handler.go 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544
  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/functional/rpcpb"
  26. "github.com/coreos/etcd/pkg/fileutil"
  27. "github.com/coreos/etcd/pkg/proxy"
  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. func (srv *Server) startProxy() error {
  105. if srv.Member.EtcdClientProxy {
  106. advertiseClientURL, advertiseClientURLPort, err := getURLAndPort(srv.Member.Etcd.AdvertiseClientURLs[0])
  107. if err != nil {
  108. return err
  109. }
  110. listenClientURL, _, err := getURLAndPort(srv.Member.Etcd.ListenClientURLs[0])
  111. if err != nil {
  112. return err
  113. }
  114. srv.advertiseClientPortToProxy[advertiseClientURLPort] = proxy.NewServer(proxy.ServerConfig{
  115. Logger: srv.lg,
  116. From: *advertiseClientURL,
  117. To: *listenClientURL,
  118. })
  119. select {
  120. case err = <-srv.advertiseClientPortToProxy[advertiseClientURLPort].Error():
  121. return err
  122. case <-time.After(2 * time.Second):
  123. srv.lg.Info("started proxy on client traffic", zap.String("url", advertiseClientURL.String()))
  124. }
  125. }
  126. if srv.Member.EtcdPeerProxy {
  127. advertisePeerURL, advertisePeerURLPort, err := getURLAndPort(srv.Member.Etcd.AdvertisePeerURLs[0])
  128. if err != nil {
  129. return err
  130. }
  131. listenPeerURL, _, err := getURLAndPort(srv.Member.Etcd.ListenPeerURLs[0])
  132. if err != nil {
  133. return err
  134. }
  135. srv.advertisePeerPortToProxy[advertisePeerURLPort] = proxy.NewServer(proxy.ServerConfig{
  136. Logger: srv.lg,
  137. From: *advertisePeerURL,
  138. To: *listenPeerURL,
  139. })
  140. select {
  141. case err = <-srv.advertisePeerPortToProxy[advertisePeerURLPort].Error():
  142. return err
  143. case <-time.After(2 * time.Second):
  144. srv.lg.Info("started proxy on peer traffic", zap.String("url", advertisePeerURL.String()))
  145. }
  146. }
  147. return nil
  148. }
  149. func (srv *Server) stopProxy() {
  150. if srv.Member.EtcdClientProxy && len(srv.advertiseClientPortToProxy) > 0 {
  151. for port, px := range srv.advertiseClientPortToProxy {
  152. if err := px.Close(); err != nil {
  153. srv.lg.Warn("failed to close proxy", zap.Int("port", port))
  154. continue
  155. }
  156. select {
  157. case <-px.Done():
  158. // enough time to release port
  159. time.Sleep(time.Second)
  160. case <-time.After(time.Second):
  161. }
  162. srv.lg.Info("closed proxy",
  163. zap.Int("port", port),
  164. zap.String("from", px.From()),
  165. zap.String("to", px.To()),
  166. )
  167. }
  168. srv.advertiseClientPortToProxy = make(map[int]proxy.Server)
  169. }
  170. if srv.Member.EtcdPeerProxy && len(srv.advertisePeerPortToProxy) > 0 {
  171. for port, px := range srv.advertisePeerPortToProxy {
  172. if err := px.Close(); err != nil {
  173. srv.lg.Warn("failed to close proxy", zap.Int("port", port))
  174. continue
  175. }
  176. select {
  177. case <-px.Done():
  178. // enough time to release port
  179. time.Sleep(time.Second)
  180. case <-time.After(time.Second):
  181. }
  182. srv.lg.Info("closed proxy",
  183. zap.Int("port", port),
  184. zap.String("from", px.From()),
  185. zap.String("to", px.To()),
  186. )
  187. }
  188. srv.advertisePeerPortToProxy = make(map[int]proxy.Server)
  189. }
  190. }
  191. func (srv *Server) saveEtcdLogFile() error {
  192. var err error
  193. srv.etcdLogFile, err = os.Create(srv.Member.EtcdLogPath)
  194. if err != nil {
  195. return err
  196. }
  197. srv.lg.Info("created etcd log file", zap.String("path", srv.Member.EtcdLogPath))
  198. return nil
  199. }
  200. func (srv *Server) creatEtcdCmd() {
  201. etcdPath, etcdFlags := srv.Member.EtcdExecPath, srv.Member.Etcd.Flags()
  202. u, _ := url.Parse(srv.Member.FailpointHTTPAddr)
  203. srv.lg.Info("creating etcd command",
  204. zap.String("etcd-exec-path", etcdPath),
  205. zap.Strings("etcd-flags", etcdFlags),
  206. zap.String("failpoint-http-addr", srv.Member.FailpointHTTPAddr),
  207. zap.String("failpoint-addr", u.Host),
  208. )
  209. srv.etcdCmd = exec.Command(etcdPath, etcdFlags...)
  210. srv.etcdCmd.Env = []string{"GOFAIL_HTTP=" + u.Host}
  211. srv.etcdCmd.Stdout = srv.etcdLogFile
  212. srv.etcdCmd.Stderr = srv.etcdLogFile
  213. }
  214. func (srv *Server) saveTLSAssets() error {
  215. // if started with manual TLS, stores TLS assets
  216. // from tester/client to disk before starting etcd process
  217. // TODO: not implemented yet
  218. if !srv.Member.Etcd.ClientAutoTLS {
  219. if srv.Member.Etcd.ClientCertAuth {
  220. return fmt.Errorf("manual TLS setup is not implemented yet, but Member.Etcd.ClientCertAuth is %v", srv.Member.Etcd.ClientCertAuth)
  221. }
  222. if srv.Member.Etcd.ClientCertFile != "" {
  223. return fmt.Errorf("manual TLS setup is not implemented yet, but Member.Etcd.ClientCertFile is %q", srv.Member.Etcd.ClientCertFile)
  224. }
  225. if srv.Member.Etcd.ClientKeyFile != "" {
  226. return fmt.Errorf("manual TLS setup is not implemented yet, but Member.Etcd.ClientKeyFile is %q", srv.Member.Etcd.ClientKeyFile)
  227. }
  228. if srv.Member.Etcd.ClientTrustedCAFile != "" {
  229. return fmt.Errorf("manual TLS setup is not implemented yet, but Member.Etcd.ClientTrustedCAFile is %q", srv.Member.Etcd.ClientTrustedCAFile)
  230. }
  231. }
  232. if !srv.Member.Etcd.PeerAutoTLS {
  233. if srv.Member.Etcd.PeerClientCertAuth {
  234. return fmt.Errorf("manual TLS setup is not implemented yet, but Member.Etcd.PeerClientCertAuth is %v", srv.Member.Etcd.PeerClientCertAuth)
  235. }
  236. if srv.Member.Etcd.PeerCertFile != "" {
  237. return fmt.Errorf("manual TLS setup is not implemented yet, but Member.Etcd.PeerCertFile is %q", srv.Member.Etcd.PeerCertFile)
  238. }
  239. if srv.Member.Etcd.PeerKeyFile != "" {
  240. return fmt.Errorf("manual TLS setup is not implemented yet, but Member.Etcd.PeerKeyFile is %q", srv.Member.Etcd.PeerKeyFile)
  241. }
  242. if srv.Member.Etcd.PeerTrustedCAFile != "" {
  243. return fmt.Errorf("manual TLS setup is not implemented yet, but Member.Etcd.PeerTrustedCAFile is %q", srv.Member.Etcd.PeerTrustedCAFile)
  244. }
  245. }
  246. // TODO
  247. return nil
  248. }
  249. func (srv *Server) loadAutoTLSAssets() error {
  250. // if started with auto TLS, sends back TLS assets to tester/client
  251. if srv.Member.Etcd.ClientAutoTLS {
  252. // in case of slow disk
  253. time.Sleep(time.Second)
  254. fdir := filepath.Join(srv.Member.Etcd.DataDir, "fixtures", "client")
  255. srv.lg.Info(
  256. "loading client TLS assets",
  257. zap.String("dir", fdir),
  258. zap.String("endpoint", srv.EtcdClientEndpoint),
  259. )
  260. certPath := filepath.Join(fdir, "cert.pem")
  261. if !fileutil.Exist(certPath) {
  262. return fmt.Errorf("cannot find %q", certPath)
  263. }
  264. certData, err := ioutil.ReadFile(certPath)
  265. if err != nil {
  266. return fmt.Errorf("cannot read %q (%v)", certPath, err)
  267. }
  268. srv.Member.ClientCertData = string(certData)
  269. keyPath := filepath.Join(fdir, "key.pem")
  270. if !fileutil.Exist(keyPath) {
  271. return fmt.Errorf("cannot find %q", keyPath)
  272. }
  273. keyData, err := ioutil.ReadFile(keyPath)
  274. if err != nil {
  275. return fmt.Errorf("cannot read %q (%v)", keyPath, err)
  276. }
  277. srv.Member.ClientKeyData = string(keyData)
  278. srv.lg.Info(
  279. "loaded client TLS assets",
  280. zap.String("peer-cert-path", certPath),
  281. zap.Int("peer-cert-length", len(certData)),
  282. zap.String("peer-key-path", keyPath),
  283. zap.Int("peer-key-length", len(keyData)),
  284. )
  285. }
  286. if srv.Member.Etcd.ClientAutoTLS {
  287. // in case of slow disk
  288. time.Sleep(time.Second)
  289. fdir := filepath.Join(srv.Member.Etcd.DataDir, "fixtures", "peer")
  290. srv.lg.Info(
  291. "loading client TLS assets",
  292. zap.String("dir", fdir),
  293. zap.String("endpoint", srv.EtcdClientEndpoint),
  294. )
  295. certPath := filepath.Join(fdir, "cert.pem")
  296. if !fileutil.Exist(certPath) {
  297. return fmt.Errorf("cannot find %q", certPath)
  298. }
  299. certData, err := ioutil.ReadFile(certPath)
  300. if err != nil {
  301. return fmt.Errorf("cannot read %q (%v)", certPath, err)
  302. }
  303. srv.Member.PeerCertData = string(certData)
  304. keyPath := filepath.Join(fdir, "key.pem")
  305. if !fileutil.Exist(keyPath) {
  306. return fmt.Errorf("cannot find %q", keyPath)
  307. }
  308. keyData, err := ioutil.ReadFile(keyPath)
  309. if err != nil {
  310. return fmt.Errorf("cannot read %q (%v)", keyPath, err)
  311. }
  312. srv.Member.PeerKeyData = string(keyData)
  313. srv.lg.Info(
  314. "loaded peer TLS assets",
  315. zap.String("peer-cert-path", certPath),
  316. zap.Int("peer-cert-length", len(certData)),
  317. zap.String("peer-key-path", keyPath),
  318. zap.Int("peer-key-length", len(keyData)),
  319. )
  320. }
  321. return nil
  322. }
  323. // start but do not wait for it to complete
  324. func (srv *Server) startEtcdCmd() error {
  325. return srv.etcdCmd.Start()
  326. }
  327. func (srv *Server) handleRestartEtcd() (*rpcpb.Response, error) {
  328. srv.creatEtcdCmd()
  329. var err error
  330. if err = srv.saveTLSAssets(); err != nil {
  331. return nil, err
  332. }
  333. if err = srv.startEtcdCmd(); err != nil {
  334. return nil, err
  335. }
  336. srv.lg.Info("restarted etcd", zap.String("command-path", srv.etcdCmd.Path))
  337. if err = srv.loadAutoTLSAssets(); err != nil {
  338. return nil, err
  339. }
  340. // wait some time for etcd listener start
  341. // before setting up proxy
  342. // TODO: local tests should handle port conflicts
  343. // with clients on restart
  344. time.Sleep(time.Second)
  345. if err = srv.startProxy(); err != nil {
  346. return nil, err
  347. }
  348. return &rpcpb.Response{
  349. Success: true,
  350. Status: "restart etcd PASS",
  351. Member: srv.Member,
  352. }, nil
  353. }
  354. func (srv *Server) handleKillEtcd() (*rpcpb.Response, error) {
  355. srv.stopProxy()
  356. err := stopWithSig(srv.etcdCmd, syscall.SIGTERM)
  357. if err != nil {
  358. return nil, err
  359. }
  360. srv.lg.Info("killed etcd", zap.String("signal", syscall.SIGTERM.String()))
  361. return &rpcpb.Response{
  362. Success: true,
  363. Status: "successfully killed etcd!",
  364. }, nil
  365. }
  366. func (srv *Server) handleFailArchive() (*rpcpb.Response, error) {
  367. srv.stopProxy()
  368. // exit with stackstrace
  369. err := stopWithSig(srv.etcdCmd, syscall.SIGQUIT)
  370. if err != nil {
  371. return nil, err
  372. }
  373. srv.lg.Info("killed etcd", zap.String("signal", syscall.SIGQUIT.String()))
  374. srv.etcdLogFile.Sync()
  375. srv.etcdLogFile.Close()
  376. // TODO: support separate WAL directory
  377. if err = archive(
  378. srv.Member.BaseDir,
  379. srv.Member.EtcdLogPath,
  380. srv.Member.Etcd.DataDir,
  381. ); err != nil {
  382. return nil, err
  383. }
  384. srv.lg.Info("archived data", zap.String("base-dir", srv.Member.BaseDir))
  385. if err = srv.saveEtcdLogFile(); err != nil {
  386. return nil, err
  387. }
  388. srv.lg.Info("cleaning up page cache")
  389. if err := cleanPageCache(); err != nil {
  390. srv.lg.Warn("failed to clean up page cache", zap.String("error", err.Error()))
  391. }
  392. srv.lg.Info("cleaned up page cache")
  393. return &rpcpb.Response{
  394. Success: true,
  395. Status: "successfully cleaned up etcd!",
  396. }, nil
  397. }
  398. // stop proxy, etcd, delete data directory
  399. func (srv *Server) handleDestroyEtcdAgent() (*rpcpb.Response, error) {
  400. err := stopWithSig(srv.etcdCmd, syscall.SIGTERM)
  401. if err != nil {
  402. return nil, err
  403. }
  404. srv.lg.Info("killed etcd", zap.String("signal", syscall.SIGTERM.String()))
  405. err = os.RemoveAll(srv.Member.BaseDir)
  406. if err != nil {
  407. return nil, err
  408. }
  409. srv.lg.Info("removed base directory", zap.String("dir", srv.Member.BaseDir))
  410. // stop agent server
  411. srv.Stop()
  412. for port, px := range srv.advertiseClientPortToProxy {
  413. err := px.Close()
  414. srv.lg.Info("closed proxy", zap.Int("client-port", port), zap.Error(err))
  415. }
  416. for port, px := range srv.advertisePeerPortToProxy {
  417. err := px.Close()
  418. srv.lg.Info("closed proxy", zap.Int("peer-port", port), zap.Error(err))
  419. }
  420. return &rpcpb.Response{
  421. Success: true,
  422. Status: "successfully destroyed etcd and agent!",
  423. }, nil
  424. }
  425. func (srv *Server) handleBlackholePeerPortTxRx() (*rpcpb.Response, error) {
  426. for port, px := range srv.advertisePeerPortToProxy {
  427. srv.lg.Info("blackholing", zap.Int("peer-port", port))
  428. px.BlackholeTx()
  429. px.BlackholeRx()
  430. srv.lg.Info("blackholed", zap.Int("peer-port", port))
  431. }
  432. return &rpcpb.Response{
  433. Success: true,
  434. Status: "successfully blackholed peer port tx/rx!",
  435. }, nil
  436. }
  437. func (srv *Server) handleUnblackholePeerPortTxRx() (*rpcpb.Response, error) {
  438. for port, px := range srv.advertisePeerPortToProxy {
  439. srv.lg.Info("unblackholing", zap.Int("peer-port", port))
  440. px.UnblackholeTx()
  441. px.UnblackholeRx()
  442. srv.lg.Info("unblackholed", zap.Int("peer-port", port))
  443. }
  444. return &rpcpb.Response{
  445. Success: true,
  446. Status: "successfully unblackholed peer port tx/rx!",
  447. }, nil
  448. }
  449. func (srv *Server) handleDelayPeerPortTxRx() (*rpcpb.Response, error) {
  450. lat := time.Duration(srv.Tester.UpdatedDelayLatencyMs) * time.Millisecond
  451. rv := time.Duration(srv.Tester.DelayLatencyMsRv) * time.Millisecond
  452. for port, px := range srv.advertisePeerPortToProxy {
  453. srv.lg.Info("delaying",
  454. zap.Int("peer-port", port),
  455. zap.Duration("latency", lat),
  456. zap.Duration("random-variable", rv),
  457. )
  458. px.DelayTx(lat, rv)
  459. px.DelayRx(lat, rv)
  460. srv.lg.Info("delayed",
  461. zap.Int("peer-port", port),
  462. zap.Duration("latency", lat),
  463. zap.Duration("random-variable", rv),
  464. )
  465. }
  466. return &rpcpb.Response{
  467. Success: true,
  468. Status: "successfully delay peer port tx/rx!",
  469. }, nil
  470. }
  471. func (srv *Server) handleUndelayPeerPortTxRx() (*rpcpb.Response, error) {
  472. for port, px := range srv.advertisePeerPortToProxy {
  473. srv.lg.Info("undelaying", zap.Int("peer-port", port))
  474. px.UndelayTx()
  475. px.UndelayRx()
  476. srv.lg.Info("undelayed", zap.Int("peer-port", port))
  477. }
  478. return &rpcpb.Response{
  479. Success: true,
  480. Status: "successfully undelay peer port tx/rx!",
  481. }, nil
  482. }