handler.go 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406
  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. "net/url"
  19. "os"
  20. "os/exec"
  21. "syscall"
  22. "time"
  23. "github.com/coreos/etcd/pkg/fileutil"
  24. "github.com/coreos/etcd/pkg/proxy"
  25. "github.com/coreos/etcd/tools/functional-tester/rpcpb"
  26. "go.uber.org/zap"
  27. )
  28. // return error for system errors (e.g. fail to create files)
  29. // return status error in response for wrong configuration/operation (e.g. start etcd twice)
  30. func (srv *Server) handleTesterRequest(req *rpcpb.Request) (resp *rpcpb.Response, err error) {
  31. defer func() {
  32. if err == nil {
  33. srv.last = req.Operation
  34. srv.lg.Info("handler success", zap.String("operation", req.Operation.String()))
  35. }
  36. }()
  37. switch req.Operation {
  38. case rpcpb.Operation_InitialStartEtcd:
  39. return srv.handleInitialStartEtcd(req)
  40. case rpcpb.Operation_RestartEtcd:
  41. return srv.handleRestartEtcd()
  42. case rpcpb.Operation_KillEtcd:
  43. return srv.handleKillEtcd()
  44. case rpcpb.Operation_FailArchive:
  45. return srv.handleFailArchive()
  46. case rpcpb.Operation_DestroyEtcdAgent:
  47. return srv.handleDestroyEtcdAgent()
  48. case rpcpb.Operation_BlackholePeerPortTxRx:
  49. return srv.handleBlackholePeerPortTxRx()
  50. case rpcpb.Operation_UnblackholePeerPortTxRx:
  51. return srv.handleUnblackholePeerPortTxRx()
  52. case rpcpb.Operation_DelayPeerPortTxRx:
  53. return srv.handleDelayPeerPortTxRx()
  54. case rpcpb.Operation_UndelayPeerPortTxRx:
  55. return srv.handleUndelayPeerPortTxRx()
  56. default:
  57. msg := fmt.Sprintf("operation not found (%v)", req.Operation)
  58. return &rpcpb.Response{Success: false, Status: msg}, errors.New(msg)
  59. }
  60. }
  61. func (srv *Server) handleInitialStartEtcd(req *rpcpb.Request) (*rpcpb.Response, error) {
  62. if srv.last != rpcpb.Operation_NotStarted {
  63. return &rpcpb.Response{
  64. Success: false,
  65. Status: fmt.Sprintf("%q is not valid; last server operation was %q", rpcpb.Operation_InitialStartEtcd.String(), srv.last.String()),
  66. }, nil
  67. }
  68. srv.Member = req.Member
  69. srv.Tester = req.Tester
  70. err := fileutil.TouchDirAll(srv.Member.BaseDir)
  71. if err != nil {
  72. return nil, err
  73. }
  74. srv.lg.Info("created base directory", zap.String("path", srv.Member.BaseDir))
  75. if err = srv.createEtcdFile(); err != nil {
  76. return nil, err
  77. }
  78. srv.creatEtcdCmd()
  79. err = srv.startEtcdCmd()
  80. if err != nil {
  81. return nil, err
  82. }
  83. srv.lg.Info("started etcd", zap.String("command-path", srv.etcdCmd.Path))
  84. // wait some time for etcd listener start
  85. // before setting up proxy
  86. time.Sleep(time.Second)
  87. if err = srv.startProxy(); err != nil {
  88. return nil, err
  89. }
  90. return &rpcpb.Response{
  91. Success: true,
  92. Status: "successfully started etcd!",
  93. }, nil
  94. }
  95. func (srv *Server) startProxy() error {
  96. if srv.Member.EtcdClientProxy {
  97. advertiseClientURL, advertiseClientURLPort, err := getURLAndPort(srv.Member.Etcd.AdvertiseClientURLs[0])
  98. if err != nil {
  99. return err
  100. }
  101. listenClientURL, _, err := getURLAndPort(srv.Member.Etcd.ListenClientURLs[0])
  102. if err != nil {
  103. return err
  104. }
  105. srv.advertiseClientPortToProxy[advertiseClientURLPort] = proxy.NewServer(proxy.ServerConfig{
  106. Logger: srv.lg,
  107. From: *advertiseClientURL,
  108. To: *listenClientURL,
  109. })
  110. select {
  111. case err = <-srv.advertiseClientPortToProxy[advertiseClientURLPort].Error():
  112. return err
  113. case <-time.After(2 * time.Second):
  114. srv.lg.Info("started proxy on client traffic", zap.String("url", advertiseClientURL.String()))
  115. }
  116. }
  117. if srv.Member.EtcdPeerProxy {
  118. advertisePeerURL, advertisePeerURLPort, err := getURLAndPort(srv.Member.Etcd.AdvertisePeerURLs[0])
  119. if err != nil {
  120. return err
  121. }
  122. listenPeerURL, _, err := getURLAndPort(srv.Member.Etcd.ListenPeerURLs[0])
  123. if err != nil {
  124. return err
  125. }
  126. srv.advertisePeerPortToProxy[advertisePeerURLPort] = proxy.NewServer(proxy.ServerConfig{
  127. Logger: srv.lg,
  128. From: *advertisePeerURL,
  129. To: *listenPeerURL,
  130. })
  131. select {
  132. case err = <-srv.advertisePeerPortToProxy[advertisePeerURLPort].Error():
  133. return err
  134. case <-time.After(2 * time.Second):
  135. srv.lg.Info("started proxy on peer traffic", zap.String("url", advertisePeerURL.String()))
  136. }
  137. }
  138. return nil
  139. }
  140. func (srv *Server) stopProxy() {
  141. if srv.Member.EtcdClientProxy && len(srv.advertiseClientPortToProxy) > 0 {
  142. for port, px := range srv.advertiseClientPortToProxy {
  143. if err := px.Close(); err != nil {
  144. srv.lg.Warn("failed to close proxy", zap.Int("port", port))
  145. continue
  146. }
  147. select {
  148. case <-px.Done():
  149. // enough time to release port
  150. time.Sleep(time.Second)
  151. case <-time.After(time.Second):
  152. }
  153. srv.lg.Info("closed proxy",
  154. zap.Int("port", port),
  155. zap.String("from", px.From()),
  156. zap.String("to", px.To()),
  157. )
  158. }
  159. srv.advertiseClientPortToProxy = make(map[int]proxy.Server)
  160. }
  161. if srv.Member.EtcdPeerProxy && len(srv.advertisePeerPortToProxy) > 0 {
  162. for port, px := range srv.advertisePeerPortToProxy {
  163. if err := px.Close(); err != nil {
  164. srv.lg.Warn("failed to close proxy", zap.Int("port", port))
  165. continue
  166. }
  167. select {
  168. case <-px.Done():
  169. // enough time to release port
  170. time.Sleep(time.Second)
  171. case <-time.After(time.Second):
  172. }
  173. srv.lg.Info("closed proxy",
  174. zap.Int("port", port),
  175. zap.String("from", px.From()),
  176. zap.String("to", px.To()),
  177. )
  178. }
  179. srv.advertisePeerPortToProxy = make(map[int]proxy.Server)
  180. }
  181. }
  182. func (srv *Server) createEtcdFile() error {
  183. var err error
  184. srv.etcdLogFile, err = os.Create(srv.Member.EtcdLogPath)
  185. if err != nil {
  186. return err
  187. }
  188. srv.lg.Info("created etcd log file", zap.String("path", srv.Member.EtcdLogPath))
  189. return nil
  190. }
  191. func (srv *Server) creatEtcdCmd() {
  192. etcdPath, etcdFlags := srv.Member.EtcdExecPath, srv.Member.Etcd.Flags()
  193. u, _ := url.Parse(srv.Member.FailpointHTTPAddr)
  194. srv.lg.Info("creating etcd command",
  195. zap.String("etcd-exec-path", etcdPath),
  196. zap.Strings("etcd-flags", etcdFlags),
  197. zap.String("failpoint-http-addr", srv.Member.FailpointHTTPAddr),
  198. zap.String("failpoint-addr", u.Host),
  199. )
  200. srv.etcdCmd = exec.Command(etcdPath, etcdFlags...)
  201. srv.etcdCmd.Env = []string{"GOFAIL_HTTP=" + u.Host}
  202. srv.etcdCmd.Stdout = srv.etcdLogFile
  203. srv.etcdCmd.Stderr = srv.etcdLogFile
  204. }
  205. // start but do not wait for it to complete
  206. func (srv *Server) startEtcdCmd() error {
  207. return srv.etcdCmd.Start()
  208. }
  209. func (srv *Server) handleRestartEtcd() (*rpcpb.Response, error) {
  210. srv.creatEtcdCmd()
  211. srv.lg.Info("restarting etcd")
  212. err := srv.startEtcdCmd()
  213. if err != nil {
  214. return nil, err
  215. }
  216. srv.lg.Info("restarted etcd", zap.String("command-path", srv.etcdCmd.Path))
  217. // wait some time for etcd listener start
  218. // before setting up proxy
  219. // TODO: local tests should handle port conflicts
  220. // with clients on restart
  221. time.Sleep(time.Second)
  222. if err = srv.startProxy(); err != nil {
  223. return nil, err
  224. }
  225. return &rpcpb.Response{
  226. Success: true,
  227. Status: "successfully restarted etcd!",
  228. }, nil
  229. }
  230. func (srv *Server) handleKillEtcd() (*rpcpb.Response, error) {
  231. srv.stopProxy()
  232. err := stopWithSig(srv.etcdCmd, syscall.SIGTERM)
  233. if err != nil {
  234. return nil, err
  235. }
  236. srv.lg.Info("killed etcd", zap.String("signal", syscall.SIGTERM.String()))
  237. return &rpcpb.Response{
  238. Success: true,
  239. Status: "successfully killed etcd!",
  240. }, nil
  241. }
  242. func (srv *Server) handleFailArchive() (*rpcpb.Response, error) {
  243. srv.stopProxy()
  244. // exit with stackstrace
  245. err := stopWithSig(srv.etcdCmd, syscall.SIGQUIT)
  246. if err != nil {
  247. return nil, err
  248. }
  249. srv.lg.Info("killed etcd", zap.String("signal", syscall.SIGQUIT.String()))
  250. srv.etcdLogFile.Sync()
  251. srv.etcdLogFile.Close()
  252. // TODO: support separate WAL directory
  253. if err = archive(
  254. srv.Member.BaseDir,
  255. srv.Member.EtcdLogPath,
  256. srv.Member.Etcd.DataDir,
  257. ); err != nil {
  258. return nil, err
  259. }
  260. srv.lg.Info("archived data", zap.String("base-dir", srv.Member.BaseDir))
  261. if err = srv.createEtcdFile(); err != nil {
  262. return nil, err
  263. }
  264. srv.lg.Info("cleaning up page cache")
  265. if err := cleanPageCache(); err != nil {
  266. srv.lg.Warn("failed to clean up page cache", zap.String("error", err.Error()))
  267. }
  268. srv.lg.Info("cleaned up page cache")
  269. return &rpcpb.Response{
  270. Success: true,
  271. Status: "successfully cleaned up etcd!",
  272. }, nil
  273. }
  274. // stop proxy, etcd, delete data directory
  275. func (srv *Server) handleDestroyEtcdAgent() (*rpcpb.Response, error) {
  276. err := stopWithSig(srv.etcdCmd, syscall.SIGTERM)
  277. if err != nil {
  278. return nil, err
  279. }
  280. srv.lg.Info("killed etcd", zap.String("signal", syscall.SIGTERM.String()))
  281. err = os.RemoveAll(srv.Member.BaseDir)
  282. if err != nil {
  283. return nil, err
  284. }
  285. srv.lg.Info("removed base directory", zap.String("dir", srv.Member.BaseDir))
  286. // stop agent server
  287. srv.Stop()
  288. for port, px := range srv.advertiseClientPortToProxy {
  289. err := px.Close()
  290. srv.lg.Info("closed proxy", zap.Int("client-port", port), zap.Error(err))
  291. }
  292. for port, px := range srv.advertisePeerPortToProxy {
  293. err := px.Close()
  294. srv.lg.Info("closed proxy", zap.Int("peer-port", port), zap.Error(err))
  295. }
  296. return &rpcpb.Response{
  297. Success: true,
  298. Status: "successfully destroyed etcd and agent!",
  299. }, nil
  300. }
  301. func (srv *Server) handleBlackholePeerPortTxRx() (*rpcpb.Response, error) {
  302. for port, px := range srv.advertisePeerPortToProxy {
  303. srv.lg.Info("blackholing", zap.Int("peer-port", port))
  304. px.BlackholeTx()
  305. px.BlackholeRx()
  306. srv.lg.Info("blackholed", zap.Int("peer-port", port))
  307. }
  308. return &rpcpb.Response{
  309. Success: true,
  310. Status: "successfully blackholed peer port tx/rx!",
  311. }, nil
  312. }
  313. func (srv *Server) handleUnblackholePeerPortTxRx() (*rpcpb.Response, error) {
  314. for port, px := range srv.advertisePeerPortToProxy {
  315. srv.lg.Info("unblackholing", zap.Int("peer-port", port))
  316. px.UnblackholeTx()
  317. px.UnblackholeRx()
  318. srv.lg.Info("unblackholed", zap.Int("peer-port", port))
  319. }
  320. return &rpcpb.Response{
  321. Success: true,
  322. Status: "successfully unblackholed peer port tx/rx!",
  323. }, nil
  324. }
  325. func (srv *Server) handleDelayPeerPortTxRx() (*rpcpb.Response, error) {
  326. lat := time.Duration(srv.Tester.UpdatedDelayLatencyMs) * time.Millisecond
  327. rv := time.Duration(srv.Tester.DelayLatencyMsRv) * time.Millisecond
  328. for port, px := range srv.advertisePeerPortToProxy {
  329. srv.lg.Info("delaying",
  330. zap.Int("peer-port", port),
  331. zap.Duration("latency", lat),
  332. zap.Duration("random-variable", rv),
  333. )
  334. px.DelayTx(lat, rv)
  335. px.DelayRx(lat, rv)
  336. srv.lg.Info("delayed",
  337. zap.Int("peer-port", port),
  338. zap.Duration("latency", lat),
  339. zap.Duration("random-variable", rv),
  340. )
  341. }
  342. return &rpcpb.Response{
  343. Success: true,
  344. Status: "successfully delay peer port tx/rx!",
  345. }, nil
  346. }
  347. func (srv *Server) handleUndelayPeerPortTxRx() (*rpcpb.Response, error) {
  348. for port, px := range srv.advertisePeerPortToProxy {
  349. srv.lg.Info("undelaying", zap.Int("peer-port", port))
  350. px.UndelayTx()
  351. px.UndelayRx()
  352. srv.lg.Info("undelayed", zap.Int("peer-port", port))
  353. }
  354. return &rpcpb.Response{
  355. Success: true,
  356. Status: "successfully undelay peer port tx/rx!",
  357. }, nil
  358. }