sslobodr | d046be8 | 2019-01-16 10:02:22 -0500 | [diff] [blame] | 1 | // Copyright 2015 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 | |
| 15 | package etcdmain |
| 16 | |
| 17 | import ( |
| 18 | "encoding/json" |
| 19 | "fmt" |
| 20 | "io/ioutil" |
| 21 | "net/http" |
| 22 | "os" |
| 23 | "path/filepath" |
| 24 | "reflect" |
| 25 | "runtime" |
| 26 | "strings" |
| 27 | "time" |
| 28 | |
| 29 | "github.com/coreos/etcd/discovery" |
| 30 | "github.com/coreos/etcd/embed" |
| 31 | "github.com/coreos/etcd/etcdserver" |
| 32 | "github.com/coreos/etcd/etcdserver/api/etcdhttp" |
| 33 | "github.com/coreos/etcd/pkg/cors" |
| 34 | "github.com/coreos/etcd/pkg/fileutil" |
| 35 | pkgioutil "github.com/coreos/etcd/pkg/ioutil" |
| 36 | "github.com/coreos/etcd/pkg/osutil" |
| 37 | "github.com/coreos/etcd/pkg/transport" |
| 38 | "github.com/coreos/etcd/pkg/types" |
| 39 | "github.com/coreos/etcd/proxy/httpproxy" |
| 40 | "github.com/coreos/etcd/version" |
| 41 | |
| 42 | "github.com/coreos/pkg/capnslog" |
| 43 | "google.golang.org/grpc" |
| 44 | ) |
| 45 | |
| 46 | type dirType string |
| 47 | |
| 48 | var plog = capnslog.NewPackageLogger("github.com/coreos/etcd", "etcdmain") |
| 49 | |
| 50 | var ( |
| 51 | dirMember = dirType("member") |
| 52 | dirProxy = dirType("proxy") |
| 53 | dirEmpty = dirType("empty") |
| 54 | ) |
| 55 | |
| 56 | func startEtcdOrProxyV2() { |
| 57 | grpc.EnableTracing = false |
| 58 | |
| 59 | cfg := newConfig() |
| 60 | defaultInitialCluster := cfg.ec.InitialCluster |
| 61 | |
| 62 | err := cfg.parse(os.Args[1:]) |
| 63 | if err != nil { |
| 64 | plog.Errorf("error verifying flags, %v. See 'etcd --help'.", err) |
| 65 | switch err { |
| 66 | case embed.ErrUnsetAdvertiseClientURLsFlag: |
| 67 | plog.Errorf("When listening on specific address(es), this etcd process must advertise accessible url(s) to each connected client.") |
| 68 | } |
| 69 | os.Exit(1) |
| 70 | } |
| 71 | cfg.ec.SetupLogging() |
| 72 | |
| 73 | var stopped <-chan struct{} |
| 74 | var errc <-chan error |
| 75 | |
| 76 | plog.Infof("etcd Version: %s\n", version.Version) |
| 77 | plog.Infof("Git SHA: %s\n", version.GitSHA) |
| 78 | plog.Infof("Go Version: %s\n", runtime.Version()) |
| 79 | plog.Infof("Go OS/Arch: %s/%s\n", runtime.GOOS, runtime.GOARCH) |
| 80 | |
| 81 | GoMaxProcs := runtime.GOMAXPROCS(0) |
| 82 | plog.Infof("setting maximum number of CPUs to %d, total number of available CPUs is %d", GoMaxProcs, runtime.NumCPU()) |
| 83 | |
| 84 | defaultHost, dhErr := (&cfg.ec).UpdateDefaultClusterFromName(defaultInitialCluster) |
| 85 | if defaultHost != "" { |
| 86 | plog.Infof("advertising using detected default host %q", defaultHost) |
| 87 | } |
| 88 | if dhErr != nil { |
| 89 | plog.Noticef("failed to detect default host (%v)", dhErr) |
| 90 | } |
| 91 | |
| 92 | if cfg.ec.Dir == "" { |
| 93 | cfg.ec.Dir = fmt.Sprintf("%v.etcd", cfg.ec.Name) |
| 94 | plog.Warningf("no data-dir provided, using default data-dir ./%s", cfg.ec.Dir) |
| 95 | } |
| 96 | |
| 97 | which := identifyDataDirOrDie(cfg.ec.Dir) |
| 98 | if which != dirEmpty { |
| 99 | plog.Noticef("the server is already initialized as %v before, starting as etcd %v...", which, which) |
| 100 | switch which { |
| 101 | case dirMember: |
| 102 | stopped, errc, err = startEtcd(&cfg.ec) |
| 103 | case dirProxy: |
| 104 | err = startProxy(cfg) |
| 105 | default: |
| 106 | plog.Panicf("unhandled dir type %v", which) |
| 107 | } |
| 108 | } else { |
| 109 | shouldProxy := cfg.isProxy() |
| 110 | if !shouldProxy { |
| 111 | stopped, errc, err = startEtcd(&cfg.ec) |
| 112 | if derr, ok := err.(*etcdserver.DiscoveryError); ok && derr.Err == discovery.ErrFullCluster { |
| 113 | if cfg.shouldFallbackToProxy() { |
| 114 | plog.Noticef("discovery cluster full, falling back to %s", fallbackFlagProxy) |
| 115 | shouldProxy = true |
| 116 | } |
| 117 | } |
| 118 | } |
| 119 | if shouldProxy { |
| 120 | err = startProxy(cfg) |
| 121 | } |
| 122 | } |
| 123 | |
| 124 | if err != nil { |
| 125 | if derr, ok := err.(*etcdserver.DiscoveryError); ok { |
| 126 | switch derr.Err { |
| 127 | case discovery.ErrDuplicateID: |
| 128 | plog.Errorf("member %q has previously registered with discovery service token (%s).", cfg.ec.Name, cfg.ec.Durl) |
| 129 | plog.Errorf("But etcd could not find valid cluster configuration in the given data dir (%s).", cfg.ec.Dir) |
| 130 | plog.Infof("Please check the given data dir path if the previous bootstrap succeeded") |
| 131 | plog.Infof("or use a new discovery token if the previous bootstrap failed.") |
| 132 | case discovery.ErrDuplicateName: |
| 133 | plog.Errorf("member with duplicated name has registered with discovery service token(%s).", cfg.ec.Durl) |
| 134 | plog.Errorf("please check (cURL) the discovery token for more information.") |
| 135 | plog.Errorf("please do not reuse the discovery token and generate a new one to bootstrap the cluster.") |
| 136 | default: |
| 137 | plog.Errorf("%v", err) |
| 138 | plog.Infof("discovery token %s was used, but failed to bootstrap the cluster.", cfg.ec.Durl) |
| 139 | plog.Infof("please generate a new discovery token and try to bootstrap again.") |
| 140 | } |
| 141 | os.Exit(1) |
| 142 | } |
| 143 | |
| 144 | if strings.Contains(err.Error(), "include") && strings.Contains(err.Error(), "--initial-cluster") { |
| 145 | plog.Infof("%v", err) |
| 146 | if cfg.ec.InitialCluster == cfg.ec.InitialClusterFromName(cfg.ec.Name) { |
| 147 | plog.Infof("forgot to set --initial-cluster flag?") |
| 148 | } |
| 149 | if types.URLs(cfg.ec.APUrls).String() == embed.DefaultInitialAdvertisePeerURLs { |
| 150 | plog.Infof("forgot to set --initial-advertise-peer-urls flag?") |
| 151 | } |
| 152 | if cfg.ec.InitialCluster == cfg.ec.InitialClusterFromName(cfg.ec.Name) && len(cfg.ec.Durl) == 0 { |
| 153 | plog.Infof("if you want to use discovery service, please set --discovery flag.") |
| 154 | } |
| 155 | os.Exit(1) |
| 156 | } |
| 157 | plog.Fatalf("%v", err) |
| 158 | } |
| 159 | |
| 160 | osutil.HandleInterrupts() |
| 161 | |
| 162 | // At this point, the initialization of etcd is done. |
| 163 | // The listeners are listening on the TCP ports and ready |
| 164 | // for accepting connections. The etcd instance should be |
| 165 | // joined with the cluster and ready to serve incoming |
| 166 | // connections. |
| 167 | notifySystemd() |
| 168 | |
| 169 | select { |
| 170 | case lerr := <-errc: |
| 171 | // fatal out on listener errors |
| 172 | plog.Fatal(lerr) |
| 173 | case <-stopped: |
| 174 | } |
| 175 | |
| 176 | osutil.Exit(0) |
| 177 | } |
| 178 | |
| 179 | // startEtcd runs StartEtcd in addition to hooks needed for standalone etcd. |
| 180 | func startEtcd(cfg *embed.Config) (<-chan struct{}, <-chan error, error) { |
| 181 | e, err := embed.StartEtcd(cfg) |
| 182 | if err != nil { |
| 183 | return nil, nil, err |
| 184 | } |
| 185 | osutil.RegisterInterruptHandler(e.Close) |
| 186 | select { |
| 187 | case <-e.Server.ReadyNotify(): // wait for e.Server to join the cluster |
| 188 | case <-e.Server.StopNotify(): // publish aborted from 'ErrStopped' |
| 189 | } |
| 190 | return e.Server.StopNotify(), e.Err(), nil |
| 191 | } |
| 192 | |
| 193 | // startProxy launches an HTTP proxy for client communication which proxies to other etcd nodes. |
| 194 | func startProxy(cfg *config) error { |
| 195 | plog.Notice("proxy: this proxy supports v2 API only!") |
| 196 | |
| 197 | clientTLSInfo := cfg.ec.ClientTLSInfo |
| 198 | if clientTLSInfo.Empty() { |
| 199 | // Support old proxy behavior of defaulting to PeerTLSInfo |
| 200 | // for both client and peer connections. |
| 201 | clientTLSInfo = cfg.ec.PeerTLSInfo |
| 202 | } |
| 203 | clientTLSInfo.InsecureSkipVerify = cfg.ec.ClientAutoTLS |
| 204 | cfg.ec.PeerTLSInfo.InsecureSkipVerify = cfg.ec.PeerAutoTLS |
| 205 | |
| 206 | pt, err := transport.NewTimeoutTransport(clientTLSInfo, time.Duration(cfg.cp.ProxyDialTimeoutMs)*time.Millisecond, time.Duration(cfg.cp.ProxyReadTimeoutMs)*time.Millisecond, time.Duration(cfg.cp.ProxyWriteTimeoutMs)*time.Millisecond) |
| 207 | if err != nil { |
| 208 | return err |
| 209 | } |
| 210 | pt.MaxIdleConnsPerHost = httpproxy.DefaultMaxIdleConnsPerHost |
| 211 | |
| 212 | if err = cfg.ec.PeerSelfCert(); err != nil { |
| 213 | plog.Fatalf("could not get certs (%v)", err) |
| 214 | } |
| 215 | tr, err := transport.NewTimeoutTransport(cfg.ec.PeerTLSInfo, time.Duration(cfg.cp.ProxyDialTimeoutMs)*time.Millisecond, time.Duration(cfg.cp.ProxyReadTimeoutMs)*time.Millisecond, time.Duration(cfg.cp.ProxyWriteTimeoutMs)*time.Millisecond) |
| 216 | if err != nil { |
| 217 | return err |
| 218 | } |
| 219 | |
| 220 | cfg.ec.Dir = filepath.Join(cfg.ec.Dir, "proxy") |
| 221 | err = os.MkdirAll(cfg.ec.Dir, fileutil.PrivateDirMode) |
| 222 | if err != nil { |
| 223 | return err |
| 224 | } |
| 225 | |
| 226 | var peerURLs []string |
| 227 | clusterfile := filepath.Join(cfg.ec.Dir, "cluster") |
| 228 | |
| 229 | b, err := ioutil.ReadFile(clusterfile) |
| 230 | switch { |
| 231 | case err == nil: |
| 232 | if cfg.ec.Durl != "" { |
| 233 | plog.Warningf("discovery token ignored since the proxy has already been initialized. Valid cluster file found at %q", clusterfile) |
| 234 | } |
| 235 | if cfg.ec.DNSCluster != "" { |
| 236 | plog.Warningf("DNS SRV discovery ignored since the proxy has already been initialized. Valid cluster file found at %q", clusterfile) |
| 237 | } |
| 238 | urls := struct{ PeerURLs []string }{} |
| 239 | err = json.Unmarshal(b, &urls) |
| 240 | if err != nil { |
| 241 | return err |
| 242 | } |
| 243 | peerURLs = urls.PeerURLs |
| 244 | plog.Infof("proxy: using peer urls %v from cluster file %q", peerURLs, clusterfile) |
| 245 | case os.IsNotExist(err): |
| 246 | var urlsmap types.URLsMap |
| 247 | urlsmap, _, err = cfg.ec.PeerURLsMapAndToken("proxy") |
| 248 | if err != nil { |
| 249 | return fmt.Errorf("error setting up initial cluster: %v", err) |
| 250 | } |
| 251 | |
| 252 | if cfg.ec.Durl != "" { |
| 253 | var s string |
| 254 | s, err = discovery.GetCluster(cfg.ec.Durl, cfg.ec.Dproxy) |
| 255 | if err != nil { |
| 256 | return err |
| 257 | } |
| 258 | if urlsmap, err = types.NewURLsMap(s); err != nil { |
| 259 | return err |
| 260 | } |
| 261 | } |
| 262 | peerURLs = urlsmap.URLs() |
| 263 | plog.Infof("proxy: using peer urls %v ", peerURLs) |
| 264 | default: |
| 265 | return err |
| 266 | } |
| 267 | |
| 268 | clientURLs := []string{} |
| 269 | uf := func() []string { |
| 270 | gcls, gerr := etcdserver.GetClusterFromRemotePeers(peerURLs, tr) |
| 271 | |
| 272 | if gerr != nil { |
| 273 | plog.Warningf("proxy: %v", gerr) |
| 274 | return []string{} |
| 275 | } |
| 276 | |
| 277 | clientURLs = gcls.ClientURLs() |
| 278 | |
| 279 | urls := struct{ PeerURLs []string }{gcls.PeerURLs()} |
| 280 | b, jerr := json.Marshal(urls) |
| 281 | if jerr != nil { |
| 282 | plog.Warningf("proxy: error on marshal peer urls %s", jerr) |
| 283 | return clientURLs |
| 284 | } |
| 285 | |
| 286 | err = pkgioutil.WriteAndSyncFile(clusterfile+".bak", b, 0600) |
| 287 | if err != nil { |
| 288 | plog.Warningf("proxy: error on writing urls %s", err) |
| 289 | return clientURLs |
| 290 | } |
| 291 | err = os.Rename(clusterfile+".bak", clusterfile) |
| 292 | if err != nil { |
| 293 | plog.Warningf("proxy: error on updating clusterfile %s", err) |
| 294 | return clientURLs |
| 295 | } |
| 296 | if !reflect.DeepEqual(gcls.PeerURLs(), peerURLs) { |
| 297 | plog.Noticef("proxy: updated peer urls in cluster file from %v to %v", peerURLs, gcls.PeerURLs()) |
| 298 | } |
| 299 | peerURLs = gcls.PeerURLs() |
| 300 | |
| 301 | return clientURLs |
| 302 | } |
| 303 | ph := httpproxy.NewHandler(pt, uf, time.Duration(cfg.cp.ProxyFailureWaitMs)*time.Millisecond, time.Duration(cfg.cp.ProxyRefreshIntervalMs)*time.Millisecond) |
| 304 | ph = &cors.CORSHandler{ |
| 305 | Handler: ph, |
| 306 | Info: cfg.ec.CorsInfo, |
| 307 | } |
| 308 | |
| 309 | if cfg.isReadonlyProxy() { |
| 310 | ph = httpproxy.NewReadonlyHandler(ph) |
| 311 | } |
| 312 | |
| 313 | // setup self signed certs when serving https |
| 314 | cHosts, cTLS := []string{}, false |
| 315 | for _, u := range cfg.ec.LCUrls { |
| 316 | cHosts = append(cHosts, u.Host) |
| 317 | cTLS = cTLS || u.Scheme == "https" |
| 318 | } |
| 319 | for _, u := range cfg.ec.ACUrls { |
| 320 | cHosts = append(cHosts, u.Host) |
| 321 | cTLS = cTLS || u.Scheme == "https" |
| 322 | } |
| 323 | listenerTLS := cfg.ec.ClientTLSInfo |
| 324 | if cfg.ec.ClientAutoTLS && cTLS { |
| 325 | listenerTLS, err = transport.SelfCert(filepath.Join(cfg.ec.Dir, "clientCerts"), cHosts) |
| 326 | if err != nil { |
| 327 | plog.Fatalf("proxy: could not initialize self-signed client certs (%v)", err) |
| 328 | } |
| 329 | } |
| 330 | |
| 331 | // Start a proxy server goroutine for each listen address |
| 332 | for _, u := range cfg.ec.LCUrls { |
| 333 | l, err := transport.NewListener(u.Host, u.Scheme, &listenerTLS) |
| 334 | if err != nil { |
| 335 | return err |
| 336 | } |
| 337 | |
| 338 | host := u.String() |
| 339 | go func() { |
| 340 | plog.Info("proxy: listening for client requests on ", host) |
| 341 | mux := http.NewServeMux() |
| 342 | etcdhttp.HandlePrometheus(mux) // v2 proxy just uses the same port |
| 343 | mux.Handle("/", ph) |
| 344 | plog.Fatal(http.Serve(l, mux)) |
| 345 | }() |
| 346 | } |
| 347 | return nil |
| 348 | } |
| 349 | |
| 350 | // identifyDataDirOrDie returns the type of the data dir. |
| 351 | // Dies if the datadir is invalid. |
| 352 | func identifyDataDirOrDie(dir string) dirType { |
| 353 | names, err := fileutil.ReadDir(dir) |
| 354 | if err != nil { |
| 355 | if os.IsNotExist(err) { |
| 356 | return dirEmpty |
| 357 | } |
| 358 | plog.Fatalf("error listing data dir: %s", dir) |
| 359 | } |
| 360 | |
| 361 | var m, p bool |
| 362 | for _, name := range names { |
| 363 | switch dirType(name) { |
| 364 | case dirMember: |
| 365 | m = true |
| 366 | case dirProxy: |
| 367 | p = true |
| 368 | default: |
| 369 | plog.Warningf("found invalid file/dir %s under data dir %s (Ignore this if you are upgrading etcd)", name, dir) |
| 370 | } |
| 371 | } |
| 372 | |
| 373 | if m && p { |
| 374 | plog.Fatal("invalid datadir. Both member and proxy directories exist.") |
| 375 | } |
| 376 | if m { |
| 377 | return dirMember |
| 378 | } |
| 379 | if p { |
| 380 | return dirProxy |
| 381 | } |
| 382 | return dirEmpty |
| 383 | } |
| 384 | |
| 385 | func checkSupportArch() { |
| 386 | // TODO qualify arm64 |
| 387 | if runtime.GOARCH == "amd64" || runtime.GOARCH == "ppc64le" { |
| 388 | return |
| 389 | } |
| 390 | // unsupported arch only configured via environment variable |
| 391 | // so unset here to not parse through flag |
| 392 | defer os.Unsetenv("ETCD_UNSUPPORTED_ARCH") |
| 393 | if env, ok := os.LookupEnv("ETCD_UNSUPPORTED_ARCH"); ok && env == runtime.GOARCH { |
| 394 | plog.Warningf("running etcd on unsupported architecture %q since ETCD_UNSUPPORTED_ARCH is set", env) |
| 395 | return |
| 396 | } |
| 397 | plog.Errorf("etcd on unsupported platform without ETCD_UNSUPPORTED_ARCH=%s set.", runtime.GOARCH) |
| 398 | os.Exit(1) |
| 399 | } |