root / Ganeti / HTools / Cluster.hs @ 9d3fada5
History | View | Annotate | Download (30.8 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 |
, Score |
19 |
-- * Generic functions |
20 |
, totalResources |
21 |
-- * First phase functions |
22 |
, computeBadItems |
23 |
-- * Second phase functions |
24 |
, computeSolution |
25 |
, applySolution |
26 |
, printSolution |
27 |
, printSolutionLine |
28 |
, formatCmds |
29 |
, printNodes |
30 |
-- * Balacing functions |
31 |
, checkMove |
32 |
, compCV |
33 |
, printStats |
34 |
-- * Loading functions |
35 |
, loadData |
36 |
, checkData |
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.Utils |
49 |
|
50 |
type NodeList = Container.Container Node.Node |
51 |
type InstanceList = Container.Container Instance.Instance |
52 |
-- | The type used to hold idx-to-name mappings |
53 |
type NameList = [(Int, String)] |
54 |
-- | A separate name for the cluster score type |
55 |
type Score = Double |
56 |
|
57 |
-- | The description of an instance placement. |
58 |
type Placement = (Int, Int, Int, Score) |
59 |
|
60 |
{- | A cluster solution described as the solution delta and the list |
61 |
of placements. |
62 |
|
63 |
-} |
64 |
data Solution = Solution Int [Placement] |
65 |
deriving (Eq, Ord, Show) |
66 |
|
67 |
-- | Returns the delta of a solution or -1 for Nothing |
68 |
solutionDelta :: Maybe Solution -> Int |
69 |
solutionDelta sol = case sol of |
70 |
Just (Solution d _) -> d |
71 |
_ -> -1 |
72 |
|
73 |
-- | A removal set. |
74 |
data Removal = Removal NodeList [Instance.Instance] |
75 |
|
76 |
-- | An instance move definition |
77 |
data IMove = Failover -- ^ Failover the instance (f) |
78 |
| ReplacePrimary Int -- ^ Replace primary (f, r:np, f) |
79 |
| ReplaceSecondary Int -- ^ Replace secondary (r:ns) |
80 |
| ReplaceAndFailover Int -- ^ Replace secondary, failover (r:np, f) |
81 |
| FailoverAndReplace Int -- ^ Failover, replace secondary (f, r:ns) |
82 |
deriving (Show) |
83 |
|
84 |
-- | The complete state for the balancing solution |
85 |
data Table = Table NodeList InstanceList Score [Placement] |
86 |
deriving (Show) |
87 |
|
88 |
-- General functions |
89 |
|
90 |
-- | Cap the removal list if needed. |
91 |
capRemovals :: [a] -> Int -> [a] |
92 |
capRemovals removals max_removals = |
93 |
if max_removals > 0 then |
94 |
take max_removals removals |
95 |
else |
96 |
removals |
97 |
|
98 |
-- | Check if the given node list fails the N+1 check. |
99 |
verifyN1Check :: [Node.Node] -> Bool |
100 |
verifyN1Check nl = any Node.failN1 nl |
101 |
|
102 |
-- | Verifies the N+1 status and return the affected nodes. |
103 |
verifyN1 :: [Node.Node] -> [Node.Node] |
104 |
verifyN1 nl = filter Node.failN1 nl |
105 |
|
106 |
{-| Add an instance and return the new node and instance maps. -} |
107 |
addInstance :: NodeList -> Instance.Instance -> |
108 |
Node.Node -> Node.Node -> Maybe NodeList |
109 |
addInstance nl idata pri sec = |
110 |
let pdx = Node.idx pri |
111 |
sdx = Node.idx sec |
112 |
in do |
113 |
pnode <- Node.addPri pri idata |
114 |
snode <- Node.addSec sec idata pdx |
115 |
new_nl <- return $ Container.addTwo sdx snode |
116 |
pdx pnode nl |
117 |
return new_nl |
118 |
|
119 |
-- | Remove an instance and return the new node and instance maps. |
120 |
removeInstance :: NodeList -> Instance.Instance -> NodeList |
121 |
removeInstance nl idata = |
122 |
let pnode = Instance.pnode idata |
123 |
snode = Instance.snode idata |
124 |
pn = Container.find pnode nl |
125 |
sn = Container.find snode nl |
126 |
new_nl = Container.addTwo |
127 |
pnode (Node.removePri pn idata) |
128 |
snode (Node.removeSec sn idata) nl in |
129 |
new_nl |
130 |
|
131 |
-- | Remove an instance and return the new node map. |
132 |
removeInstances :: NodeList -> [Instance.Instance] -> NodeList |
133 |
removeInstances = foldl' removeInstance |
134 |
|
135 |
-- | Compute the total free disk and memory in the cluster. |
136 |
totalResources :: Container.Container Node.Node -> (Int, Int) |
137 |
totalResources nl = |
138 |
foldl' |
139 |
(\ (mem, dsk) node -> (mem + (Node.f_mem node), |
140 |
dsk + (Node.f_dsk node))) |
141 |
(0, 0) (Container.elems nl) |
142 |
|
143 |
{- | Compute a new version of a cluster given a solution. |
144 |
|
145 |
This is not used for computing the solutions, but for applying a |
146 |
(known-good) solution to the original cluster for final display. |
147 |
|
148 |
It first removes the relocated instances after which it places them on |
149 |
their new nodes. |
150 |
|
151 |
-} |
152 |
applySolution :: NodeList -> InstanceList -> [Placement] -> NodeList |
153 |
applySolution nl il sol = |
154 |
let odxes = map (\ (a, b, c, _) -> (Container.find a il, |
155 |
Node.idx (Container.find b nl), |
156 |
Node.idx (Container.find c nl)) |
157 |
) sol |
158 |
idxes = (\ (x, _, _) -> x) (unzip3 odxes) |
159 |
nc = removeInstances nl idxes |
160 |
in |
161 |
foldl' (\ nz (a, b, c) -> |
162 |
let new_p = Container.find b nz |
163 |
new_s = Container.find c nz in |
164 |
fromJust (addInstance nz a new_p new_s) |
165 |
) nc odxes |
166 |
|
167 |
|
168 |
-- First phase functions |
169 |
|
170 |
{- | Given a list 1,2,3..n build a list of pairs [(1, [2..n]), (2, |
171 |
[3..n]), ...] |
172 |
|
173 |
-} |
174 |
genParts :: [a] -> Int -> [(a, [a])] |
175 |
genParts l count = |
176 |
case l of |
177 |
[] -> [] |
178 |
x:xs -> |
179 |
if length l < count then |
180 |
[] |
181 |
else |
182 |
(x, xs) : (genParts xs count) |
183 |
|
184 |
-- | Generates combinations of count items from the names list. |
185 |
genNames :: Int -> [b] -> [[b]] |
186 |
genNames count1 names1 = |
187 |
let aux_fn count names current = |
188 |
case count of |
189 |
0 -> [current] |
190 |
_ -> |
191 |
concatMap |
192 |
(\ (x, xs) -> aux_fn (count - 1) xs (x:current)) |
193 |
(genParts names count) |
194 |
in |
195 |
aux_fn count1 names1 [] |
196 |
|
197 |
{- | Computes the pair of bad nodes and instances. |
198 |
|
199 |
The bad node list is computed via a simple 'verifyN1' check, and the |
200 |
bad instance list is the list of primary and secondary instances of |
201 |
those nodes. |
202 |
|
203 |
-} |
204 |
computeBadItems :: NodeList -> InstanceList -> |
205 |
([Node.Node], [Instance.Instance]) |
206 |
computeBadItems nl il = |
207 |
let bad_nodes = verifyN1 $ filter (not . Node.offline) $ Container.elems nl |
208 |
bad_instances = map (\idx -> Container.find idx il) $ |
209 |
sort $ nub $ concat $ |
210 |
map (\ n -> (Node.slist n) ++ (Node.plist n)) bad_nodes |
211 |
in |
212 |
(bad_nodes, bad_instances) |
213 |
|
214 |
|
215 |
{- | Checks if removal of instances results in N+1 pass. |
216 |
|
217 |
Note: the check removal cannot optimize by scanning only the affected |
218 |
nodes, since the cluster is known to be not healthy; only the check |
219 |
placement can make this shortcut. |
220 |
|
221 |
-} |
222 |
checkRemoval :: NodeList -> [Instance.Instance] -> Maybe Removal |
223 |
checkRemoval nl victims = |
224 |
let nx = removeInstances nl victims |
225 |
failN1 = verifyN1Check (Container.elems nx) |
226 |
in |
227 |
if failN1 then |
228 |
Nothing |
229 |
else |
230 |
Just $ Removal nx victims |
231 |
|
232 |
|
233 |
-- | Computes the removals list for a given depth |
234 |
computeRemovals :: NodeList |
235 |
-> [Instance.Instance] |
236 |
-> Int |
237 |
-> [Maybe Removal] |
238 |
computeRemovals nl bad_instances depth = |
239 |
map (checkRemoval nl) $ genNames depth bad_instances |
240 |
|
241 |
-- Second phase functions |
242 |
|
243 |
-- | Single-node relocation cost |
244 |
nodeDelta :: Int -> Int -> Int -> Int |
245 |
nodeDelta i p s = |
246 |
if i == p || i == s then |
247 |
0 |
248 |
else |
249 |
1 |
250 |
|
251 |
{-| Compute best solution. |
252 |
|
253 |
This function compares two solutions, choosing the minimum valid |
254 |
solution. |
255 |
-} |
256 |
compareSolutions :: Maybe Solution -> Maybe Solution -> Maybe Solution |
257 |
compareSolutions a b = case (a, b) of |
258 |
(Nothing, x) -> x |
259 |
(x, Nothing) -> x |
260 |
(x, y) -> min x y |
261 |
|
262 |
-- | Compute best table. Note that the ordering of the arguments is important. |
263 |
compareTables :: Table -> Table -> Table |
264 |
compareTables a@(Table _ _ a_cv _) b@(Table _ _ b_cv _ ) = |
265 |
if a_cv > b_cv then b else a |
266 |
|
267 |
-- | Check if a given delta is worse then an existing solution. |
268 |
tooHighDelta :: Maybe Solution -> Int -> Int -> Bool |
269 |
tooHighDelta sol new_delta max_delta = |
270 |
if new_delta > max_delta && max_delta >=0 then |
271 |
True |
272 |
else |
273 |
case sol of |
274 |
Nothing -> False |
275 |
Just (Solution old_delta _) -> old_delta <= new_delta |
276 |
|
277 |
{-| Check if placement of instances still keeps the cluster N+1 compliant. |
278 |
|
279 |
This is the workhorse of the allocation algorithm: given the |
280 |
current node and instance maps, the list of instances to be |
281 |
placed, and the current solution, this will return all possible |
282 |
solution by recursing until all target instances are placed. |
283 |
|
284 |
-} |
285 |
checkPlacement :: NodeList -- ^ The current node list |
286 |
-> [Instance.Instance] -- ^ List of instances still to place |
287 |
-> [Placement] -- ^ Partial solution until now |
288 |
-> Int -- ^ The delta of the partial solution |
289 |
-> Maybe Solution -- ^ The previous solution |
290 |
-> Int -- ^ Abort if the we go above this delta |
291 |
-> Maybe Solution -- ^ The new solution |
292 |
checkPlacement nl victims current current_delta prev_sol max_delta = |
293 |
let target = head victims |
294 |
opdx = Instance.pnode target |
295 |
osdx = Instance.snode target |
296 |
vtail = tail victims |
297 |
have_tail = (length vtail) > 0 |
298 |
nodes = Container.elems nl |
299 |
iidx = Instance.idx target |
300 |
in |
301 |
foldl' |
302 |
(\ accu_p pri -> |
303 |
let |
304 |
pri_idx = Node.idx pri |
305 |
upri_delta = current_delta + nodeDelta pri_idx opdx osdx |
306 |
new_pri = Node.addPri pri target |
307 |
fail_delta1 = tooHighDelta accu_p upri_delta max_delta |
308 |
in |
309 |
if fail_delta1 || isNothing(new_pri) then accu_p |
310 |
else let pri_nl = Container.add pri_idx (fromJust new_pri) nl in |
311 |
foldl' |
312 |
(\ accu sec -> |
313 |
let |
314 |
sec_idx = Node.idx sec |
315 |
upd_delta = upri_delta + |
316 |
nodeDelta sec_idx opdx osdx |
317 |
fail_delta2 = tooHighDelta accu upd_delta max_delta |
318 |
new_sec = Node.addSec sec target pri_idx |
319 |
in |
320 |
if sec_idx == pri_idx || fail_delta2 || |
321 |
isNothing new_sec then accu |
322 |
else let |
323 |
nx = Container.add sec_idx (fromJust new_sec) pri_nl |
324 |
upd_cv = compCV nx |
325 |
plc = (iidx, pri_idx, sec_idx, upd_cv) |
326 |
c2 = plc:current |
327 |
result = |
328 |
if have_tail then |
329 |
checkPlacement nx vtail c2 upd_delta |
330 |
accu max_delta |
331 |
else |
332 |
Just (Solution upd_delta c2) |
333 |
in compareSolutions accu result |
334 |
) accu_p nodes |
335 |
) prev_sol nodes |
336 |
|
337 |
-- | Apply a move |
338 |
applyMove :: NodeList -> Instance.Instance |
339 |
-> IMove -> (Maybe NodeList, Instance.Instance, Int, Int) |
340 |
-- Failover (f) |
341 |
applyMove nl inst Failover = |
342 |
let old_pdx = Instance.pnode inst |
343 |
old_sdx = Instance.snode inst |
344 |
old_p = Container.find old_pdx nl |
345 |
old_s = Container.find old_sdx nl |
346 |
int_p = Node.removePri old_p inst |
347 |
int_s = Node.removeSec old_s inst |
348 |
new_nl = do -- Maybe monad |
349 |
new_p <- Node.addPri int_s inst |
350 |
new_s <- Node.addSec int_p inst old_sdx |
351 |
return $ Container.addTwo old_pdx new_s old_sdx new_p nl |
352 |
in (new_nl, Instance.setBoth inst old_sdx old_pdx, old_sdx, old_pdx) |
353 |
|
354 |
-- Replace the primary (f:, r:np, f) |
355 |
applyMove nl inst (ReplacePrimary new_pdx) = |
356 |
let old_pdx = Instance.pnode inst |
357 |
old_sdx = Instance.snode inst |
358 |
old_p = Container.find old_pdx nl |
359 |
old_s = Container.find old_sdx nl |
360 |
tgt_n = Container.find new_pdx nl |
361 |
int_p = Node.removePri old_p inst |
362 |
int_s = Node.removeSec old_s inst |
363 |
new_nl = do -- Maybe monad |
364 |
new_p <- Node.addPri tgt_n inst |
365 |
new_s <- Node.addSec int_s inst new_pdx |
366 |
return $ Container.add new_pdx new_p $ |
367 |
Container.addTwo old_pdx int_p old_sdx 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_nl = Node.addSec tgt_n inst old_pdx >>= |
378 |
\new_s -> return $ Container.addTwo new_sdx |
379 |
new_s old_sdx int_s nl |
380 |
in (new_nl, Instance.setSec inst new_sdx, old_pdx, new_sdx) |
381 |
|
382 |
-- Replace the secondary and failover (r:np, f) |
383 |
applyMove nl inst (ReplaceAndFailover new_pdx) = |
384 |
let old_pdx = Instance.pnode inst |
385 |
old_sdx = Instance.snode inst |
386 |
old_p = Container.find old_pdx nl |
387 |
old_s = Container.find old_sdx nl |
388 |
tgt_n = Container.find new_pdx nl |
389 |
int_p = Node.removePri old_p inst |
390 |
int_s = Node.removeSec old_s inst |
391 |
new_nl = do -- Maybe monad |
392 |
new_p <- Node.addPri tgt_n inst |
393 |
new_s <- Node.addSec int_p inst new_pdx |
394 |
return $ Container.add new_pdx new_p $ |
395 |
Container.addTwo old_pdx new_s old_sdx int_s nl |
396 |
in (new_nl, Instance.setBoth inst new_pdx old_pdx, new_pdx, old_pdx) |
397 |
|
398 |
-- Failver and replace the secondary (f, r:ns) |
399 |
applyMove nl inst (FailoverAndReplace new_sdx) = |
400 |
let old_pdx = Instance.pnode inst |
401 |
old_sdx = Instance.snode inst |
402 |
old_p = Container.find old_pdx nl |
403 |
old_s = Container.find old_sdx nl |
404 |
tgt_n = Container.find new_sdx nl |
405 |
int_p = Node.removePri old_p inst |
406 |
int_s = Node.removeSec old_s inst |
407 |
new_nl = do -- Maybe monad |
408 |
new_p <- Node.addPri int_s inst |
409 |
new_s <- Node.addSec tgt_n inst old_sdx |
410 |
return $ Container.add new_sdx new_s $ |
411 |
Container.addTwo old_sdx new_p old_pdx int_p nl |
412 |
in (new_nl, Instance.setBoth inst old_sdx new_sdx, old_sdx, new_sdx) |
413 |
|
414 |
checkSingleStep :: Table -- ^ The original table |
415 |
-> Instance.Instance -- ^ The instance to move |
416 |
-> Table -- ^ The current best table |
417 |
-> IMove -- ^ The move to apply |
418 |
-> Table -- ^ The final best table |
419 |
checkSingleStep ini_tbl target cur_tbl move = |
420 |
let |
421 |
Table ini_nl ini_il _ ini_plc = ini_tbl |
422 |
(tmp_nl, new_inst, pri_idx, sec_idx) = applyMove ini_nl target move |
423 |
in |
424 |
if isNothing tmp_nl then cur_tbl |
425 |
else |
426 |
let tgt_idx = Instance.idx target |
427 |
upd_nl = fromJust tmp_nl |
428 |
upd_cvar = compCV upd_nl |
429 |
upd_il = Container.add tgt_idx new_inst ini_il |
430 |
upd_plc = (tgt_idx, pri_idx, sec_idx, upd_cvar):ini_plc |
431 |
upd_tbl = Table upd_nl upd_il upd_cvar upd_plc |
432 |
in |
433 |
compareTables cur_tbl upd_tbl |
434 |
|
435 |
-- | Given the status of the current secondary as a valid new node |
436 |
-- and the current candidate target node, |
437 |
-- generate the possible moves for a instance. |
438 |
possibleMoves :: Bool -> Int -> [IMove] |
439 |
possibleMoves True tdx = |
440 |
[ReplaceSecondary tdx, |
441 |
ReplaceAndFailover tdx, |
442 |
ReplacePrimary tdx, |
443 |
FailoverAndReplace tdx] |
444 |
|
445 |
possibleMoves False tdx = |
446 |
[ReplaceSecondary tdx, |
447 |
ReplaceAndFailover tdx] |
448 |
|
449 |
-- | Compute the best move for a given instance. |
450 |
checkInstanceMove :: [Int] -- Allowed target node indices |
451 |
-> Table -- Original table |
452 |
-> Instance.Instance -- Instance to move |
453 |
-> Table -- Best new table for this instance |
454 |
checkInstanceMove nodes_idx ini_tbl target = |
455 |
let |
456 |
opdx = Instance.pnode target |
457 |
osdx = Instance.snode target |
458 |
nodes = filter (\idx -> idx /= opdx && idx /= osdx) nodes_idx |
459 |
use_secondary = elem osdx nodes_idx |
460 |
aft_failover = if use_secondary -- if allowed to failover |
461 |
then checkSingleStep ini_tbl target ini_tbl Failover |
462 |
else ini_tbl |
463 |
all_moves = concatMap (possibleMoves use_secondary) nodes |
464 |
in |
465 |
-- iterate over the possible nodes for this instance |
466 |
foldl' (checkSingleStep ini_tbl target) aft_failover all_moves |
467 |
|
468 |
-- | Compute the best next move. |
469 |
checkMove :: [Int] -- ^ Allowed target node indices |
470 |
-> Table -- ^ The current solution |
471 |
-> [Instance.Instance] -- ^ List of instances still to move |
472 |
-> Table -- ^ The new solution |
473 |
checkMove nodes_idx ini_tbl victims = |
474 |
let Table _ _ _ ini_plc = ini_tbl |
475 |
-- iterate over all instances, computing the best move |
476 |
best_tbl = |
477 |
foldl' |
478 |
(\ step_tbl elem -> compareTables step_tbl $ |
479 |
checkInstanceMove nodes_idx ini_tbl elem) |
480 |
ini_tbl victims |
481 |
Table _ _ _ best_plc = best_tbl |
482 |
in |
483 |
if length best_plc == length ini_plc then -- no advancement |
484 |
ini_tbl |
485 |
else |
486 |
best_tbl |
487 |
|
488 |
{- | Auxiliary function for solution computation. |
489 |
|
490 |
We write this in an explicit recursive fashion in order to control |
491 |
early-abort in case we have met the min delta. We can't use foldr |
492 |
instead of explicit recursion since we need the accumulator for the |
493 |
abort decision. |
494 |
|
495 |
-} |
496 |
advanceSolution :: [Maybe Removal] -- ^ The removal to process |
497 |
-> Int -- ^ Minimum delta parameter |
498 |
-> Int -- ^ Maximum delta parameter |
499 |
-> Maybe Solution -- ^ Current best solution |
500 |
-> Maybe Solution -- ^ New best solution |
501 |
advanceSolution [] _ _ sol = sol |
502 |
advanceSolution (Nothing:xs) m n sol = advanceSolution xs m n sol |
503 |
advanceSolution ((Just (Removal nx removed)):xs) min_d max_d prev_sol = |
504 |
let new_sol = checkPlacement nx removed [] 0 prev_sol max_d |
505 |
new_delta = solutionDelta $! new_sol |
506 |
in |
507 |
if new_delta >= 0 && new_delta <= min_d then |
508 |
new_sol |
509 |
else |
510 |
advanceSolution xs min_d max_d new_sol |
511 |
|
512 |
-- | Computes the placement solution. |
513 |
solutionFromRemovals :: [Maybe Removal] -- ^ The list of (possible) removals |
514 |
-> Int -- ^ Minimum delta parameter |
515 |
-> Int -- ^ Maximum delta parameter |
516 |
-> Maybe Solution -- ^ The best solution found |
517 |
solutionFromRemovals removals min_delta max_delta = |
518 |
advanceSolution removals min_delta max_delta Nothing |
519 |
|
520 |
{- | Computes the solution at the given depth. |
521 |
|
522 |
This is a wrapper over both computeRemovals and |
523 |
solutionFromRemovals. In case we have no solution, we return Nothing. |
524 |
|
525 |
-} |
526 |
computeSolution :: NodeList -- ^ The original node data |
527 |
-> [Instance.Instance] -- ^ The list of /bad/ instances |
528 |
-> Int -- ^ The /depth/ of removals |
529 |
-> Int -- ^ Maximum number of removals to process |
530 |
-> Int -- ^ Minimum delta parameter |
531 |
-> Int -- ^ Maximum delta parameter |
532 |
-> Maybe Solution -- ^ The best solution found (or Nothing) |
533 |
computeSolution nl bad_instances depth max_removals min_delta max_delta = |
534 |
let |
535 |
removals = computeRemovals nl bad_instances depth |
536 |
removals' = capRemovals removals max_removals |
537 |
in |
538 |
solutionFromRemovals removals' min_delta max_delta |
539 |
|
540 |
-- Solution display functions (pure) |
541 |
|
542 |
-- | Given the original and final nodes, computes the relocation description. |
543 |
computeMoves :: String -- ^ The instance name |
544 |
-> String -- ^ Original primary |
545 |
-> String -- ^ Original secondary |
546 |
-> String -- ^ New primary |
547 |
-> String -- ^ New secondary |
548 |
-> (String, [String]) |
549 |
-- ^ Tuple of moves and commands list; moves is containing |
550 |
-- either @/f/@ for failover or @/r:name/@ for replace |
551 |
-- secondary, while the command list holds gnt-instance |
552 |
-- commands (without that prefix), e.g \"@failover instance1@\" |
553 |
computeMoves i a b c d = |
554 |
if c == a then {- Same primary -} |
555 |
if d == b then {- Same sec??! -} |
556 |
("-", []) |
557 |
else {- Change of secondary -} |
558 |
(printf "r:%s" d, |
559 |
[printf "replace-disks -n %s %s" d i]) |
560 |
else |
561 |
if c == b then {- Failover and ... -} |
562 |
if d == a then {- that's all -} |
563 |
("f", [printf "migrate -f %s" i]) |
564 |
else |
565 |
(printf "f r:%s" d, |
566 |
[printf "migrate -f %s" i, |
567 |
printf "replace-disks -n %s %s" d i]) |
568 |
else |
569 |
if d == a then {- ... and keep primary as secondary -} |
570 |
(printf "r:%s f" c, |
571 |
[printf "replace-disks -n %s %s" c i, |
572 |
printf "migrate -f %s" i]) |
573 |
else |
574 |
if d == b then {- ... keep same secondary -} |
575 |
(printf "f r:%s f" c, |
576 |
[printf "migrate -f %s" i, |
577 |
printf "replace-disks -n %s %s" c i, |
578 |
printf "migrate -f %s" i]) |
579 |
|
580 |
else {- Nothing in common -} |
581 |
(printf "r:%s f r:%s" c d, |
582 |
[printf "replace-disks -n %s %s" c i, |
583 |
printf "migrate -f %s" i, |
584 |
printf "replace-disks -n %s %s" d i]) |
585 |
|
586 |
{-| Converts a placement to string format -} |
587 |
printSolutionLine :: InstanceList |
588 |
-> NameList |
589 |
-> NameList |
590 |
-> Int |
591 |
-> Int |
592 |
-> Placement |
593 |
-> Int |
594 |
-> (String, [String]) |
595 |
printSolutionLine il ktn kti nmlen imlen plc pos = |
596 |
let |
597 |
pmlen = (2*nmlen + 1) |
598 |
(i, p, s, c) = plc |
599 |
inst = Container.find i il |
600 |
inam = fromJust $ lookup (Instance.idx inst) kti |
601 |
npri = fromJust $ lookup p ktn |
602 |
nsec = fromJust $ lookup s ktn |
603 |
opri = fromJust $ lookup (Instance.pnode inst) ktn |
604 |
osec = fromJust $ lookup (Instance.snode inst) ktn |
605 |
(moves, cmds) = computeMoves inam opri osec npri nsec |
606 |
ostr = (printf "%s:%s" opri osec)::String |
607 |
nstr = (printf "%s:%s" npri nsec)::String |
608 |
in |
609 |
(printf " %3d. %-*s %-*s => %-*s %.8f a=%s" |
610 |
pos imlen inam pmlen ostr |
611 |
pmlen nstr c moves, |
612 |
cmds) |
613 |
|
614 |
formatCmds :: [[String]] -> String |
615 |
formatCmds cmd_strs = |
616 |
unlines $ |
617 |
concat $ map (\(a, b) -> |
618 |
(printf "echo step %d" (a::Int)): |
619 |
(printf "check"): |
620 |
(map ("gnt-instance " ++) b)) $ |
621 |
zip [1..] cmd_strs |
622 |
|
623 |
{-| Converts a solution to string format -} |
624 |
printSolution :: InstanceList |
625 |
-> NameList |
626 |
-> NameList |
627 |
-> [Placement] |
628 |
-> ([String], [[String]]) |
629 |
printSolution il ktn kti sol = |
630 |
let |
631 |
mlen_fn = maximum . (map length) . snd . unzip |
632 |
imlen = mlen_fn kti |
633 |
nmlen = mlen_fn ktn |
634 |
in |
635 |
unzip $ map (uncurry $ printSolutionLine il ktn kti nmlen imlen) $ |
636 |
zip sol [1..] |
637 |
|
638 |
-- | Print the node list. |
639 |
printNodes :: NameList -> NodeList -> String |
640 |
printNodes ktn nl = |
641 |
let snl = sortBy (compare `on` Node.idx) (Container.elems nl) |
642 |
snl' = map (\ n -> ((fromJust $ lookup (Node.idx n) ktn), n)) snl |
643 |
m_name = maximum . (map length) . fst . unzip $ snl' |
644 |
helper = Node.list m_name |
645 |
header = printf |
646 |
"%2s %-*s %5s %5s %5s %5s %5s %5s %5s %5s %3s %3s %7s %7s" |
647 |
" F" m_name "Name" |
648 |
"t_mem" "n_mem" "i_mem" "x_mem" "f_mem" "r_mem" |
649 |
"t_dsk" "f_dsk" |
650 |
"pri" "sec" "p_fmem" "p_fdsk" |
651 |
in unlines $ (header:map (uncurry helper) snl') |
652 |
|
653 |
-- | Compute the mem and disk covariance. |
654 |
compDetailedCV :: NodeList -> (Double, Double, Double, Double, Double) |
655 |
compDetailedCV nl = |
656 |
let |
657 |
all_nodes = Container.elems nl |
658 |
(offline, nodes) = partition Node.offline all_nodes |
659 |
mem_l = map Node.p_mem nodes |
660 |
dsk_l = map Node.p_dsk nodes |
661 |
mem_cv = varianceCoeff mem_l |
662 |
dsk_cv = varianceCoeff dsk_l |
663 |
n1_l = length $ filter Node.failN1 nodes |
664 |
n1_score = (fromIntegral n1_l) / (fromIntegral $ length nodes) |
665 |
res_l = map Node.p_rem nodes |
666 |
res_cv = varianceCoeff res_l |
667 |
offline_inst = sum . map (\n -> (length . Node.plist $ n) + |
668 |
(length . Node.slist $ n)) $ offline |
669 |
online_inst = sum . map (\n -> (length . Node.plist $ n) + |
670 |
(length . Node.slist $ n)) $ nodes |
671 |
off_score = (fromIntegral offline_inst) / |
672 |
(fromIntegral $ online_inst + offline_inst) |
673 |
in (mem_cv, dsk_cv, n1_score, res_cv, off_score) |
674 |
|
675 |
-- | Compute the 'total' variance. |
676 |
compCV :: NodeList -> Double |
677 |
compCV nl = |
678 |
let (mem_cv, dsk_cv, n1_score, res_cv, off_score) = compDetailedCV nl |
679 |
in mem_cv + dsk_cv + n1_score + res_cv + off_score |
680 |
|
681 |
printStats :: NodeList -> String |
682 |
printStats nl = |
683 |
let (mem_cv, dsk_cv, n1_score, res_cv, off_score) = compDetailedCV nl |
684 |
in printf "f_mem=%.8f, r_mem=%.8f, f_dsk=%.8f, n1=%.3f, uf=%.3f" |
685 |
mem_cv res_cv dsk_cv n1_score off_score |
686 |
|
687 |
-- Balancing functions |
688 |
|
689 |
-- Loading functions |
690 |
|
691 |
{- | Convert newline and delimiter-separated text. |
692 |
|
693 |
This function converts a text in tabular format as generated by |
694 |
@gnt-instance list@ and @gnt-node list@ to a list of objects using a |
695 |
supplied conversion function. |
696 |
|
697 |
-} |
698 |
loadTabular :: (Monad m) => String -> ([String] -> m (String, a)) |
699 |
-> (a -> Int -> a) -> m ([(String, Int)], [(Int, a)]) |
700 |
loadTabular text_data convert_fn set_fn = do |
701 |
let lines_data = lines text_data |
702 |
rows = map (sepSplit '|') lines_data |
703 |
kerows <- mapM convert_fn rows |
704 |
let idxrows = map (\ (idx, (k, v)) -> ((k, idx), (idx, set_fn v idx))) |
705 |
(zip [0..] kerows) |
706 |
return $ unzip idxrows |
707 |
|
708 |
-- | For each instance, add its index to its primary and secondary nodes |
709 |
fixNodes :: [(Int, Node.Node)] |
710 |
-> [(Int, Instance.Instance)] |
711 |
-> [(Int, Node.Node)] |
712 |
fixNodes nl il = |
713 |
foldl' (\accu (idx, inst) -> |
714 |
let |
715 |
assocEqual = (\ (i, _) (j, _) -> i == j) |
716 |
pdx = Instance.pnode inst |
717 |
sdx = Instance.snode inst |
718 |
pold = fromJust $ lookup pdx accu |
719 |
sold = fromJust $ lookup sdx accu |
720 |
pnew = Node.setPri pold idx |
721 |
snew = Node.setSec sold idx |
722 |
ac1 = deleteBy assocEqual (pdx, pold) accu |
723 |
ac2 = deleteBy assocEqual (sdx, sold) ac1 |
724 |
ac3 = (pdx, pnew):(sdx, snew):ac2 |
725 |
in ac3) nl il |
726 |
|
727 |
-- | Compute the longest common suffix of a NameList list that |
728 |
-- | starts with a dot |
729 |
longestDomain :: NameList -> String |
730 |
longestDomain [] = "" |
731 |
longestDomain ((_,x):xs) = |
732 |
let |
733 |
onlyStrings = snd $ unzip xs |
734 |
in |
735 |
foldr (\ suffix accu -> if all (isSuffixOf suffix) onlyStrings |
736 |
then suffix |
737 |
else accu) |
738 |
"" $ filter (isPrefixOf ".") (tails x) |
739 |
|
740 |
-- | Remove tails from the (Int, String) lists |
741 |
stripSuffix :: String -> NameList -> NameList |
742 |
stripSuffix suffix lst = |
743 |
let sflen = length suffix in |
744 |
map (\ (key, name) -> (key, take ((length name) - sflen) name)) lst |
745 |
|
746 |
-- | Lookups a node into an assoc list |
747 |
lookupNode :: (Monad m) => String -> String -> [(String, Int)] -> m Int |
748 |
lookupNode node inst ktn = |
749 |
case lookup node ktn of |
750 |
Nothing -> fail $ "Unknown node " ++ node ++ " for instance " ++ inst |
751 |
Just idx -> return idx |
752 |
|
753 |
{-| Initializer function that loads the data from a node and list file |
754 |
and massages it into the correct format. -} |
755 |
loadData :: String -- ^ Node data in text format |
756 |
-> String -- ^ Instance data in text format |
757 |
-> Result (Container.Container Node.Node, |
758 |
Container.Container Instance.Instance, |
759 |
String, NameList, NameList) |
760 |
loadData ndata idata = do |
761 |
{- node file: name t_mem n_mem f_mem t_disk f_disk -} |
762 |
(ktn, nl) <- loadTabular ndata |
763 |
(\ (name:tm:nm:fm:td:fd:fo:[]) -> |
764 |
return (name, |
765 |
if any (== "?") [tm,nm,fm,td,fd] || fo == "Y" then |
766 |
Node.create 0 0 0 0 0 True |
767 |
else |
768 |
Node.create (read tm) (read nm) (read fm) |
769 |
(read td) (read fd) False |
770 |
)) |
771 |
Node.setIdx |
772 |
{- instance file: name mem disk status pnode snode -} |
773 |
(kti, il) <- loadTabular idata |
774 |
(\ (name:mem:dsk:status:pnode:snode:[]) -> do |
775 |
pidx <- lookupNode pnode name ktn |
776 |
sidx <- lookupNode snode name ktn |
777 |
let newinst = Instance.create (read mem) (read dsk) |
778 |
status pidx sidx |
779 |
return (name, newinst) |
780 |
) |
781 |
Instance.setIdx |
782 |
let |
783 |
nl2 = fixNodes nl il |
784 |
il3 = Container.fromAssocList il |
785 |
nl3 = Container.fromAssocList |
786 |
(map (\ (k, v) -> (k, Node.buildPeers v il3 (length nl2))) nl2) |
787 |
xtn = swapPairs ktn |
788 |
xti = swapPairs kti |
789 |
common_suffix = longestDomain (xti ++ xtn) |
790 |
stn = stripSuffix common_suffix xtn |
791 |
sti = stripSuffix common_suffix xti |
792 |
return (nl3, il3, common_suffix, stn, sti) |
793 |
|
794 |
-- | Compute the amount of memory used by primary instances on a node. |
795 |
nodeImem :: Node.Node -> InstanceList -> Int |
796 |
nodeImem node il = |
797 |
let rfind = flip Container.find $ il |
798 |
in sum . map Instance.mem . |
799 |
map rfind $ Node.plist node |
800 |
|
801 |
-- | Compute the amount of disk used by instances on a node (either primary |
802 |
-- or secondary). |
803 |
nodeIdsk :: Node.Node -> InstanceList -> Int |
804 |
nodeIdsk node il = |
805 |
let rfind = flip Container.find $ il |
806 |
in sum . map Instance.dsk . |
807 |
map rfind $ (Node.plist node) ++ (Node.slist node) |
808 |
|
809 |
|
810 |
-- | Check cluster data for consistency |
811 |
checkData :: NodeList -> InstanceList -> NameList -> NameList |
812 |
-> ([String], NodeList) |
813 |
checkData nl il ktn _ = |
814 |
Container.mapAccum |
815 |
(\ msgs node -> |
816 |
let nname = fromJust $ lookup (Node.idx node) ktn |
817 |
nilst = map (flip Container.find $ il) (Node.plist node) |
818 |
dilst = filter (not . Instance.running) nilst |
819 |
adj_mem = sum . map Instance.mem $ dilst |
820 |
delta_mem = (truncate $ Node.t_mem node) |
821 |
- (Node.n_mem node) |
822 |
- (Node.f_mem node) |
823 |
- (nodeImem node il) |
824 |
+ adj_mem |
825 |
delta_dsk = (truncate $ Node.t_dsk node) |
826 |
- (Node.f_dsk node) |
827 |
- (nodeIdsk node il) |
828 |
newn = Node.setFmem (Node.setXmem node delta_mem) |
829 |
(Node.f_mem node - adj_mem) |
830 |
umsg1 = if delta_mem > 512 || delta_dsk > 1024 |
831 |
then [printf "node %s is missing %d MB ram \ |
832 |
\and %d GB disk" |
833 |
nname delta_mem (delta_dsk `div` 1024)] |
834 |
else [] |
835 |
in (msgs ++ umsg1, newn) |
836 |
) [] nl |