subnet.go 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512
  1. // Copyright 2015 CoreOS, Inc.
  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 subnet
  15. import (
  16. "encoding/json"
  17. "errors"
  18. "fmt"
  19. "net"
  20. "regexp"
  21. "strconv"
  22. "time"
  23. "github.com/coreos/flannel/Godeps/_workspace/src/github.com/coreos/go-etcd/etcd"
  24. log "github.com/coreos/flannel/Godeps/_workspace/src/github.com/golang/glog"
  25. "github.com/coreos/flannel/pkg/ip"
  26. "github.com/coreos/flannel/pkg/task"
  27. )
  28. const (
  29. registerRetries = 10
  30. subnetTTL = 24 * 3600
  31. renewMargin = time.Hour
  32. )
  33. // etcd error codes
  34. const (
  35. etcdKeyNotFound = 100
  36. etcdKeyAlreadyExists = 105
  37. etcdEventIndexCleared = 401
  38. )
  39. const (
  40. SubnetAdded = iota
  41. SubnetRemoved
  42. SubnetListed
  43. )
  44. var (
  45. subnetRegex *regexp.Regexp = regexp.MustCompile(`(\d+\.\d+.\d+.\d+)-(\d+)`)
  46. )
  47. type LeaseAttrs struct {
  48. PublicIP ip.IP4
  49. BackendType string `json:",omitempty"`
  50. BackendData json.RawMessage `json:",omitempty"`
  51. }
  52. type SubnetLease struct {
  53. Network ip.IP4Net
  54. Attrs LeaseAttrs
  55. }
  56. type SubnetManager struct {
  57. registry subnetRegistry
  58. config *Config
  59. myLease SubnetLease
  60. leaseExp time.Time
  61. lastIndex uint64
  62. leases []SubnetLease
  63. }
  64. type EventType int
  65. type Event struct {
  66. Type EventType
  67. Lease SubnetLease
  68. }
  69. type EventBatch []Event
  70. func NewSubnetManager(config *EtcdConfig) (*SubnetManager, error) {
  71. esr, err := newEtcdSubnetRegistry(config)
  72. if err != nil {
  73. return nil, err
  74. }
  75. return newSubnetManager(esr)
  76. }
  77. func (sm *SubnetManager) AcquireLease(attrs *LeaseAttrs, cancel chan bool) (ip.IP4Net, error) {
  78. for {
  79. sn, err := sm.acquireLeaseOnce(attrs, cancel)
  80. switch {
  81. case err == nil:
  82. log.Info("Subnet lease acquired: ", sn)
  83. return sn, nil
  84. case err == task.ErrCanceled:
  85. return ip.IP4Net{}, err
  86. default:
  87. log.Error("Failed to acquire subnet: ", err)
  88. }
  89. select {
  90. case <-time.After(time.Second):
  91. case <-cancel:
  92. return ip.IP4Net{}, task.ErrCanceled
  93. }
  94. }
  95. }
  96. func findLeaseByIP(leases []SubnetLease, pubIP ip.IP4) *SubnetLease {
  97. for _, l := range leases {
  98. if pubIP == l.Attrs.PublicIP {
  99. return &l
  100. }
  101. }
  102. return nil
  103. }
  104. func (sm *SubnetManager) tryAcquireLease(extIP ip.IP4, attrs *LeaseAttrs) (ip.IP4Net, error) {
  105. var err error
  106. sm.leases, err = sm.getLeases()
  107. if err != nil {
  108. return ip.IP4Net{}, err
  109. }
  110. attrBytes, err := json.Marshal(attrs)
  111. if err != nil {
  112. log.Errorf("marshal failed: %#v, %v", attrs, err)
  113. return ip.IP4Net{}, err
  114. }
  115. // try to reuse a subnet if there's one that matches our IP
  116. if l := findLeaseByIP(sm.leases, extIP); l != nil {
  117. resp, err := sm.registry.updateSubnet(l.Network.StringSep(".", "-"), string(attrBytes), subnetTTL)
  118. if err != nil {
  119. return ip.IP4Net{}, err
  120. }
  121. sm.myLease.Network = l.Network
  122. sm.myLease.Attrs = *attrs
  123. sm.leaseExp = *resp.Node.Expiration
  124. return l.Network, nil
  125. }
  126. // no existing match, grab a new one
  127. sn, err := sm.allocateSubnet()
  128. if err != nil {
  129. return ip.IP4Net{}, err
  130. }
  131. resp, err := sm.registry.createSubnet(sn.StringSep(".", "-"), string(attrBytes), subnetTTL)
  132. switch {
  133. case err == nil:
  134. sm.myLease.Network = sn
  135. sm.myLease.Attrs = *attrs
  136. sm.leaseExp = *resp.Node.Expiration
  137. return sn, nil
  138. // if etcd returned Key Already Exists, try again.
  139. case err.(*etcd.EtcdError).ErrorCode == etcdKeyAlreadyExists:
  140. return ip.IP4Net{}, nil
  141. default:
  142. return ip.IP4Net{}, err
  143. }
  144. }
  145. func (sm *SubnetManager) acquireLeaseOnce(attrs *LeaseAttrs, cancel chan bool) (ip.IP4Net, error) {
  146. for i := 0; i < registerRetries; i++ {
  147. sn, err := sm.tryAcquireLease(attrs.PublicIP, attrs)
  148. switch {
  149. case err != nil:
  150. return ip.IP4Net{}, err
  151. case sn.IP != 0:
  152. return sn, nil
  153. }
  154. // before moving on, check for cancel
  155. if interrupted(cancel) {
  156. return ip.IP4Net{}, task.ErrCanceled
  157. }
  158. }
  159. return ip.IP4Net{}, errors.New("Max retries reached trying to acquire a subnet")
  160. }
  161. func (sm *SubnetManager) GetConfig() *Config {
  162. return sm.config
  163. }
  164. /// Implementation
  165. func parseSubnetKey(s string) (ip.IP4Net, error) {
  166. if parts := subnetRegex.FindStringSubmatch(s); len(parts) == 3 {
  167. snIp := net.ParseIP(parts[1]).To4()
  168. prefixLen, err := strconv.ParseUint(parts[2], 10, 5)
  169. if snIp != nil && err == nil {
  170. return ip.IP4Net{IP: ip.FromIP(snIp), PrefixLen: uint(prefixLen)}, nil
  171. }
  172. }
  173. return ip.IP4Net{}, errors.New("Error parsing IP Subnet")
  174. }
  175. func newSubnetManager(r subnetRegistry) (*SubnetManager, error) {
  176. cfgResp, err := r.getConfig()
  177. if err != nil {
  178. return nil, err
  179. }
  180. cfg, err := ParseConfig(cfgResp.Node.Value)
  181. if err != nil {
  182. return nil, err
  183. }
  184. sm := SubnetManager{
  185. registry: r,
  186. config: cfg,
  187. }
  188. return &sm, nil
  189. }
  190. func (sm *SubnetManager) getLeases() ([]SubnetLease, error) {
  191. resp, err := sm.registry.getSubnets()
  192. var leases []SubnetLease
  193. switch {
  194. case err == nil:
  195. for _, node := range resp.Node.Nodes {
  196. sn, err := parseSubnetKey(node.Key)
  197. if err == nil {
  198. var attrs LeaseAttrs
  199. if err = json.Unmarshal([]byte(node.Value), &attrs); err == nil {
  200. lease := SubnetLease{sn, attrs}
  201. leases = append(leases, lease)
  202. }
  203. }
  204. }
  205. sm.lastIndex = resp.EtcdIndex
  206. case err.(*etcd.EtcdError).ErrorCode == etcdKeyNotFound:
  207. // key not found: treat it as empty set
  208. sm.lastIndex = err.(*etcd.EtcdError).Index
  209. default:
  210. return nil, err
  211. }
  212. return leases, nil
  213. }
  214. func deleteLease(l []SubnetLease, i int) []SubnetLease {
  215. l[i], l = l[len(l)-1], l[:len(l)-1]
  216. return l
  217. }
  218. func (sm *SubnetManager) applyLeases(newLeases []SubnetLease) EventBatch {
  219. var batch EventBatch
  220. for _, l := range newLeases {
  221. // skip self
  222. if l.Network.Equal(sm.myLease.Network) {
  223. continue
  224. }
  225. found := false
  226. for i, c := range sm.leases {
  227. if c.Network.Equal(l.Network) {
  228. sm.leases = deleteLease(sm.leases, i)
  229. found = true
  230. break
  231. }
  232. }
  233. if !found {
  234. // new subnet
  235. batch = append(batch, Event{SubnetAdded, l})
  236. }
  237. }
  238. // everything left in sm.leases has been deleted
  239. for _, c := range sm.leases {
  240. batch = append(batch, Event{SubnetRemoved, c})
  241. }
  242. sm.leases = newLeases
  243. return batch
  244. }
  245. func (sm *SubnetManager) applySubnetChange(action string, ipn ip.IP4Net, data string) (Event, error) {
  246. switch action {
  247. case "delete", "expire":
  248. for i, l := range sm.leases {
  249. if l.Network.Equal(ipn) {
  250. deleteLease(sm.leases, i)
  251. return Event{SubnetRemoved, l}, nil
  252. }
  253. }
  254. log.Errorf("Removed subnet (%s) was not found", ipn)
  255. return Event{
  256. SubnetRemoved,
  257. SubnetLease{ipn, LeaseAttrs{}},
  258. }, nil
  259. default:
  260. var attrs LeaseAttrs
  261. err := json.Unmarshal([]byte(data), &attrs)
  262. if err != nil {
  263. return Event{}, err
  264. }
  265. for i, l := range sm.leases {
  266. if l.Network.Equal(ipn) {
  267. sm.leases[i] = SubnetLease{ipn, attrs}
  268. return Event{SubnetAdded, sm.leases[i]}, nil
  269. }
  270. }
  271. sm.leases = append(sm.leases, SubnetLease{ipn, attrs})
  272. return Event{SubnetAdded, sm.leases[len(sm.leases)-1]}, nil
  273. }
  274. }
  275. func (sm *SubnetManager) allocateSubnet() (ip.IP4Net, error) {
  276. log.Infof("Picking subnet in range %s ... %s", sm.config.SubnetMin, sm.config.SubnetMax)
  277. var bag []ip.IP4
  278. sn := ip.IP4Net{IP: sm.config.SubnetMin, PrefixLen: sm.config.SubnetLen}
  279. OuterLoop:
  280. for ; sn.IP <= sm.config.SubnetMax && len(bag) < 100; sn = sn.Next() {
  281. for _, l := range sm.leases {
  282. if sn.Overlaps(l.Network) {
  283. continue OuterLoop
  284. }
  285. }
  286. bag = append(bag, sn.IP)
  287. }
  288. if len(bag) == 0 {
  289. return ip.IP4Net{}, errors.New("out of subnets")
  290. } else {
  291. i := randInt(0, len(bag))
  292. return ip.IP4Net{IP: bag[i], PrefixLen: sm.config.SubnetLen}, nil
  293. }
  294. }
  295. func (sm *SubnetManager) WatchLeases(receiver chan EventBatch, cancel chan bool) {
  296. // "catch up" by replaying all the leases we discovered during
  297. // AcquireLease
  298. var batch EventBatch
  299. for _, l := range sm.leases {
  300. if !sm.myLease.Network.Equal(l.Network) {
  301. batch = append(batch, Event{SubnetAdded, l})
  302. }
  303. }
  304. if len(batch) > 0 {
  305. receiver <- batch
  306. }
  307. for {
  308. resp, err := sm.registry.watchSubnets(sm.lastIndex+1, cancel)
  309. // watchSubnets exited by cancel chan being signaled
  310. if err == nil && resp == nil {
  311. return
  312. }
  313. var batch *EventBatch
  314. if err == nil {
  315. batch, err = sm.parseSubnetWatchResponse(resp)
  316. } else {
  317. batch, err = sm.parseSubnetWatchError(err)
  318. }
  319. if err != nil {
  320. log.Errorf("%v", err)
  321. time.Sleep(time.Second)
  322. continue
  323. }
  324. if batch != nil {
  325. receiver <- *batch
  326. }
  327. }
  328. }
  329. func (sm *SubnetManager) ListLeases(receiver chan EventBatch, cancel chan bool) {
  330. //periodly list leases
  331. for {
  332. log.Info("begin to colletc leases")
  333. resp, err := sm.registry.watchSubnets(sm.lastIndex+1, cancel)
  334. // watchSubnets exited by cancel chan being signaled
  335. if err == nil && resp == nil {
  336. return
  337. }
  338. leases, err := sm.getLeases()
  339. if err == nil {
  340. var batch EventBatch
  341. for _, l := range leases {
  342. // skip self
  343. if l.Network.Equal(sm.myLease.Network) {
  344. continue
  345. }
  346. batch = append(batch, Event{SubnetListed, l})
  347. }
  348. log.Info("finish collection leases")
  349. if &batch != nil {
  350. receiver <- batch
  351. }
  352. log.Info("finish passing")
  353. }
  354. time.Sleep(10 * time.Second)
  355. }
  356. }
  357. func (sm *SubnetManager) parseSubnetWatchResponse(resp *etcd.Response) (batch *EventBatch, err error) {
  358. sm.lastIndex = resp.Node.ModifiedIndex
  359. sn, err := parseSubnetKey(resp.Node.Key)
  360. if err != nil {
  361. err = fmt.Errorf("Error parsing subnet IP: %s", resp.Node.Key)
  362. return
  363. }
  364. // Don't process our own changes
  365. if !sm.myLease.Network.Equal(sn) {
  366. evt, err := sm.applySubnetChange(resp.Action, sn, resp.Node.Value)
  367. if err != nil {
  368. return nil, err
  369. }
  370. batch = &EventBatch{evt}
  371. }
  372. return
  373. }
  374. func (sm *SubnetManager) parseSubnetWatchError(err error) (batch *EventBatch, out error) {
  375. etcdErr, ok := err.(*etcd.EtcdError)
  376. if ok && etcdErr.ErrorCode == etcdEventIndexCleared {
  377. // etcd maintains a history window for events and it's possible to fall behind.
  378. // to recover, get the current state and then "diff" against our cache to generate
  379. // events for the caller
  380. log.Warning("Watch of subnet leases failed because etcd index outside history window")
  381. leases, err := sm.getLeases()
  382. if err == nil {
  383. lb := sm.applyLeases(leases)
  384. batch = &lb
  385. } else {
  386. out = fmt.Errorf("Failed to retrieve subnet leases: %v", err)
  387. }
  388. } else {
  389. out = fmt.Errorf("Watch of subnet leases failed: %v", err)
  390. }
  391. return
  392. }
  393. func (sm *SubnetManager) LeaseRenewer(cancel chan bool) {
  394. for {
  395. dur := sm.leaseExp.Sub(time.Now()) - renewMargin
  396. select {
  397. case <-time.After(dur):
  398. attrBytes, err := json.Marshal(&sm.myLease.Attrs)
  399. if err != nil {
  400. log.Error("Error renewing lease (trying again in 1 min): ", err)
  401. dur = time.Minute
  402. continue
  403. }
  404. resp, err := sm.registry.updateSubnet(sm.myLease.Network.StringSep(".", "-"), string(attrBytes), subnetTTL)
  405. if err != nil {
  406. log.Error("Error renewing lease (trying again in 1 min): ", err)
  407. dur = time.Minute
  408. continue
  409. }
  410. sm.leaseExp = *resp.Node.Expiration
  411. log.Info("Lease renewed, new expiration: ", sm.leaseExp)
  412. dur = sm.leaseExp.Sub(time.Now()) - renewMargin
  413. case <-cancel:
  414. return
  415. }
  416. }
  417. }
  418. func interrupted(cancel chan bool) bool {
  419. select {
  420. case <-cancel:
  421. return true
  422. default:
  423. return false
  424. }
  425. }