handler.go 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784
  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. "go.etcd.io/etcd/embed"
  26. "go.etcd.io/etcd/functional/rpcpb"
  27. "go.etcd.io/etcd/pkg/fileutil"
  28. "go.etcd.io/etcd/pkg/proxy"
  29. "go.uber.org/zap"
  30. )
  31. // return error for system errors (e.g. fail to create files)
  32. // return status error in response for wrong configuration/operation (e.g. start etcd twice)
  33. func (srv *Server) handleTesterRequest(req *rpcpb.Request) (resp *rpcpb.Response, err error) {
  34. defer func() {
  35. if err == nil && req != nil {
  36. srv.last = req.Operation
  37. srv.lg.Info("handler success", zap.String("operation", req.Operation.String()))
  38. }
  39. }()
  40. if req != nil {
  41. srv.Member = req.Member
  42. srv.Tester = req.Tester
  43. }
  44. switch req.Operation {
  45. case rpcpb.Operation_INITIAL_START_ETCD:
  46. return srv.handle_INITIAL_START_ETCD(req)
  47. case rpcpb.Operation_RESTART_ETCD:
  48. return srv.handle_RESTART_ETCD()
  49. case rpcpb.Operation_SIGTERM_ETCD:
  50. return srv.handle_SIGTERM_ETCD()
  51. case rpcpb.Operation_SIGQUIT_ETCD_AND_REMOVE_DATA:
  52. return srv.handle_SIGQUIT_ETCD_AND_REMOVE_DATA()
  53. case rpcpb.Operation_SAVE_SNAPSHOT:
  54. return srv.handle_SAVE_SNAPSHOT()
  55. case rpcpb.Operation_RESTORE_RESTART_FROM_SNAPSHOT:
  56. return srv.handle_RESTORE_RESTART_FROM_SNAPSHOT()
  57. case rpcpb.Operation_RESTART_FROM_SNAPSHOT:
  58. return srv.handle_RESTART_FROM_SNAPSHOT()
  59. case rpcpb.Operation_SIGQUIT_ETCD_AND_ARCHIVE_DATA:
  60. return srv.handle_SIGQUIT_ETCD_AND_ARCHIVE_DATA()
  61. case rpcpb.Operation_SIGQUIT_ETCD_AND_REMOVE_DATA_AND_STOP_AGENT:
  62. return srv.handle_SIGQUIT_ETCD_AND_REMOVE_DATA_AND_STOP_AGENT()
  63. case rpcpb.Operation_BLACKHOLE_PEER_PORT_TX_RX:
  64. return srv.handle_BLACKHOLE_PEER_PORT_TX_RX(), nil
  65. case rpcpb.Operation_UNBLACKHOLE_PEER_PORT_TX_RX:
  66. return srv.handle_UNBLACKHOLE_PEER_PORT_TX_RX(), nil
  67. case rpcpb.Operation_DELAY_PEER_PORT_TX_RX:
  68. return srv.handle_DELAY_PEER_PORT_TX_RX(), nil
  69. case rpcpb.Operation_UNDELAY_PEER_PORT_TX_RX:
  70. return srv.handle_UNDELAY_PEER_PORT_TX_RX(), nil
  71. default:
  72. msg := fmt.Sprintf("operation not found (%v)", req.Operation)
  73. return &rpcpb.Response{Success: false, Status: msg}, errors.New(msg)
  74. }
  75. }
  76. // just archive the first file
  77. func (srv *Server) createEtcdLogFile() error {
  78. var err error
  79. srv.etcdLogFile, err = os.Create(srv.Member.Etcd.LogOutputs[0])
  80. if err != nil {
  81. return err
  82. }
  83. srv.lg.Info("created etcd log file", zap.String("path", srv.Member.Etcd.LogOutputs[0]))
  84. return nil
  85. }
  86. func (srv *Server) creatEtcd(fromSnapshot bool) error {
  87. if !fileutil.Exist(srv.Member.EtcdExec) && srv.Member.EtcdExec != "embed" {
  88. return fmt.Errorf("unknown etcd exec %q or path does not exist", srv.Member.EtcdExec)
  89. }
  90. if srv.Member.EtcdExec != "embed" {
  91. etcdPath, etcdFlags := srv.Member.EtcdExec, srv.Member.Etcd.Flags()
  92. if fromSnapshot {
  93. etcdFlags = srv.Member.EtcdOnSnapshotRestore.Flags()
  94. }
  95. u, _ := url.Parse(srv.Member.FailpointHTTPAddr)
  96. srv.lg.Info(
  97. "creating etcd command",
  98. zap.String("etcd-exec", etcdPath),
  99. zap.Strings("etcd-flags", etcdFlags),
  100. zap.String("failpoint-http-addr", srv.Member.FailpointHTTPAddr),
  101. zap.String("failpoint-addr", u.Host),
  102. )
  103. srv.etcdCmd = exec.Command(etcdPath, etcdFlags...)
  104. srv.etcdCmd.Env = []string{"GOFAIL_HTTP=" + u.Host}
  105. srv.etcdCmd.Stdout = srv.etcdLogFile
  106. srv.etcdCmd.Stderr = srv.etcdLogFile
  107. return nil
  108. }
  109. cfg, err := srv.Member.Etcd.EmbedConfig()
  110. if err != nil {
  111. return err
  112. }
  113. srv.lg.Info("starting embedded etcd", zap.String("name", cfg.Name))
  114. srv.etcdServer, err = embed.StartEtcd(cfg)
  115. if err != nil {
  116. return err
  117. }
  118. srv.lg.Info("started embedded etcd", zap.String("name", cfg.Name))
  119. return nil
  120. }
  121. // start but do not wait for it to complete
  122. func (srv *Server) runEtcd() error {
  123. errc := make(chan error)
  124. go func() {
  125. time.Sleep(5 * time.Second)
  126. // server advertise client/peer listener had to start first
  127. // before setting up proxy listener
  128. errc <- srv.startProxy()
  129. }()
  130. if srv.etcdCmd != nil {
  131. srv.lg.Info(
  132. "starting etcd command",
  133. zap.String("command-path", srv.etcdCmd.Path),
  134. )
  135. err := srv.etcdCmd.Start()
  136. perr := <-errc
  137. srv.lg.Info(
  138. "started etcd command",
  139. zap.String("command-path", srv.etcdCmd.Path),
  140. zap.Errors("errors", []error{err, perr}),
  141. )
  142. if err != nil {
  143. return err
  144. }
  145. return perr
  146. }
  147. select {
  148. case <-srv.etcdServer.Server.ReadyNotify():
  149. srv.lg.Info("embedded etcd is ready")
  150. case <-time.After(time.Minute):
  151. srv.etcdServer.Close()
  152. return fmt.Errorf("took too long to start %v", <-srv.etcdServer.Err())
  153. }
  154. return <-errc
  155. }
  156. // SIGQUIT to exit with stackstrace
  157. func (srv *Server) stopEtcd(sig os.Signal) error {
  158. srv.stopProxy()
  159. if srv.etcdCmd != nil {
  160. srv.lg.Info(
  161. "stopping etcd command",
  162. zap.String("command-path", srv.etcdCmd.Path),
  163. zap.String("signal", sig.String()),
  164. )
  165. err := srv.etcdCmd.Process.Signal(sig)
  166. if err != nil {
  167. return err
  168. }
  169. errc := make(chan error)
  170. go func() {
  171. _, ew := srv.etcdCmd.Process.Wait()
  172. errc <- ew
  173. close(errc)
  174. }()
  175. select {
  176. case <-time.After(5 * time.Second):
  177. srv.etcdCmd.Process.Kill()
  178. case e := <-errc:
  179. return e
  180. }
  181. err = <-errc
  182. srv.lg.Info(
  183. "stopped etcd command",
  184. zap.String("command-path", srv.etcdCmd.Path),
  185. zap.String("signal", sig.String()),
  186. zap.Error(err),
  187. )
  188. return err
  189. }
  190. srv.lg.Info("stopping embedded etcd")
  191. srv.etcdServer.Server.HardStop()
  192. srv.etcdServer.Close()
  193. srv.lg.Info("stopped embedded etcd")
  194. return nil
  195. }
  196. func (srv *Server) startProxy() error {
  197. if srv.Member.EtcdClientProxy {
  198. advertiseClientURL, advertiseClientURLPort, err := getURLAndPort(srv.Member.Etcd.AdvertiseClientURLs[0])
  199. if err != nil {
  200. return err
  201. }
  202. listenClientURL, _, err := getURLAndPort(srv.Member.Etcd.ListenClientURLs[0])
  203. if err != nil {
  204. return err
  205. }
  206. srv.lg.Info("starting proxy on client traffic", zap.String("url", advertiseClientURL.String()))
  207. srv.advertiseClientPortToProxy[advertiseClientURLPort] = proxy.NewServer(proxy.ServerConfig{
  208. Logger: srv.lg,
  209. From: *advertiseClientURL,
  210. To: *listenClientURL,
  211. })
  212. select {
  213. case err = <-srv.advertiseClientPortToProxy[advertiseClientURLPort].Error():
  214. return err
  215. case <-time.After(2 * time.Second):
  216. srv.lg.Info("started proxy on client traffic", zap.String("url", advertiseClientURL.String()))
  217. }
  218. }
  219. if srv.Member.EtcdPeerProxy {
  220. advertisePeerURL, advertisePeerURLPort, err := getURLAndPort(srv.Member.Etcd.AdvertisePeerURLs[0])
  221. if err != nil {
  222. return err
  223. }
  224. listenPeerURL, _, err := getURLAndPort(srv.Member.Etcd.ListenPeerURLs[0])
  225. if err != nil {
  226. return err
  227. }
  228. srv.lg.Info("starting proxy on peer traffic", zap.String("url", advertisePeerURL.String()))
  229. srv.advertisePeerPortToProxy[advertisePeerURLPort] = proxy.NewServer(proxy.ServerConfig{
  230. Logger: srv.lg,
  231. From: *advertisePeerURL,
  232. To: *listenPeerURL,
  233. })
  234. select {
  235. case err = <-srv.advertisePeerPortToProxy[advertisePeerURLPort].Error():
  236. return err
  237. case <-time.After(2 * time.Second):
  238. srv.lg.Info("started proxy on peer traffic", zap.String("url", advertisePeerURL.String()))
  239. }
  240. }
  241. return nil
  242. }
  243. func (srv *Server) stopProxy() {
  244. if srv.Member.EtcdClientProxy && len(srv.advertiseClientPortToProxy) > 0 {
  245. for port, px := range srv.advertiseClientPortToProxy {
  246. if err := px.Close(); err != nil {
  247. srv.lg.Warn("failed to close proxy", zap.Int("port", port))
  248. continue
  249. }
  250. select {
  251. case <-px.Done():
  252. // enough time to release port
  253. time.Sleep(time.Second)
  254. case <-time.After(time.Second):
  255. }
  256. srv.lg.Info("closed proxy",
  257. zap.Int("port", port),
  258. zap.String("from", px.From()),
  259. zap.String("to", px.To()),
  260. )
  261. }
  262. srv.advertiseClientPortToProxy = make(map[int]proxy.Server)
  263. }
  264. if srv.Member.EtcdPeerProxy && len(srv.advertisePeerPortToProxy) > 0 {
  265. for port, px := range srv.advertisePeerPortToProxy {
  266. if err := px.Close(); err != nil {
  267. srv.lg.Warn("failed to close proxy", zap.Int("port", port))
  268. continue
  269. }
  270. select {
  271. case <-px.Done():
  272. // enough time to release port
  273. time.Sleep(time.Second)
  274. case <-time.After(time.Second):
  275. }
  276. srv.lg.Info("closed proxy",
  277. zap.Int("port", port),
  278. zap.String("from", px.From()),
  279. zap.String("to", px.To()),
  280. )
  281. }
  282. srv.advertisePeerPortToProxy = make(map[int]proxy.Server)
  283. }
  284. }
  285. // if started with manual TLS, stores TLS assets
  286. // from tester/client to disk before starting etcd process
  287. func (srv *Server) saveTLSAssets() error {
  288. if srv.Member.PeerCertPath != "" {
  289. if srv.Member.PeerCertData == "" {
  290. return fmt.Errorf("got empty data for %q", srv.Member.PeerCertPath)
  291. }
  292. if err := ioutil.WriteFile(srv.Member.PeerCertPath, []byte(srv.Member.PeerCertData), 0644); err != nil {
  293. return err
  294. }
  295. }
  296. if srv.Member.PeerKeyPath != "" {
  297. if srv.Member.PeerKeyData == "" {
  298. return fmt.Errorf("got empty data for %q", srv.Member.PeerKeyPath)
  299. }
  300. if err := ioutil.WriteFile(srv.Member.PeerKeyPath, []byte(srv.Member.PeerKeyData), 0644); err != nil {
  301. return err
  302. }
  303. }
  304. if srv.Member.PeerTrustedCAPath != "" {
  305. if srv.Member.PeerTrustedCAData == "" {
  306. return fmt.Errorf("got empty data for %q", srv.Member.PeerTrustedCAPath)
  307. }
  308. if err := ioutil.WriteFile(srv.Member.PeerTrustedCAPath, []byte(srv.Member.PeerTrustedCAData), 0644); err != nil {
  309. return err
  310. }
  311. }
  312. if srv.Member.PeerCertPath != "" &&
  313. srv.Member.PeerKeyPath != "" &&
  314. srv.Member.PeerTrustedCAPath != "" {
  315. srv.lg.Info(
  316. "wrote",
  317. zap.String("peer-cert", srv.Member.PeerCertPath),
  318. zap.String("peer-key", srv.Member.PeerKeyPath),
  319. zap.String("peer-trusted-ca", srv.Member.PeerTrustedCAPath),
  320. )
  321. }
  322. if srv.Member.ClientCertPath != "" {
  323. if srv.Member.ClientCertData == "" {
  324. return fmt.Errorf("got empty data for %q", srv.Member.ClientCertPath)
  325. }
  326. if err := ioutil.WriteFile(srv.Member.ClientCertPath, []byte(srv.Member.ClientCertData), 0644); err != nil {
  327. return err
  328. }
  329. }
  330. if srv.Member.ClientKeyPath != "" {
  331. if srv.Member.ClientKeyData == "" {
  332. return fmt.Errorf("got empty data for %q", srv.Member.ClientKeyPath)
  333. }
  334. if err := ioutil.WriteFile(srv.Member.ClientKeyPath, []byte(srv.Member.ClientKeyData), 0644); err != nil {
  335. return err
  336. }
  337. }
  338. if srv.Member.ClientTrustedCAPath != "" {
  339. if srv.Member.ClientTrustedCAData == "" {
  340. return fmt.Errorf("got empty data for %q", srv.Member.ClientTrustedCAPath)
  341. }
  342. if err := ioutil.WriteFile(srv.Member.ClientTrustedCAPath, []byte(srv.Member.ClientTrustedCAData), 0644); err != nil {
  343. return err
  344. }
  345. }
  346. if srv.Member.ClientCertPath != "" &&
  347. srv.Member.ClientKeyPath != "" &&
  348. srv.Member.ClientTrustedCAPath != "" {
  349. srv.lg.Info(
  350. "wrote",
  351. zap.String("client-cert", srv.Member.ClientCertPath),
  352. zap.String("client-key", srv.Member.ClientKeyPath),
  353. zap.String("client-trusted-ca", srv.Member.ClientTrustedCAPath),
  354. )
  355. }
  356. return nil
  357. }
  358. func (srv *Server) loadAutoTLSAssets() error {
  359. if srv.Member.Etcd.PeerAutoTLS {
  360. // in case of slow disk
  361. time.Sleep(time.Second)
  362. fdir := filepath.Join(srv.Member.Etcd.DataDir, "fixtures", "peer")
  363. srv.lg.Info(
  364. "loading client auto TLS assets",
  365. zap.String("dir", fdir),
  366. zap.String("endpoint", srv.EtcdClientEndpoint),
  367. )
  368. certPath := filepath.Join(fdir, "cert.pem")
  369. if !fileutil.Exist(certPath) {
  370. return fmt.Errorf("cannot find %q", certPath)
  371. }
  372. certData, err := ioutil.ReadFile(certPath)
  373. if err != nil {
  374. return fmt.Errorf("cannot read %q (%v)", certPath, err)
  375. }
  376. srv.Member.PeerCertData = string(certData)
  377. keyPath := filepath.Join(fdir, "key.pem")
  378. if !fileutil.Exist(keyPath) {
  379. return fmt.Errorf("cannot find %q", keyPath)
  380. }
  381. keyData, err := ioutil.ReadFile(keyPath)
  382. if err != nil {
  383. return fmt.Errorf("cannot read %q (%v)", keyPath, err)
  384. }
  385. srv.Member.PeerKeyData = string(keyData)
  386. srv.lg.Info(
  387. "loaded peer auto TLS assets",
  388. zap.String("peer-cert-path", certPath),
  389. zap.Int("peer-cert-length", len(certData)),
  390. zap.String("peer-key-path", keyPath),
  391. zap.Int("peer-key-length", len(keyData)),
  392. )
  393. }
  394. if srv.Member.Etcd.ClientAutoTLS {
  395. // in case of slow disk
  396. time.Sleep(time.Second)
  397. fdir := filepath.Join(srv.Member.Etcd.DataDir, "fixtures", "client")
  398. srv.lg.Info(
  399. "loading client TLS assets",
  400. zap.String("dir", fdir),
  401. zap.String("endpoint", srv.EtcdClientEndpoint),
  402. )
  403. certPath := filepath.Join(fdir, "cert.pem")
  404. if !fileutil.Exist(certPath) {
  405. return fmt.Errorf("cannot find %q", certPath)
  406. }
  407. certData, err := ioutil.ReadFile(certPath)
  408. if err != nil {
  409. return fmt.Errorf("cannot read %q (%v)", certPath, err)
  410. }
  411. srv.Member.ClientCertData = string(certData)
  412. keyPath := filepath.Join(fdir, "key.pem")
  413. if !fileutil.Exist(keyPath) {
  414. return fmt.Errorf("cannot find %q", keyPath)
  415. }
  416. keyData, err := ioutil.ReadFile(keyPath)
  417. if err != nil {
  418. return fmt.Errorf("cannot read %q (%v)", keyPath, err)
  419. }
  420. srv.Member.ClientKeyData = string(keyData)
  421. srv.lg.Info(
  422. "loaded client TLS assets",
  423. zap.String("peer-cert-path", certPath),
  424. zap.Int("peer-cert-length", len(certData)),
  425. zap.String("peer-key-path", keyPath),
  426. zap.Int("peer-key-length", len(keyData)),
  427. )
  428. }
  429. return nil
  430. }
  431. func (srv *Server) handle_INITIAL_START_ETCD(req *rpcpb.Request) (*rpcpb.Response, error) {
  432. if srv.last != rpcpb.Operation_NOT_STARTED {
  433. return &rpcpb.Response{
  434. Success: false,
  435. Status: fmt.Sprintf("%q is not valid; last server operation was %q", rpcpb.Operation_INITIAL_START_ETCD.String(), srv.last.String()),
  436. Member: req.Member,
  437. }, nil
  438. }
  439. err := fileutil.TouchDirAll(srv.Member.BaseDir)
  440. if err != nil {
  441. return nil, err
  442. }
  443. srv.lg.Info("created base directory", zap.String("path", srv.Member.BaseDir))
  444. if srv.etcdServer == nil {
  445. if err = srv.createEtcdLogFile(); err != nil {
  446. return nil, err
  447. }
  448. }
  449. if err = srv.saveTLSAssets(); err != nil {
  450. return nil, err
  451. }
  452. if err = srv.creatEtcd(false); err != nil {
  453. return nil, err
  454. }
  455. if err = srv.runEtcd(); err != nil {
  456. return nil, err
  457. }
  458. if err = srv.loadAutoTLSAssets(); err != nil {
  459. return nil, err
  460. }
  461. return &rpcpb.Response{
  462. Success: true,
  463. Status: "start etcd PASS",
  464. Member: srv.Member,
  465. }, nil
  466. }
  467. func (srv *Server) handle_RESTART_ETCD() (*rpcpb.Response, error) {
  468. var err error
  469. if !fileutil.Exist(srv.Member.BaseDir) {
  470. err = fileutil.TouchDirAll(srv.Member.BaseDir)
  471. if err != nil {
  472. return nil, err
  473. }
  474. }
  475. if err = srv.saveTLSAssets(); err != nil {
  476. return nil, err
  477. }
  478. if err = srv.creatEtcd(false); err != nil {
  479. return nil, err
  480. }
  481. if err = srv.runEtcd(); err != nil {
  482. return nil, err
  483. }
  484. if err = srv.loadAutoTLSAssets(); err != nil {
  485. return nil, err
  486. }
  487. return &rpcpb.Response{
  488. Success: true,
  489. Status: "restart etcd PASS",
  490. Member: srv.Member,
  491. }, nil
  492. }
  493. func (srv *Server) handle_SIGTERM_ETCD() (*rpcpb.Response, error) {
  494. if err := srv.stopEtcd(syscall.SIGTERM); err != nil {
  495. return nil, err
  496. }
  497. if srv.etcdServer != nil {
  498. srv.etcdServer.GetLogger().Sync()
  499. } else {
  500. srv.etcdLogFile.Sync()
  501. }
  502. return &rpcpb.Response{
  503. Success: true,
  504. Status: "killed etcd",
  505. }, nil
  506. }
  507. func (srv *Server) handle_SIGQUIT_ETCD_AND_REMOVE_DATA() (*rpcpb.Response, error) {
  508. err := srv.stopEtcd(syscall.SIGQUIT)
  509. if err != nil {
  510. return nil, err
  511. }
  512. if srv.etcdServer != nil {
  513. srv.etcdServer.GetLogger().Sync()
  514. } else {
  515. srv.etcdLogFile.Sync()
  516. srv.etcdLogFile.Close()
  517. }
  518. // for debugging purposes, rename instead of removing
  519. if err = os.RemoveAll(srv.Member.BaseDir + ".backup"); err != nil {
  520. return nil, err
  521. }
  522. if err = os.Rename(srv.Member.BaseDir, srv.Member.BaseDir+".backup"); err != nil {
  523. return nil, err
  524. }
  525. srv.lg.Info(
  526. "renamed",
  527. zap.String("base-dir", srv.Member.BaseDir),
  528. zap.String("new-dir", srv.Member.BaseDir+".backup"),
  529. )
  530. // create a new log file for next new member restart
  531. if !fileutil.Exist(srv.Member.BaseDir) {
  532. err = fileutil.TouchDirAll(srv.Member.BaseDir)
  533. if err != nil {
  534. return nil, err
  535. }
  536. }
  537. return &rpcpb.Response{
  538. Success: true,
  539. Status: "killed etcd and removed base directory",
  540. }, nil
  541. }
  542. func (srv *Server) handle_SAVE_SNAPSHOT() (*rpcpb.Response, error) {
  543. err := srv.Member.SaveSnapshot(srv.lg)
  544. if err != nil {
  545. return nil, err
  546. }
  547. return &rpcpb.Response{
  548. Success: true,
  549. Status: "saved snapshot",
  550. SnapshotInfo: srv.Member.SnapshotInfo,
  551. }, nil
  552. }
  553. func (srv *Server) handle_RESTORE_RESTART_FROM_SNAPSHOT() (resp *rpcpb.Response, err error) {
  554. err = srv.Member.RestoreSnapshot(srv.lg)
  555. if err != nil {
  556. return nil, err
  557. }
  558. resp, err = srv.handle_RESTART_FROM_SNAPSHOT()
  559. if resp != nil && err == nil {
  560. resp.Status = "restored snapshot and " + resp.Status
  561. }
  562. return resp, err
  563. }
  564. func (srv *Server) handle_RESTART_FROM_SNAPSHOT() (resp *rpcpb.Response, err error) {
  565. if err = srv.saveTLSAssets(); err != nil {
  566. return nil, err
  567. }
  568. if err = srv.creatEtcd(true); err != nil {
  569. return nil, err
  570. }
  571. if err = srv.runEtcd(); err != nil {
  572. return nil, err
  573. }
  574. if err = srv.loadAutoTLSAssets(); err != nil {
  575. return nil, err
  576. }
  577. return &rpcpb.Response{
  578. Success: true,
  579. Status: "restarted etcd from snapshot",
  580. SnapshotInfo: srv.Member.SnapshotInfo,
  581. }, nil
  582. }
  583. func (srv *Server) handle_SIGQUIT_ETCD_AND_ARCHIVE_DATA() (*rpcpb.Response, error) {
  584. err := srv.stopEtcd(syscall.SIGQUIT)
  585. if err != nil {
  586. return nil, err
  587. }
  588. if srv.etcdServer != nil {
  589. srv.etcdServer.GetLogger().Sync()
  590. } else {
  591. srv.etcdLogFile.Sync()
  592. srv.etcdLogFile.Close()
  593. }
  594. // TODO: support separate WAL directory
  595. if err = archive(
  596. srv.Member.BaseDir,
  597. srv.Member.Etcd.LogOutputs[0],
  598. srv.Member.Etcd.DataDir,
  599. ); err != nil {
  600. return nil, err
  601. }
  602. srv.lg.Info("archived data", zap.String("base-dir", srv.Member.BaseDir))
  603. if srv.etcdServer == nil {
  604. if err = srv.createEtcdLogFile(); err != nil {
  605. return nil, err
  606. }
  607. }
  608. srv.lg.Info("cleaning up page cache")
  609. if err := cleanPageCache(); err != nil {
  610. srv.lg.Warn("failed to clean up page cache", zap.String("error", err.Error()))
  611. }
  612. srv.lg.Info("cleaned up page cache")
  613. return &rpcpb.Response{
  614. Success: true,
  615. Status: "cleaned up etcd",
  616. }, nil
  617. }
  618. // stop proxy, etcd, delete data directory
  619. func (srv *Server) handle_SIGQUIT_ETCD_AND_REMOVE_DATA_AND_STOP_AGENT() (*rpcpb.Response, error) {
  620. err := srv.stopEtcd(syscall.SIGQUIT)
  621. if err != nil {
  622. return nil, err
  623. }
  624. if srv.etcdServer != nil {
  625. srv.etcdServer.GetLogger().Sync()
  626. } else {
  627. srv.etcdLogFile.Sync()
  628. srv.etcdLogFile.Close()
  629. }
  630. err = os.RemoveAll(srv.Member.BaseDir)
  631. if err != nil {
  632. return nil, err
  633. }
  634. srv.lg.Info("removed base directory", zap.String("dir", srv.Member.BaseDir))
  635. // stop agent server
  636. srv.Stop()
  637. return &rpcpb.Response{
  638. Success: true,
  639. Status: "destroyed etcd and agent",
  640. }, nil
  641. }
  642. func (srv *Server) handle_BLACKHOLE_PEER_PORT_TX_RX() *rpcpb.Response {
  643. for port, px := range srv.advertisePeerPortToProxy {
  644. srv.lg.Info("blackholing", zap.Int("peer-port", port))
  645. px.BlackholeTx()
  646. px.BlackholeRx()
  647. srv.lg.Info("blackholed", zap.Int("peer-port", port))
  648. }
  649. return &rpcpb.Response{
  650. Success: true,
  651. Status: "blackholed peer port tx/rx",
  652. }
  653. }
  654. func (srv *Server) handle_UNBLACKHOLE_PEER_PORT_TX_RX() *rpcpb.Response {
  655. for port, px := range srv.advertisePeerPortToProxy {
  656. srv.lg.Info("unblackholing", zap.Int("peer-port", port))
  657. px.UnblackholeTx()
  658. px.UnblackholeRx()
  659. srv.lg.Info("unblackholed", zap.Int("peer-port", port))
  660. }
  661. return &rpcpb.Response{
  662. Success: true,
  663. Status: "unblackholed peer port tx/rx",
  664. }
  665. }
  666. func (srv *Server) handle_DELAY_PEER_PORT_TX_RX() *rpcpb.Response {
  667. lat := time.Duration(srv.Tester.UpdatedDelayLatencyMs) * time.Millisecond
  668. rv := time.Duration(srv.Tester.DelayLatencyMsRv) * time.Millisecond
  669. for port, px := range srv.advertisePeerPortToProxy {
  670. srv.lg.Info("delaying",
  671. zap.Int("peer-port", port),
  672. zap.Duration("latency", lat),
  673. zap.Duration("random-variable", rv),
  674. )
  675. px.DelayTx(lat, rv)
  676. px.DelayRx(lat, rv)
  677. srv.lg.Info("delayed",
  678. zap.Int("peer-port", port),
  679. zap.Duration("latency", lat),
  680. zap.Duration("random-variable", rv),
  681. )
  682. }
  683. return &rpcpb.Response{
  684. Success: true,
  685. Status: "delayed peer port tx/rx",
  686. }
  687. }
  688. func (srv *Server) handle_UNDELAY_PEER_PORT_TX_RX() *rpcpb.Response {
  689. for port, px := range srv.advertisePeerPortToProxy {
  690. srv.lg.Info("undelaying", zap.Int("peer-port", port))
  691. px.UndelayTx()
  692. px.UndelayRx()
  693. srv.lg.Info("undelayed", zap.Int("peer-port", port))
  694. }
  695. return &rpcpb.Response{
  696. Success: true,
  697. Status: "undelayed peer port tx/rx",
  698. }
  699. }