tester.go 7.5 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280
  1. // Copyright 2015 The etcd Authors
  2. //
  3. // Licensed under the Apache License, Version 2.0 (the "License");
  4. // you may not use this file except in compliance with the License.
  5. // You may obtain a copy of the License at
  6. //
  7. // http://www.apache.org/licenses/LICENSE-2.0
  8. //
  9. // Unless required by applicable law or agreed to in writing, software
  10. // distributed under the License is distributed on an "AS IS" BASIS,
  11. // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  12. // See the License for the specific language governing permissions and
  13. // limitations under the License.
  14. package main
  15. import (
  16. "fmt"
  17. "time"
  18. )
  19. type tester struct {
  20. failures []failure
  21. cluster *cluster
  22. limit int
  23. status Status
  24. currentRevision int64
  25. Stressers []Stresser
  26. stressBuilder stressBuilder
  27. leaseStresserBuilder leaseStresserBuilder
  28. Checker Checker
  29. consistencyCheck bool
  30. }
  31. // compactQPS is rough number of compact requests per second.
  32. // Previous tests showed etcd can compact about 60,000 entries per second.
  33. const compactQPS = 50000
  34. func (tt *tester) runLoop() {
  35. tt.status.Since = time.Now()
  36. tt.status.RoundLimit = tt.limit
  37. tt.status.cluster = tt.cluster
  38. for _, f := range tt.failures {
  39. tt.status.Failures = append(tt.status.Failures, f.Desc())
  40. }
  41. tt.setupStressers()
  42. var prevCompactRev int64
  43. for round := 0; round < tt.limit || tt.limit == -1; round++ {
  44. tt.status.setRound(round)
  45. roundTotalCounter.Inc()
  46. if err := tt.doRound(round); err != nil {
  47. plog.Warningf("%s functional-tester returning with error (%v)", tt.logPrefix(), err)
  48. if tt.cleanup() != nil {
  49. return
  50. }
  51. prevCompactRev = 0 // reset after clean up
  52. continue
  53. }
  54. // -1 so that logPrefix doesn't print out 'case'
  55. tt.status.setCase(-1)
  56. revToCompact := max(0, tt.currentRevision-10000)
  57. compactN := revToCompact - prevCompactRev
  58. timeout := 10 * time.Second
  59. if compactN > 0 {
  60. timeout += time.Duration(compactN/compactQPS) * time.Second
  61. }
  62. prevCompactRev = revToCompact
  63. plog.Printf("%s compacting %d entries (timeout %v)", tt.logPrefix(), compactN, timeout)
  64. if err := tt.compact(revToCompact, timeout); err != nil {
  65. plog.Warningf("%s functional-tester compact got error (%v)", tt.logPrefix(), err)
  66. if tt.cleanup() != nil {
  67. return
  68. }
  69. prevCompactRev = 0 // reset after clean up
  70. }
  71. if round > 0 && round%500 == 0 { // every 500 rounds
  72. if err := tt.defrag(); err != nil {
  73. plog.Warningf("%s functional-tester returning with error (%v)", tt.logPrefix(), err)
  74. return
  75. }
  76. }
  77. }
  78. plog.Printf("%s functional-tester is finished", tt.logPrefix())
  79. }
  80. func (tt *tester) setupStressers() {
  81. tt.Stressers = make([]Stresser, 0)
  82. leaseStressers := make([]Stresser, 0, 2*len(tt.cluster.Members))
  83. for i, m := range tt.cluster.Members {
  84. lStresser := tt.leaseStresserBuilder(m)
  85. leaseStressers[i] = lStresser
  86. tt.Stressers = append(tt.Stressers, tt.stressBuilder(m), lStresser)
  87. }
  88. for i := range tt.Stressers {
  89. go tt.Stressers[i].Stress()
  90. }
  91. if !tt.consistencyCheck || tt.cluster.v2Only {
  92. tt.Checker = newNoChecker()
  93. return
  94. }
  95. tt.Checker = newCompositeChecker([]Checker{
  96. newHashChecker(hashAndRevGetter(tt.cluster)),
  97. newLeaseChecker(leaseStressers)},
  98. )
  99. }
  100. func (tt *tester) doRound(round int) error {
  101. for j, f := range tt.failures {
  102. caseTotalCounter.WithLabelValues(f.Desc()).Inc()
  103. tt.status.setCase(j)
  104. if err := tt.cluster.WaitHealth(); err != nil {
  105. return fmt.Errorf("wait full health error: %v", err)
  106. }
  107. plog.Printf("%s injecting failure %q", tt.logPrefix(), f.Desc())
  108. if err := f.Inject(tt.cluster, round); err != nil {
  109. return fmt.Errorf("injection error: %v", err)
  110. }
  111. plog.Printf("%s injected failure", tt.logPrefix())
  112. plog.Printf("%s recovering failure %q", tt.logPrefix(), f.Desc())
  113. if err := f.Recover(tt.cluster, round); err != nil {
  114. return fmt.Errorf("recovery error: %v", err)
  115. }
  116. plog.Printf("%s wait until cluster is healthy", tt.logPrefix())
  117. if err := tt.cluster.WaitHealth(); err != nil {
  118. return fmt.Errorf("wait full health error: %v", err)
  119. }
  120. plog.Printf("%s recovered failure", tt.logPrefix())
  121. if err := tt.checkConsistency(); err != nil {
  122. return fmt.Errorf("tt.checkConsistency error (%v)", err)
  123. }
  124. plog.Printf("%s succeed!", tt.logPrefix())
  125. }
  126. return nil
  127. }
  128. func (tt *tester) updateRevision() error {
  129. if tt.cluster.v2Only {
  130. return nil
  131. }
  132. revs, _, err := tt.cluster.getRevisionHash()
  133. for _, rev := range revs {
  134. tt.currentRevision = rev
  135. break // just need get one of the current revisions
  136. }
  137. plog.Printf("%s updated current revision to %d", tt.logPrefix(), tt.currentRevision)
  138. return err
  139. }
  140. func (tt *tester) checkConsistency() (err error) {
  141. tt.cancelStressers()
  142. defer func() {
  143. if err != nil {
  144. return
  145. }
  146. if err = tt.updateRevision(); err != nil {
  147. plog.Warningf("%s functional-tester returning with tt.updateRevision error (%v)", tt.logPrefix(), err)
  148. return
  149. }
  150. err = tt.startStressers()
  151. }()
  152. if err = tt.Checker.Check(); err != nil {
  153. plog.Printf("%s %v", tt.logPrefix(), err)
  154. }
  155. return err
  156. }
  157. func (tt *tester) compact(rev int64, timeout time.Duration) (err error) {
  158. tt.cancelStressers()
  159. defer func() {
  160. if err == nil {
  161. err = tt.startStressers()
  162. }
  163. }()
  164. plog.Printf("%s compacting storage (current revision %d, compact revision %d)", tt.logPrefix(), tt.currentRevision, rev)
  165. if err = tt.cluster.compactKV(rev, timeout); err != nil {
  166. return err
  167. }
  168. plog.Printf("%s compacted storage (compact revision %d)", tt.logPrefix(), rev)
  169. plog.Printf("%s checking compaction (compact revision %d)", tt.logPrefix(), rev)
  170. if err = tt.cluster.checkCompact(rev); err != nil {
  171. plog.Warningf("%s checkCompact error (%v)", tt.logPrefix(), err)
  172. return err
  173. }
  174. plog.Printf("%s confirmed compaction (compact revision %d)", tt.logPrefix(), rev)
  175. return nil
  176. }
  177. func (tt *tester) defrag() error {
  178. plog.Printf("%s defragmenting...", tt.logPrefix())
  179. if err := tt.cluster.defrag(); err != nil {
  180. plog.Warningf("%s defrag error (%v)", tt.logPrefix(), err)
  181. if cerr := tt.cleanup(); cerr != nil {
  182. return fmt.Errorf("%s, %s", err, cerr)
  183. }
  184. return err
  185. }
  186. plog.Printf("%s defragmented...", tt.logPrefix())
  187. return nil
  188. }
  189. func (tt *tester) logPrefix() string {
  190. var (
  191. rd = tt.status.getRound()
  192. cs = tt.status.getCase()
  193. prefix = fmt.Sprintf("[round#%d case#%d]", rd, cs)
  194. )
  195. if cs == -1 {
  196. prefix = fmt.Sprintf("[round#%d]", rd)
  197. }
  198. return prefix
  199. }
  200. func (tt *tester) cleanup() error {
  201. roundFailedTotalCounter.Inc()
  202. desc := "compact/defrag"
  203. if tt.status.Case != -1 {
  204. desc = tt.failures[tt.status.Case].Desc()
  205. }
  206. caseFailedTotalCounter.WithLabelValues(desc).Inc()
  207. tt.cancelStressers()
  208. plog.Printf("%s cleaning up...", tt.logPrefix())
  209. if err := tt.cluster.Cleanup(); err != nil {
  210. plog.Warningf("%s cleanup error: %v", tt.logPrefix(), err)
  211. return err
  212. }
  213. if err := tt.cluster.Reset(); err != nil {
  214. plog.Warningf("%s cleanup Bootstrap error: %v", tt.logPrefix(), err)
  215. return err
  216. }
  217. return nil
  218. }
  219. func (tt *tester) cancelStressers() {
  220. plog.Printf("%s canceling the stressers...", tt.logPrefix())
  221. for _, s := range tt.Stressers {
  222. s.Cancel()
  223. }
  224. plog.Printf("%s canceled stressers", tt.logPrefix())
  225. }
  226. func (tt *tester) startStressers() error {
  227. plog.Printf("%s starting the stressers...", tt.logPrefix())
  228. for _, s := range tt.Stressers {
  229. if err := s.Stress(); err != nil {
  230. return err
  231. }
  232. }
  233. plog.Printf("%s started stressers", tt.logPrefix())
  234. return nil
  235. }
  236. func (tt *tester) Report() (success, failure int) {
  237. for _, stresser := range tt.Stressers {
  238. s, f := stresser.Report()
  239. success += s
  240. failure += f
  241. }
  242. return
  243. }