 |
-} |
283 |
checkPlacement :: NodeList -- ^ The current node list |
284 |
-> [Instance.Instance] -- ^ List of instances still to place |
285 |
-> [Placement] -- ^ Partial solution until now |
286 |
-> Int -- ^ The delta of the partial solution |
287 |
-> Maybe Solution -- ^ The previous solution |
288 |
-> Int -- ^ Abort if the we go above this delta |
289 |
-> Maybe Solution -- ^ The new solution |
290 |
checkPlacement nl victims current current_delta prev_sol max_delta = |
291 |
let target = head victims |
292 |
opdx = Instance.pnode target |
293 |
osdx = Instance.snode target |
294 |
vtail = tail victims |
295 |
have_tail = (length vtail) > 0 |
296 |
nodes = Container.elems nl |
297 |
iidx = Instance.idx target |
298 |
in |
299 |
foldl' |
300 |
(\ accu_p pri -> |
301 |
let |
302 |
pri_idx = Node.idx pri |
303 |
upri_delta = current_delta + nodeDelta pri_idx opdx osdx |
304 |
new_pri = Node.addPri pri target |
305 |
fail_delta1 = tooHighDelta accu_p upri_delta max_delta |
306 |
in |
307 |
if fail_delta1 || isNothing(new_pri) then accu_p |
308 |
else let pri_nl = Container.add pri_idx (fromJust new_pri) nl in |
309 |
foldl' |
310 |
(\ accu sec -> |
311 |
let |
312 |
sec_idx = Node.idx sec |
313 |
upd_delta = upri_delta + |
314 |
nodeDelta sec_idx opdx osdx |
315 |
fail_delta2 = tooHighDelta accu upd_delta max_delta |
316 |
new_sec = Node.addSec sec target pri_idx |
317 |
in |
318 |
if sec_idx == pri_idx || fail_delta2 || |
319 |
isNothing new_sec then accu |
320 |
else let |
321 |
nx = Container.add sec_idx (fromJust new_sec) pri_nl |
322 |
upd_cv = compCV nx |
323 |
plc = (iidx, pri_idx, sec_idx, upd_cv) |
324 |
c2 = plc:current |
325 |
result = |
326 |
if have_tail then |
327 |
checkPlacement nx vtail c2 upd_delta |
328 |
accu max_delta |
329 |
else |
330 |
Just (Solution upd_delta c2) |
331 |
in compareSolutions accu result |
332 |
) accu_p nodes |
333 |
) prev_sol nodes |
334 |
|
335 |
-- | Apply a move |
336 |
applyMove :: NodeList -> Instance.Instance |
337 |
-> IMove -> (Maybe NodeList, Instance.Instance, Int, Int) |
338 |
-- Failover (f) |
339 |
applyMove nl inst Failover = |
340 |
let old_pdx = Instance.pnode inst |
341 |
old_sdx = Instance.snode inst |
342 |
old_p = Container.find old_pdx nl |
343 |
old_s = Container.find old_sdx nl |
344 |
int_p = Node.removePri old_p inst |
345 |
int_s = Node.removeSec old_s inst |
346 |
new_p = Node.addPri int_s inst |
347 |
new_s = Node.addSec int_p inst old_sdx |
348 |
new_nl = if isNothing(new_p) || isNothing(new_s) then Nothing |
349 |
else Just $ Container.addTwo old_pdx (fromJust new_s) |
350 |
old_sdx (fromJust new_p) nl |
351 |
in (new_nl, Instance.setBoth inst old_sdx old_pdx, old_sdx, old_pdx) |
352 |
|
353 |
-- Replace the primary (f:, r:np, f) |
354 |
applyMove nl inst (ReplacePrimary new_pdx) = |
355 |
let old_pdx = Instance.pnode inst |
356 |
old_sdx = Instance.snode inst |
357 |
old_p = Container.find old_pdx nl |
358 |
old_s = Container.find old_sdx nl |
359 |
tgt_n = Container.find new_pdx nl |
360 |
int_p = Node.removePri old_p inst |
361 |
int_s = Node.removeSec old_s inst |
362 |
new_p = Node.addPri tgt_n inst |
363 |
new_s = Node.addSec int_s inst new_pdx |
364 |
new_nl = if isNothing(new_p) || isNothing(new_s) then Nothing |
365 |
else Just $ Container.add new_pdx (fromJust new_p) $ |
366 |
Container.addTwo old_pdx int_p |
367 |
old_sdx (fromJust new_s) nl |
368 |
in (new_nl, Instance.setPri inst new_pdx, new_pdx, old_sdx) |
369 |
|
370 |
-- Replace the secondary (r:ns) |
371 |
applyMove nl inst (ReplaceSecondary new_sdx) = |
372 |
let old_pdx = Instance.pnode inst |
373 |
old_sdx = Instance.snode inst |
374 |
old_s = Container.find old_sdx nl |
375 |
tgt_n = Container.find new_sdx nl |
376 |
int_s = Node.removeSec old_s inst |
377 |
new_s = Node.addSec tgt_n inst old_pdx |
378 |
new_nl = if isNothing(new_s) then Nothing |
379 |
else Just $ Container.addTwo new_sdx (fromJust new_s) |
380 |
old_sdx int_s nl |
381 |
in (new_nl, Instance.setSec inst new_sdx, old_pdx, new_sdx) |
382 |
|
383 |
-- Replace the secondary and failover (r:np, f) |
384 |
applyMove nl inst (ReplaceAndFailover new_pdx) = |
385 |
let old_pdx = Instance.pnode inst |
386 |
old_sdx = Instance.snode inst |
387 |
old_p = Container.find old_pdx nl |
388 |
old_s = Container.find old_sdx nl |
389 |
tgt_n = Container.find new_pdx nl |
390 |
int_p = Node.removePri old_p inst |
391 |
int_s = Node.removeSec old_s inst |
392 |
new_p = Node.addPri tgt_n inst |
393 |
new_s = Node.addSec int_p inst new_pdx |
394 |
new_nl = if isNothing(new_p) || isNothing(new_s) then Nothing |
395 |
else Just $ Container.add new_pdx (fromJust new_p) $ |
396 |
Container.addTwo old_pdx (fromJust new_s) |
397 |
old_sdx int_s nl |
398 |
in (new_nl, Instance.setBoth inst new_pdx old_pdx, new_pdx, old_pdx) |
399 |
|
400 |
-- Failver and replace the secondary (f, r:ns) |
401 |
applyMove nl inst (FailoverAndReplace new_sdx) = |
402 |
let old_pdx = Instance.pnode inst |
403 |
old_sdx = Instance.snode inst |
404 |
old_p = Container.find old_pdx nl |
405 |
old_s = Container.find old_sdx nl |
406 |
tgt_n = Container.find new_sdx nl |
407 |
int_p = Node.removePri old_p inst |
408 |
int_s = Node.removeSec old_s inst |
409 |
new_p = Node.addPri int_s inst |
410 |
new_s = Node.addSec tgt_n inst old_sdx |
411 |
new_nl = if isNothing(new_p) || isNothing(new_s) then Nothing |
412 |
else Just $ Container.add new_sdx (fromJust new_s) $ |
413 |
Container.addTwo old_sdx (fromJust new_p) |
414 |
old_pdx int_p nl |
415 |
in (new_nl, Instance.setBoth inst old_sdx new_sdx, old_sdx, new_sdx) |
416 |
|
417 |
checkSingleStep :: Table -- ^ The original table |
418 |
-> Instance.Instance -- ^ The instance to move |
419 |
-> Table -- ^ The current best table |
420 |
-> IMove -- ^ The move to apply |
421 |
-> Table -- ^ The final best table |
422 |
checkSingleStep ini_tbl target cur_tbl move = |
423 |
let |
424 |
Table ini_nl ini_il _ ini_plc = ini_tbl |
425 |
(tmp_nl, new_inst, pri_idx, sec_idx) = applyMove ini_nl target move |
426 |
in |
427 |
if isNothing tmp_nl then cur_tbl |
428 |
else |
429 |
let tgt_idx = Instance.idx target |
430 |
upd_nl = fromJust tmp_nl |
431 |
upd_cvar = compCV upd_nl |
432 |
upd_il = Container.add tgt_idx new_inst ini_il |
433 |
upd_plc = (tgt_idx, pri_idx, sec_idx, upd_cvar):ini_plc |
434 |
upd_tbl = Table upd_nl upd_il upd_cvar upd_plc |
435 |
in |
436 |
compareTables cur_tbl upd_tbl |
437 |
|
438 |
-- | Given the status of the current secondary as a valid new node |
439 |
-- and the current candidate target node, |
440 |
-- generate the possible moves for a instance. |
441 |
possibleMoves :: Bool -> Int -> [IMove] |
442 |
possibleMoves True tdx = |
443 |
[ReplaceSecondary tdx, |
444 |
ReplaceAndFailover tdx, |
445 |
ReplacePrimary tdx, |
446 |
FailoverAndReplace tdx] |
447 |
|
448 |
possibleMoves False tdx = |
449 |
[ReplaceSecondary tdx, |
450 |
ReplaceAndFailover tdx] |
451 |
|
452 |
-- | Compute the best move for a given instance. |
453 |
checkInstanceMove :: [Int] -- Allowed target node indices |
454 |
-> Table -- Original table |
455 |
-> Instance.Instance -- Instance to move |
456 |
-> Table -- Best new table for this instance |
457 |
checkInstanceMove nodes_idx ini_tbl target = |
458 |
let |
459 |
opdx = Instance.pnode target |
460 |
osdx = Instance.snode target |
461 |
nodes = filter (\idx -> idx /= opdx && idx /= osdx) nodes_idx |
462 |
use_secondary = elem osdx nodes_idx |
463 |
aft_failover = if use_secondary -- if allowed to failover |
464 |
then checkSingleStep ini_tbl target ini_tbl Failover |
465 |
else ini_tbl |
466 |
all_moves = concatMap (possibleMoves use_secondary) nodes |
467 |
in |
468 |
-- iterate over the possible nodes for this instance |
469 |
foldl' (checkSingleStep ini_tbl target) aft_failover all_moves |
470 |
|
471 |
-- | Compute the best next move. |
472 |
checkMove :: [Int] -- ^ Allowed target node indices |
473 |
-> Table -- ^ The current solution |
474 |
-> [Instance.Instance] -- ^ List of instances still to move |
475 |
-> Table -- ^ The new solution |
476 |
checkMove nodes_idx ini_tbl victims = |
477 |
let Table _ _ _ ini_plc = ini_tbl |
478 |
-- iterate over all instances, computing the best move |
479 |
best_tbl = |
480 |
foldl' |
481 |
(\ step_tbl elem -> compareTables step_tbl $ |
482 |
checkInstanceMove nodes_idx ini_tbl elem) |
483 |
ini_tbl victims |
484 |
Table _ _ _ best_plc = best_tbl |
485 |
in |
486 |
if length best_plc == length ini_plc then -- no advancement |
487 |
ini_tbl |
488 |
else |
489 |
best_tbl |
490 |
|
491 |
{- | Auxiliary function for solution computation. |
492 |
|
493 |
We write this in an explicit recursive fashion in order to control |
494 |
early-abort in case we have met the min delta. We can't use foldr |
495 |
instead of explicit recursion since we need the accumulator for the |
496 |
abort decision. |
497 |
|
498 |
-} |
499 |
advanceSolution :: [Maybe Removal] -- ^ The removal to process |
500 |
-> Int -- ^ Minimum delta parameter |
501 |
-> Int -- ^ Maximum delta parameter |
502 |
-> Maybe Solution -- ^ Current best solution |
503 |
-> Maybe Solution -- ^ New best solution |
504 |
advanceSolution [] _ _ sol = sol |
505 |
advanceSolution (Nothing:xs) m n sol = advanceSolution xs m n sol |
506 |
advanceSolution ((Just (Removal nx removed)):xs) min_d max_d prev_sol = |
507 |
let new_sol = checkPlacement nx removed [] 0 prev_sol max_d |
508 |
new_delta = solutionDelta $! new_sol |
509 |
in |
510 |
if new_delta >= 0 && new_delta <= min_d then |
511 |
new_sol |
512 |
else |
513 |
advanceSolution xs min_d max_d new_sol |
514 |
|
515 |
-- | Computes the placement solution. |
516 |
solutionFromRemovals :: [Maybe Removal] -- ^ The list of (possible) removals |
517 |
-> Int -- ^ Minimum delta parameter |
518 |
-> Int -- ^ Maximum delta parameter |
519 |
-> Maybe Solution -- ^ The best solution found |
520 |
solutionFromRemovals removals min_delta max_delta = |
521 |
advanceSolution removals min_delta max_delta Nothing |
522 |
|
523 |
{- | Computes the solution at the given depth. |
524 |
|
525 |
This is a wrapper over both computeRemovals and |
526 |
solutionFromRemovals. In case we have no solution, we return Nothing. |
527 |
|
528 |
-} |
529 |
computeSolution :: NodeList -- ^ The original node data |
530 |
-> [Instance.Instance] -- ^ The list of /bad/ instances |
531 |
-> Int -- ^ The /depth/ of removals |
532 |
-> Int -- ^ Maximum number of removals to process |
533 |
-> Int -- ^ Minimum delta parameter |
534 |
-> Int -- ^ Maximum delta parameter |
535 |
-> Maybe Solution -- ^ The best solution found (or Nothing) |
536 |
computeSolution nl bad_instances depth max_removals min_delta max_delta = |
537 |
let |
538 |
removals = computeRemovals nl bad_instances depth |
539 |
removals' = capRemovals removals max_removals |
540 |
in |
541 |
solutionFromRemovals removals' min_delta max_delta |
542 |
|
543 |
-- Solution display functions (pure) |
544 |
|
545 |
-- | Given the original and final nodes, computes the relocation description. |
546 |
computeMoves :: String -- ^ The instance name |
547 |
-> String -- ^ Original primary |
548 |
-> String -- ^ Original secondary |
549 |
-> String -- ^ New primary |
550 |
-> String -- ^ New secondary |
551 |
-> (String, [String]) |
552 |
-- ^ Tuple of moves and commands list; moves is containing |
553 |
-- either @/f/@ for failover or @/r:name/@ for replace |
554 |
-- secondary, while the command list holds gnt-instance |
555 |
-- commands (without that prefix), e.g \"@failover instance1@\" |
556 |
computeMoves i a b c d = |
557 |
if c == a then {- Same primary -} |
558 |
if d == b then {- Same sec??! -} |
559 |
("-", []) |
560 |
else {- Change of secondary -} |
561 |
(printf "r:%s" d, |
562 |
[printf "replace-disks -n %s %s" d i]) |
563 |
else |
564 |
if c == b then {- Failover and ... -} |
565 |
if d == a then {- that's all -} |
566 |
("f", [printf "migrate %s" i]) |
567 |
else |
568 |
(printf "f r:%s" d, |
569 |
[printf "migrate %s" i, |
570 |
printf "replace-disks -n %s %s" d i]) |
571 |
else |
572 |
if d == a then {- ... and keep primary as secondary -} |
573 |
(printf "r:%s f" c, |
574 |
[printf "replace-disks -n %s %s" c i, |
575 |
printf "migrate %s" i]) |
576 |
else |
577 |
if d == b then {- ... keep same secondary -} |
578 |
(printf "f r:%s f" c, |
579 |
[printf "migrate %s" i, |
580 |
printf "replace-disks -n %s %s" c i, |
581 |
printf "migrate %s" i]) |
582 |
|
583 |
else {- Nothing in common -} |
584 |
(printf "r:%s f r:%s" c d, |
585 |
[printf "replace-disks -n %s %s" c i, |
586 |
printf "migrate %s" i, |
587 |
printf "replace-disks -n %s %s" d i]) |
588 |
|
589 |
{-| Converts a placement to string format -} |
590 |
printSolutionLine :: InstanceList |
591 |
-> NameList |
592 |
-> NameList |
593 |
-> Int |
594 |
-> Int |
595 |
-> Placement |
596 |
-> Int |
597 |
-> (String, [String]) |
598 |
printSolutionLine il ktn kti nmlen imlen plc pos = |
599 |
let |
600 |
pmlen = (2*nmlen + 1) |
601 |
(i, p, s, c) = plc |
602 |
inst = Container.find i il |
603 |
inam = fromJust $ lookup (Instance.idx inst) kti |
604 |
npri = fromJust $ lookup p ktn |
605 |
nsec = fromJust $ lookup s ktn |
606 |
opri = fromJust $ lookup (Instance.pnode inst) ktn |
607 |
osec = fromJust $ lookup (Instance.snode inst) ktn |
608 |
(moves, cmds) = computeMoves inam opri osec npri nsec |
609 |
ostr = (printf "%s:%s" opri osec)::String |
610 |
nstr = (printf "%s:%s" npri nsec)::String |
611 |
in |
612 |
(printf " %3d. %-*s %-*s => %-*s %.8f a=%s" |
613 |
pos imlen inam pmlen ostr |
614 |
pmlen nstr c moves, |
615 |
cmds) |
616 |
|
617 |
formatCmds :: [[String]] -> String |
618 |
formatCmds cmd_strs = |
619 |
unlines $ map (" echo " ++) $ |
620 |
concat $ map (\(a, b) -> |
621 |
(printf "step %d" (a::Int)):(map ("gnt-instance " ++) b)) $ |
622 |
zip [1..] cmd_strs |
623 |
|
624 |
{-| Converts a solution to string format -} |
625 |
printSolution :: InstanceList |
626 |
-> NameList |
627 |
-> NameList |
628 |
-> [Placement] |
629 |
-> ([String], [[String]]) |
630 |
printSolution il ktn kti sol = |
631 |
let |
632 |
mlen_fn = maximum . (map length) . snd . unzip |
633 |
imlen = mlen_fn kti |
634 |
nmlen = mlen_fn ktn |
635 |
in |
636 |
unzip $ map (uncurry $ printSolutionLine il ktn kti nmlen imlen) $ |
637 |
zip sol [1..] |
638 |
|
639 |
-- | Print the node list. |
640 |
printNodes :: NameList -> NodeList -> String |
641 |
printNodes ktn nl = |
642 |
let snl = sortBy (compare `on` Node.idx) (Container.elems nl) |
643 |
snl' = map (\ n -> ((fromJust $ lookup (Node.idx n) ktn), n)) snl |
644 |
m_name = maximum . (map length) . fst . unzip $ snl' |
645 |
helper = Node.list m_name |
646 |
header = printf "%2s %-*s %5s %5s %5s %5s %5s %5s %3s %3s %7s %7s" |
647 |
" F" m_name "Name" "t_mem" "n_mem" "f_mem" "r_mem" |
648 |
"t_dsk" "f_dsk" |
649 |
"pri" "sec" "p_fmem" "p_fdsk" |
650 |
in unlines $ (header:map (uncurry helper) snl') |
651 |
|
652 |
-- | Compute the mem and disk covariance. |
653 |
compDetailedCV :: NodeList -> (Double, Double, Double, Double, Double) |
654 |
compDetailedCV nl = |
655 |
let |
656 |
all_nodes = Container.elems nl |
657 |
(offline, nodes) = partition Node.offline all_nodes |
658 |
mem_l = map Node.p_mem nodes |
659 |
dsk_l = map Node.p_dsk nodes |
660 |
mem_cv = varianceCoeff mem_l |
661 |
dsk_cv = varianceCoeff dsk_l |
662 |
n1_l = length $ filter Node.failN1 nodes |
663 |
n1_score = (fromIntegral n1_l) / (fromIntegral $ length nodes) |
664 |
res_l = map Node.p_rem nodes |
665 |
res_cv = varianceCoeff res_l |
666 |
offline_inst = sum . map (\n -> (length . Node.plist $ n) + |
667 |
(length . Node.slist $ n)) $ offline |
668 |
online_inst = sum . map (\n -> (length . Node.plist $ n) + |
669 |
(length . Node.slist $ n)) $ nodes |
670 |
off_score = (fromIntegral offline_inst) / |
671 |
(fromIntegral $ online_inst + offline_inst) |
672 |
in (mem_cv, dsk_cv, n1_score, res_cv, off_score) |
673 |
|
674 |
-- | Compute the 'total' variance. |
675 |
compCV :: NodeList -> Double |
676 |
compCV nl = |
677 |
let (mem_cv, dsk_cv, n1_score, res_cv, off_score) = compDetailedCV nl |
678 |
in mem_cv + dsk_cv + n1_score + res_cv + off_score |
679 |
|
680 |
printStats :: NodeList -> String |
681 |
printStats nl = |
682 |
let (mem_cv, dsk_cv, n1_score, res_cv, off_score) = compDetailedCV nl |
683 |
in printf "f_mem=%.8f, r_mem=%.8f, f_dsk=%.8f, n1=%.3f, uf=%.3f" |
684 |
mem_cv res_cv dsk_cv n1_score off_score |
685 |
|
686 |
-- Balancing functions |
687 |
|
688 |
-- Loading functions |
689 |
|
690 |
{- | Convert newline and delimiter-separated text. |
691 |
|
692 |
This function converts a text in tabular format as generated by |
693 |
@gnt-instance list@ and @gnt-node list@ to a list of objects using a |
694 |
supplied conversion function. |
695 |
|
696 |
-} |
697 |
loadTabular :: String -> ([String] -> (String, a)) |
698 |
-> (a -> Int -> a) -> ([(String, Int)], [(Int, a)]) |
699 |
loadTabular text_data convert_fn set_fn = |
700 |
let lines_data = lines text_data |
701 |
rows = map (sepSplit '|') lines_data |
702 |
kerows = (map convert_fn rows) |
703 |
idxrows = map (\ (idx, (k, v)) -> ((k, idx), (idx, set_fn v idx))) |
704 |
(zip [0..] kerows) |
705 |
in unzip idxrows |
706 |
|
707 |
-- | For each instance, add its index to its primary and secondary nodes |
708 |
fixNodes :: [(Int, Node.Node)] |
709 |
-> [(Int, Instance.Instance)] |
710 |
-> [(Int, Node.Node)] |
711 |
fixNodes nl il = |
712 |
foldl' (\accu (idx, inst) -> |
713 |
let |
714 |
assocEqual = (\ (i, _) (j, _) -> i == j) |
715 |
pdx = Instance.pnode inst |
716 |
sdx = Instance.snode inst |
717 |
pold = fromJust $ lookup pdx accu |
718 |
sold = fromJust $ lookup sdx accu |
719 |
pnew = Node.setPri pold idx |
720 |
snew = Node.setSec sold idx |
721 |
ac1 = deleteBy assocEqual (pdx, pold) accu |
722 |
ac2 = deleteBy assocEqual (sdx, sold) ac1 |
723 |
ac3 = (pdx, pnew):(sdx, snew):ac2 |
724 |
in ac3) nl il |
725 |
|
726 |
-- | Compute the longest common suffix of a NameList list that |
727 |
-- | starts with a dot |
728 |
longestDomain :: NameList -> String |
729 |
longestDomain [] = "" |
730 |
longestDomain ((_,x):xs) = |
731 |
let |
732 |
onlyStrings = snd $ unzip xs |
733 |
in |
734 |
foldr (\ suffix accu -> if all (isSuffixOf suffix) onlyStrings |
735 |
then suffix |
736 |
else accu) |
737 |
"" $ filter (isPrefixOf ".") (tails x) |
738 |
|
739 |
-- | Remove tails from the (Int, String) lists |
740 |
stripSuffix :: String -> NameList -> NameList |
741 |
stripSuffix suffix lst = |
742 |
let sflen = length suffix in |
743 |
map (\ (key, name) -> (key, take ((length name) - sflen) name)) lst |
744 |
|
745 |
{-| Initializer function that loads the data from a node and list file |
746 |
and massages it into the correct format. -} |
747 |
loadData :: String -- ^ Node data in text format |
748 |
-> String -- ^ Instance data in text format |
749 |
-> (Container.Container Node.Node, |
750 |
Container.Container Instance.Instance, |
751 |
String, NameList, NameList) |
752 |
loadData ndata idata = |
753 |
let |
754 |
{- node file: name t_mem n_mem f_mem t_disk f_disk -} |
755 |
(ktn, nl) = loadTabular ndata |
756 |
(\ (name:tm:nm:fm:td:fd:[]) -> |
757 |
(name, |
758 |
Node.create (read tm) (read nm) |
759 |
(read fm) (read td) (read fd))) |
760 |
Node.setIdx |
761 |
{- instance file: name mem disk pnode snode -} |
762 |
(kti, il) = loadTabular idata |
763 |
(\ (name:mem:dsk:pnode:snode:[]) -> |
764 |
(name, |
765 |
Instance.create (read mem) (read dsk) |
766 |
(fromJust $ lookup pnode ktn) |
767 |
(fromJust $ lookup snode ktn))) |
768 |
Instance.setIdx |
769 |
nl2 = fixNodes nl il |
770 |
il3 = Container.fromAssocList il |
771 |
nl3 = Container.fromAssocList |
772 |
(map (\ (k, v) -> (k, Node.buildPeers v il3 (length nl2))) nl2) |
773 |
xtn = swapPairs ktn |
774 |
xti = swapPairs kti |
775 |
common_suffix = longestDomain (xti ++ xtn) |
776 |
stn = stripSuffix common_suffix xtn |
777 |
sti = stripSuffix common_suffix xti |
778 |
in |
779 |
(nl3, il3, common_suffix, stn, sti) |
780 |
|
781 |
-- | Compute the amount of memory used by primary instances on a node. |
782 |
nodeImem :: Node.Node -> InstanceList -> Int |
783 |
nodeImem node il = |
784 |
let rfind = flip Container.find $ il |
785 |
in sum . map Instance.mem . |
786 |
map rfind $ Node.plist node |
787 |
|
788 |
|
789 |
-- | Check cluster data for consistency |
790 |
checkData :: NodeList -> InstanceList -> NameList -> NameList |
791 |
-> ([String], NodeList) |
792 |
checkData nl il ktn kti = |
793 |
Container.mapAccum |
794 |
(\ msgs node -> |
795 |
let nname = fromJust $ lookup (Node.idx node) ktn |
796 |
delta_mem = (truncate $ Node.t_mem node) - |
797 |
(Node.n_mem node) - |
798 |
(Node.f_mem node) - |
799 |
(nodeImem node il) |
800 |
newn = Node.setXmem node delta_mem |
801 |
umsg = if delta_mem > 16 |
802 |
then (printf "node %s has %6d MB of unaccounted \ |
803 |
\memory " |
804 |
nname delta_mem):msgs |
805 |
else msgs |
806 |
in (umsg, newn) |
807 |
) [] nl |