handler.go 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545
  1. // Copyright 2018 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 agent
  15. import (
  16. "errors"
  17. "fmt"
  18. "io/ioutil"
  19. "net/url"
  20. "os"
  21. "os/exec"
  22. "path/filepath"
  23. "syscall"
  24. "time"
  25. "github.com/coreos/etcd/pkg/fileutil"
  26. "github.com/coreos/etcd/pkg/proxy"
  27. "github.com/coreos/etcd/tools/functional-tester/rpcpb"
  28. "go.uber.org/zap"
  29. )
  30. // return error for system errors (e.g. fail to create files)
  31. // return status error in response for wrong configuration/operation (e.g. start etcd twice)
  32. func (srv *Server) handleTesterRequest(req *rpcpb.Request) (resp *rpcpb.Response, err error) {
  33. defer func() {
  34. if err == nil {
  35. srv.last = req.Operation
  36. srv.lg.Info("handler success", zap.String("operation", req.Operation.String()))
  37. }
  38. }()
  39. switch req.Operation {
  40. case rpcpb.Operation_InitialStartEtcd:
  41. return srv.handleInitialStartEtcd(req)
  42. case rpcpb.Operation_RestartEtcd:
  43. return srv.handleRestartEtcd()
  44. case rpcpb.Operation_KillEtcd:
  45. return srv.handleKillEtcd()
  46. case rpcpb.Operation_FailArchive:
  47. return srv.handleFailArchive()
  48. case rpcpb.Operation_DestroyEtcdAgent:
  49. return srv.handleDestroyEtcdAgent()
  50. case rpcpb.Operation_BlackholePeerPortTxRx:
  51. return srv.handleBlackholePeerPortTxRx()
  52. case rpcpb.Operation_UnblackholePeerPortTxRx:
  53. return srv.handleUnblackholePeerPortTxRx()
  54. case rpcpb.Operation_DelayPeerPortTxRx:
  55. return srv.handleDelayPeerPortTxRx()
  56. case rpcpb.Operation_UndelayPeerPortTxRx:
  57. return srv.handleUndelayPeerPortTxRx()
  58. default:
  59. msg := fmt.Sprintf("operation not found (%v)", req.Operation)
  60. return &rpcpb.Response{Success: false, Status: msg}, errors.New(msg)
  61. }
  62. }
  63. func (srv *Server) handleInitialStartEtcd(req *rpcpb.Request) (*rpcpb.Response, error) {
  64. if srv.last != rpcpb.Operation_NotStarted {
  65. return &rpcpb.Response{
  66. Success: false,
  67. Status: fmt.Sprintf("%q is not valid; last server operation was %q", rpcpb.Operation_InitialStartEtcd.String(), srv.last.String()),
  68. Member: req.Member,
  69. }, nil
  70. }
  71. srv.Member = req.Member
  72. srv.Tester = req.Tester
  73. err := fileutil.TouchDirAll(srv.Member.BaseDir)
  74. if err != nil {
  75. return nil, err
  76. }
  77. srv.lg.Info("created base directory", zap.String("path", srv.Member.BaseDir))
  78. if err = srv.saveEtcdLogFile(); err != nil {
  79. return nil, err
  80. }
  81. srv.creatEtcdCmd()
  82. if err = srv.saveTLSAssets(); err != nil {
  83. return nil, err
  84. }
  85. if err = srv.startEtcdCmd(); err != nil {
  86. return nil, err
  87. }
  88. srv.lg.Info("started etcd", zap.String("command-path", srv.etcdCmd.Path))
  89. if err = srv.loadAutoTLSAssets(); err != nil {
  90. return nil, err
  91. }
  92. // wait some time for etcd listener start
  93. // before setting up proxy
  94. time.Sleep(time.Second)
  95. if err = srv.startProxy(); err != nil {
  96. return nil, err
  97. }
  98. return &rpcpb.Response{
  99. Success: true,
  100. Status: "start etcd PASS",
  101. Member: srv.Member,
  102. }, nil
  103. }
  104. // TODO: support TLS
  105. func (srv *Server) startProxy() error {
  106. if srv.Member.EtcdClientProxy {
  107. advertiseClientURL, advertiseClientURLPort, err := getURLAndPort(srv.Member.Etcd.AdvertiseClientURLs[0])
  108. if err != nil {
  109. return err
  110. }
  111. listenClientURL, _, err := getURLAndPort(srv.Member.Etcd.ListenClientURLs[0])
  112. if err != nil {
  113. return err
  114. }
  115. srv.advertiseClientPortToProxy[advertiseClientURLPort] = proxy.NewServer(proxy.ServerConfig{
  116. Logger: srv.lg,
  117. From: *advertiseClientURL,
  118. To: *listenClientURL,
  119. })
  120. select {
  121. case err = <-srv.advertiseClientPortToProxy[advertiseClientURLPort].Error():
  122. return err
  123. case <-time.After(2 * time.Second):
  124. srv.lg.Info("started proxy on client traffic", zap.String("url", advertiseClientURL.String()))
  125. }
  126. }
  127. if srv.Member.EtcdPeerProxy {
  128. advertisePeerURL, advertisePeerURLPort, err := getURLAndPort(srv.Member.Etcd.AdvertisePeerURLs[0])
  129. if err != nil {
  130. return err
  131. }
  132. listenPeerURL, _, err := getURLAndPort(srv.Member.Etcd.ListenPeerURLs[0])
  133. if err != nil {
  134. return err
  135. }
  136. srv.advertisePeerPortToProxy[advertisePeerURLPort] = proxy.NewServer(proxy.ServerConfig{
  137. Logger: srv.lg,
  138. From: *advertisePeerURL,
  139. To: *listenPeerURL,
  140. })
  141. select {
  142. case err = <-srv.advertisePeerPortToProxy[advertisePeerURLPort].Error():
  143. return err
  144. case <-time.After(2 * time.Second):
  145. srv.lg.Info("started proxy on peer traffic", zap.String("url", advertisePeerURL.String()))
  146. }
  147. }
  148. return nil
  149. }
  150. func (srv *Server) stopProxy() {
  151. if srv.Member.EtcdClientProxy && len(srv.advertiseClientPortToProxy) > 0 {
  152. for port, px := range srv.advertiseClientPortToProxy {
  153. if err := px.Close(); err != nil {
  154. srv.lg.Warn("failed to close proxy", zap.Int("port", port))
  155. continue
  156. }
  157. select {
  158. case <-px.Done():
  159. // enough time to release port
  160. time.Sleep(time.Second)
  161. case <-time.After(time.Second):
  162. }
  163. srv.lg.Info("closed proxy",
  164. zap.Int("port", port),
  165. zap.String("from", px.From()),
  166. zap.String("to", px.To()),
  167. )
  168. }
  169. srv.advertiseClientPortToProxy = make(map[int]proxy.Server)
  170. }
  171. if srv.Member.EtcdPeerProxy && len(srv.advertisePeerPortToProxy) > 0 {
  172. for port, px := range srv.advertisePeerPortToProxy {
  173. if err := px.Close(); err != nil {
  174. srv.lg.Warn("failed to close proxy", zap.Int("port", port))
  175. continue
  176. }
  177. select {
  178. case <-px.Done():
  179. // enough time to release port
  180. time.Sleep(time.Second)
  181. case <-time.After(time.Second):
  182. }
  183. srv.lg.Info("closed proxy",
  184. zap.Int("port", port),
  185. zap.String("from", px.From()),
  186. zap.String("to", px.To()),
  187. )
  188. }
  189. srv.advertisePeerPortToProxy = make(map[int]proxy.Server)
  190. }
  191. }
  192. func (srv *Server) saveEtcdLogFile() error {
  193. var err error
  194. srv.etcdLogFile, err = os.Create(srv.Member.EtcdLogPath)
  195. if err != nil {
  196. return err
  197. }
  198. srv.lg.Info("created etcd log file", zap.String("path", srv.Member.EtcdLogPath))
  199. return nil
  200. }
  201. func (srv *Server) creatEtcdCmd() {
  202. etcdPath, etcdFlags := srv.Member.EtcdExecPath, srv.Member.Etcd.Flags()
  203. u, _ := url.Parse(srv.Member.FailpointHTTPAddr)
  204. srv.lg.Info("creating etcd command",
  205. zap.String("etcd-exec-path", etcdPath),
  206. zap.Strings("etcd-flags", etcdFlags),
  207. zap.String("failpoint-http-addr", srv.Member.FailpointHTTPAddr),
  208. zap.String("failpoint-addr", u.Host),
  209. )
  210. srv.etcdCmd = exec.Command(etcdPath, etcdFlags...)
  211. srv.etcdCmd.Env = []string{"GOFAIL_HTTP=" + u.Host}
  212. srv.etcdCmd.Stdout = srv.etcdLogFile
  213. srv.etcdCmd.Stderr = srv.etcdLogFile
  214. }
  215. func (srv *Server) saveTLSAssets() error {
  216. // if started with manual TLS, stores TLS assets
  217. // from tester/client to disk before starting etcd process
  218. // TODO: not implemented yet
  219. if !srv.Member.Etcd.ClientAutoTLS {
  220. if srv.Member.Etcd.ClientCertAuth {
  221. return fmt.Errorf("manual TLS setup is not implemented yet, but Member.Etcd.ClientCertAuth is %v", srv.Member.Etcd.ClientCertAuth)
  222. }
  223. if srv.Member.Etcd.ClientCertFile != "" {
  224. return fmt.Errorf("manual TLS setup is not implemented yet, but Member.Etcd.ClientCertFile is %q", srv.Member.Etcd.ClientCertFile)
  225. }
  226. if srv.Member.Etcd.ClientKeyFile != "" {
  227. return fmt.Errorf("manual TLS setup is not implemented yet, but Member.Etcd.ClientKeyFile is %q", srv.Member.Etcd.ClientKeyFile)
  228. }
  229. if srv.Member.Etcd.ClientTrustedCAFile != "" {
  230. return fmt.Errorf("manual TLS setup is not implemented yet, but Member.Etcd.ClientTrustedCAFile is %q", srv.Member.Etcd.ClientTrustedCAFile)
  231. }
  232. }
  233. if !srv.Member.Etcd.PeerAutoTLS {
  234. if srv.Member.Etcd.PeerClientCertAuth {
  235. return fmt.Errorf("manual TLS setup is not implemented yet, but Member.Etcd.PeerClientCertAuth is %v", srv.Member.Etcd.PeerClientCertAuth)
  236. }
  237. if srv.Member.Etcd.PeerCertFile != "" {
  238. return fmt.Errorf("manual TLS setup is not implemented yet, but Member.Etcd.PeerCertFile is %q", srv.Member.Etcd.PeerCertFile)
  239. }
  240. if srv.Member.Etcd.PeerKeyFile != "" {
  241. return fmt.Errorf("manual TLS setup is not implemented yet, but Member.Etcd.PeerKeyFile is %q", srv.Member.Etcd.PeerKeyFile)
  242. }
  243. if srv.Member.Etcd.PeerTrustedCAFile != "" {
  244. return fmt.Errorf("manual TLS setup is not implemented yet, but Member.Etcd.PeerTrustedCAFile is %q", srv.Member.Etcd.PeerTrustedCAFile)
  245. }
  246. }
  247. // TODO
  248. return nil
  249. }
  250. func (srv *Server) loadAutoTLSAssets() error {
  251. // if started with auto TLS, sends back TLS assets to tester/client
  252. if srv.Member.Etcd.ClientAutoTLS {
  253. // in case of slow disk
  254. time.Sleep(time.Second)
  255. fdir := filepath.Join(srv.Member.Etcd.DataDir, "fixtures", "client")
  256. srv.lg.Info(
  257. "loading client TLS assets",
  258. zap.String("dir", fdir),
  259. zap.String("endpoint", srv.EtcdClientEndpoint),
  260. )
  261. certPath := filepath.Join(fdir, "cert.pem")
  262. if !fileutil.Exist(certPath) {
  263. return fmt.Errorf("cannot find %q", certPath)
  264. }
  265. certData, err := ioutil.ReadFile(certPath)
  266. if err != nil {
  267. return fmt.Errorf("cannot read %q (%v)", certPath, err)
  268. }
  269. srv.Member.ClientCertData = string(certData)
  270. keyPath := filepath.Join(fdir, "key.pem")
  271. if !fileutil.Exist(keyPath) {
  272. return fmt.Errorf("cannot find %q", keyPath)
  273. }
  274. keyData, err := ioutil.ReadFile(keyPath)
  275. if err != nil {
  276. return fmt.Errorf("cannot read %q (%v)", keyPath, err)
  277. }
  278. srv.Member.ClientKeyData = string(keyData)
  279. srv.lg.Info(
  280. "loaded client TLS assets",
  281. zap.String("peer-cert-path", certPath),
  282. zap.Int("peer-cert-length", len(certData)),
  283. zap.String("peer-key-path", keyPath),
  284. zap.Int("peer-key-length", len(keyData)),
  285. )
  286. }
  287. if srv.Member.Etcd.ClientAutoTLS {
  288. // in case of slow disk
  289. time.Sleep(time.Second)
  290. fdir := filepath.Join(srv.Member.Etcd.DataDir, "fixtures", "peer")
  291. srv.lg.Info(
  292. "loading client TLS assets",
  293. zap.String("dir", fdir),
  294. zap.String("endpoint", srv.EtcdClientEndpoint),
  295. )
  296. certPath := filepath.Join(fdir, "cert.pem")
  297. if !fileutil.Exist(certPath) {
  298. return fmt.Errorf("cannot find %q", certPath)
  299. }
  300. certData, err := ioutil.ReadFile(certPath)
  301. if err != nil {
  302. return fmt.Errorf("cannot read %q (%v)", certPath, err)
  303. }
  304. srv.Member.PeerCertData = string(certData)
  305. keyPath := filepath.Join(fdir, "key.pem")
  306. if !fileutil.Exist(keyPath) {
  307. return fmt.Errorf("cannot find %q", keyPath)
  308. }
  309. keyData, err := ioutil.ReadFile(keyPath)
  310. if err != nil {
  311. return fmt.Errorf("cannot read %q (%v)", keyPath, err)
  312. }
  313. srv.Member.PeerKeyData = string(keyData)
  314. srv.lg.Info(
  315. "loaded peer TLS assets",
  316. zap.String("peer-cert-path", certPath),
  317. zap.Int("peer-cert-length", len(certData)),
  318. zap.String("peer-key-path", keyPath),
  319. zap.Int("peer-key-length", len(keyData)),
  320. )
  321. }
  322. return nil
  323. }
  324. // start but do not wait for it to complete
  325. func (srv *Server) startEtcdCmd() error {
  326. return srv.etcdCmd.Start()
  327. }
  328. func (srv *Server) handleRestartEtcd() (*rpcpb.Response, error) {
  329. srv.creatEtcdCmd()
  330. var err error
  331. if err = srv.saveTLSAssets(); err != nil {
  332. return nil, err
  333. }
  334. if err = srv.startEtcdCmd(); err != nil {
  335. return nil, err
  336. }
  337. srv.lg.Info("restarted etcd", zap.String("command-path", srv.etcdCmd.Path))
  338. if err = srv.loadAutoTLSAssets(); err != nil {
  339. return nil, err
  340. }
  341. // wait some time for etcd listener start
  342. // before setting up proxy
  343. // TODO: local tests should handle port conflicts
  344. // with clients on restart
  345. time.Sleep(time.Second)
  346. if err = srv.startProxy(); err != nil {
  347. return nil, err
  348. }
  349. return &rpcpb.Response{
  350. Success: true,
  351. Status: "restart etcd PASS",
  352. Member: srv.Member,
  353. }, nil
  354. }
  355. func (srv *Server) handleKillEtcd() (*rpcpb.Response, error) {
  356. srv.stopProxy()
  357. err := stopWithSig(srv.etcdCmd, syscall.SIGTERM)
  358. if err != nil {
  359. return nil, err
  360. }
  361. srv.lg.Info("killed etcd", zap.String("signal", syscall.SIGTERM.String()))
  362. return &rpcpb.Response{
  363. Success: true,
  364. Status: "successfully killed etcd!",
  365. }, nil
  366. }
  367. func (srv *Server) handleFailArchive() (*rpcpb.Response, error) {
  368. srv.stopProxy()
  369. // exit with stackstrace
  370. err := stopWithSig(srv.etcdCmd, syscall.SIGQUIT)
  371. if err != nil {
  372. return nil, err
  373. }
  374. srv.lg.Info("killed etcd", zap.String("signal", syscall.SIGQUIT.String()))
  375. srv.etcdLogFile.Sync()
  376. srv.etcdLogFile.Close()
  377. // TODO: support separate WAL directory
  378. if err = archive(
  379. srv.Member.BaseDir,
  380. srv.Member.EtcdLogPath,
  381. srv.Member.Etcd.DataDir,
  382. ); err != nil {
  383. return nil, err
  384. }
  385. srv.lg.Info("archived data", zap.String("base-dir", srv.Member.BaseDir))
  386. if err = srv.saveEtcdLogFile(); err != nil {
  387. return nil, err
  388. }
  389. srv.lg.Info("cleaning up page cache")
  390. if err := cleanPageCache(); err != nil {
  391. srv.lg.Warn("failed to clean up page cache", zap.String("error", err.Error()))
  392. }
  393. srv.lg.Info("cleaned up page cache")
  394. return &rpcpb.Response{
  395. Success: true,
  396. Status: "successfully cleaned up etcd!",
  397. }, nil
  398. }
  399. // stop proxy, etcd, delete data directory
  400. func (srv *Server) handleDestroyEtcdAgent() (*rpcpb.Response, error) {
  401. err := stopWithSig(srv.etcdCmd, syscall.SIGTERM)
  402. if err != nil {
  403. return nil, err
  404. }
  405. srv.lg.Info("killed etcd", zap.String("signal", syscall.SIGTERM.String()))
  406. err = os.RemoveAll(srv.Member.BaseDir)
  407. if err != nil {
  408. return nil, err
  409. }
  410. srv.lg.Info("removed base directory", zap.String("dir", srv.Member.BaseDir))
  411. // stop agent server
  412. srv.Stop()
  413. for port, px := range srv.advertiseClientPortToProxy {
  414. err := px.Close()
  415. srv.lg.Info("closed proxy", zap.Int("client-port", port), zap.Error(err))
  416. }
  417. for port, px := range srv.advertisePeerPortToProxy {
  418. err := px.Close()
  419. srv.lg.Info("closed proxy", zap.Int("peer-port", port), zap.Error(err))
  420. }
  421. return &rpcpb.Response{
  422. Success: true,
  423. Status: "successfully destroyed etcd and agent!",
  424. }, nil
  425. }
  426. func (srv *Server) handleBlackholePeerPortTxRx() (*rpcpb.Response, error) {
  427. for port, px := range srv.advertisePeerPortToProxy {
  428. srv.lg.Info("blackholing", zap.Int("peer-port", port))
  429. px.BlackholeTx()
  430. px.BlackholeRx()
  431. srv.lg.Info("blackholed", zap.Int("peer-port", port))
  432. }
  433. return &rpcpb.Response{
  434. Success: true,
  435. Status: "successfully blackholed peer port tx/rx!",
  436. }, nil
  437. }
  438. func (srv *Server) handleUnblackholePeerPortTxRx() (*rpcpb.Response, error) {
  439. for port, px := range srv.advertisePeerPortToProxy {
  440. srv.lg.Info("unblackholing", zap.Int("peer-port", port))
  441. px.UnblackholeTx()
  442. px.UnblackholeRx()
  443. srv.lg.Info("unblackholed", zap.Int("peer-port", port))
  444. }
  445. return &rpcpb.Response{
  446. Success: true,
  447. Status: "successfully unblackholed peer port tx/rx!",
  448. }, nil
  449. }
  450. func (srv *Server) handleDelayPeerPortTxRx() (*rpcpb.Response, error) {
  451. lat := time.Duration(srv.Tester.UpdatedDelayLatencyMs) * time.Millisecond
  452. rv := time.Duration(srv.Tester.DelayLatencyMsRv) * time.Millisecond
  453. for port, px := range srv.advertisePeerPortToProxy {
  454. srv.lg.Info("delaying",
  455. zap.Int("peer-port", port),
  456. zap.Duration("latency", lat),
  457. zap.Duration("random-variable", rv),
  458. )
  459. px.DelayTx(lat, rv)
  460. px.DelayRx(lat, rv)
  461. srv.lg.Info("delayed",
  462. zap.Int("peer-port", port),
  463. zap.Duration("latency", lat),
  464. zap.Duration("random-variable", rv),
  465. )
  466. }
  467. return &rpcpb.Response{
  468. Success: true,
  469. Status: "successfully delay peer port tx/rx!",
  470. }, nil
  471. }
  472. func (srv *Server) handleUndelayPeerPortTxRx() (*rpcpb.Response, error) {
  473. for port, px := range srv.advertisePeerPortToProxy {
  474. srv.lg.Info("undelaying", zap.Int("peer-port", port))
  475. px.UndelayTx()
  476. px.UndelayRx()
  477. srv.lg.Info("undelayed", zap.Int("peer-port", port))
  478. }
  479. return &rpcpb.Response{
  480. Success: true,
  481. Status: "successfully undelay peer port tx/rx!",
  482. }, nil
  483. }