sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 1 | /* |
| 2 | * Copyright 2018-present Open Networking Foundation |
| 3 | |
| 4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | * you may not use this file except in compliance with the License. |
| 6 | * You may obtain a copy of the License at |
| 7 | |
| 8 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | |
| 10 | * Unless required by applicable law or agreed to in writing, software |
| 11 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | * See the License for the specific language governing permissions and |
| 14 | * limitations under the License. |
| 15 | */ |
| 16 | |
| 17 | package main |
| 18 | |
| 19 | import ( |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 20 | "time" |
| 21 | "regexp" |
| 22 | "errors" |
| 23 | "strconv" |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 24 | |
| 25 | "k8s.io/client-go/rest" |
| 26 | "google.golang.org/grpc" |
| 27 | "golang.org/x/net/context" |
| 28 | "k8s.io/client-go/kubernetes" |
| 29 | "github.com/golang/protobuf/ptypes" |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 30 | "github.com/opencord/voltha-go/common/log" |
| 31 | kafka "github.com/opencord/voltha-go/kafka" |
| 32 | metav1 "k8s.io/apimachinery/pkg/apis/meta/v1" |
| 33 | empty "github.com/golang/protobuf/ptypes/empty" |
| 34 | vpb "github.com/opencord/voltha-go/protos/voltha" |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 35 | cmn "github.com/opencord/voltha-go/protos/common" |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 36 | pb "github.com/opencord/voltha-go/protos/afrouter" |
| 37 | ic "github.com/opencord/voltha-go/protos/inter_container" |
| 38 | ) |
| 39 | |
| 40 | type configConn struct { |
| 41 | Server string `json:"Server"` |
| 42 | Cluster string `json:"Cluster"` |
| 43 | Backend string `json:"Backend"` |
| 44 | connections map[string]connection |
| 45 | } |
| 46 | |
| 47 | type connection struct { |
| 48 | Name string `json:"Connection"` |
| 49 | Addr string `json:"Addr"` |
| 50 | Port uint64 `json:"Port"` |
| 51 | } |
| 52 | |
| 53 | type rwPod struct { |
| 54 | name string |
| 55 | ipAddr string |
| 56 | node string |
| 57 | devIds map[string]struct{} |
| 58 | backend string |
| 59 | connection string |
| 60 | } |
| 61 | |
| 62 | type podTrack struct { |
| 63 | pod *rwPod |
| 64 | dn bool |
| 65 | } |
| 66 | |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 67 | var nPods int = 6 |
| 68 | |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 69 | // Topic is affinityRouter |
| 70 | // port: 9092 |
| 71 | |
| 72 | func newKafkaClient(clientType string, host string, port int, instanceID string) (kafka.Client, error) { |
| 73 | |
| 74 | log.Infow("kafka-client-type", log.Fields{"client": clientType}) |
| 75 | switch clientType { |
| 76 | case "sarama": |
| 77 | return kafka.NewSaramaClient( |
| 78 | kafka.Host(host), |
| 79 | kafka.Port(port), |
| 80 | kafka.ConsumerType(kafka.GroupCustomer), |
| 81 | kafka.ProducerReturnOnErrors(true), |
| 82 | kafka.ProducerReturnOnSuccess(true), |
| 83 | kafka.ProducerMaxRetries(6), |
| 84 | kafka.NumPartitions(3), |
| 85 | kafka.ConsumerGroupName(instanceID), |
| 86 | kafka.ConsumerGroupPrefix(instanceID), |
| 87 | kafka.AutoCreateTopic(false), |
| 88 | kafka.ProducerFlushFrequency(5), |
| 89 | kafka.ProducerRetryBackoff(time.Millisecond*30)), nil |
| 90 | } |
| 91 | return nil, errors.New("unsupported-client-type") |
| 92 | } |
| 93 | |
| 94 | |
| 95 | func k8sClientSet() *kubernetes.Clientset { |
| 96 | // creates the in-cluster config |
| 97 | config, err := rest.InClusterConfig() |
| 98 | if err != nil { |
| 99 | panic(err.Error()) |
| 100 | } |
| 101 | // creates the clientset |
| 102 | clientset, err := kubernetes.NewForConfig(config) |
| 103 | if err != nil { |
| 104 | panic(err.Error()) |
| 105 | } |
| 106 | |
| 107 | return clientset |
| 108 | } |
| 109 | |
| 110 | |
| 111 | func connect(addr string) (*grpc.ClientConn, error) { |
| 112 | for ctr :=0 ; ctr < 100; ctr++ { |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 113 | log.Debugf("Trying to connect to %s", addr) |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 114 | conn, err := grpc.Dial(addr, grpc.WithInsecure()) |
| 115 | if err != nil { |
| 116 | log.Debugf("Attempt to connect failed, retrying %v:", err) |
| 117 | } else { |
| 118 | log.Debugf("Connection succeeded") |
| 119 | return conn,err |
| 120 | } |
| 121 | time.Sleep(10 * time.Second) |
| 122 | } |
| 123 | log.Debugf("Too many connection attempts, giving up!") |
| 124 | return nil,errors.New("Timeout attempting to conect") |
| 125 | } |
| 126 | |
| 127 | func getRwPods(cs *kubernetes.Clientset, coreFilter * regexp.Regexp) []*rwPod { |
| 128 | var rtrn []*rwPod |
| 129 | |
| 130 | pods, err := cs.CoreV1().Pods("").List(metav1.ListOptions{}) |
| 131 | if err != nil { |
| 132 | panic(err.Error()) |
| 133 | } |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 134 | //log.Debugf("There are a total of %d pods in the cluster\n", len(pods.Items)) |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 135 | |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 136 | for _,v := range pods.Items { |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 137 | if v.Namespace == "voltha" && coreFilter.MatchString(v.Name) { |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 138 | log.Debugf("Namespace: %s, PodName: %s, PodIP: %s, Host: %s\n", v.Namespace, v.Name, |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 139 | v.Status.PodIP, v.Spec.NodeName) |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 140 | // Only add the pod if it has an IP address. If it doesn't then it likely crashed and |
| 141 | // and is still in the process of getting re-started. |
| 142 | if v.Status.PodIP != "" { |
| 143 | rtrn = append(rtrn, &rwPod{name:v.Name,ipAddr:v.Status.PodIP,node:v.Spec.NodeName, |
| 144 | devIds:make(map[string]struct{}), backend:"", connection:""}) |
| 145 | } |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 146 | } |
| 147 | } |
| 148 | return rtrn |
| 149 | } |
| 150 | |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 151 | func reconcilePodDeviceIds(pod * rwPod, ids map[string]struct{}) bool { |
| 152 | var idList cmn.IDs |
sslobodr | 6c1689c | 2019-01-24 07:31:15 -0500 | [diff] [blame] | 153 | for k,_ := range ids { |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 154 | idList.Items = append(idList.Items, &cmn.ID{Id:k}) |
| 155 | } |
| 156 | conn,err := connect(pod.ipAddr+":50057") |
| 157 | defer conn.Close() |
sslobodr | 6c1689c | 2019-01-24 07:31:15 -0500 | [diff] [blame] | 158 | if err != nil { |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 159 | log.Debugf("Could not query devices from %s, could not connect", pod.name) |
| 160 | return false |
| 161 | } |
| 162 | client := vpb.NewVolthaServiceClient(conn) |
| 163 | _,err = client.ReconcileDevices(context.Background(), &idList) |
| 164 | if err != nil { |
| 165 | log.Error(err) |
| 166 | return false |
| 167 | } |
| 168 | |
| 169 | return true |
| 170 | } |
| 171 | |
| 172 | func queryPodDeviceIds(pod * rwPod) map[string]struct{} { |
| 173 | var rtrn map[string]struct{} = make(map[string]struct{}) |
| 174 | // Open a connection to the pod |
| 175 | // port 50057 |
| 176 | conn, err := connect(pod.ipAddr+":50057") |
sslobodr | 6c1689c | 2019-01-24 07:31:15 -0500 | [diff] [blame] | 177 | if err != nil { |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 178 | log.Debugf("Could not query devices from %s, could not connect", pod.name) |
| 179 | return rtrn |
| 180 | } |
| 181 | defer conn.Close() |
| 182 | client := vpb.NewVolthaServiceClient(conn) |
| 183 | devs,err := client.ListDeviceIds(context.Background(), &empty.Empty{}) |
| 184 | if err != nil { |
| 185 | log.Error(err) |
| 186 | return rtrn |
| 187 | } |
| 188 | for _,dv := range devs.Items { |
| 189 | rtrn[dv.Id]=struct{}{} |
| 190 | } |
| 191 | |
| 192 | return rtrn |
| 193 | } |
| 194 | |
| 195 | func queryDeviceIds(pods []*rwPod) { |
| 196 | for pk,_ := range pods { |
| 197 | // Keep the old Id list if a new list is not returned |
| 198 | if idList := queryPodDeviceIds(pods[pk]); len(idList) != 0 { |
| 199 | pods[pk].devIds = idList |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 200 | } |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 201 | } |
| 202 | } |
| 203 | |
| 204 | func allEmpty(pods []*rwPod) bool { |
| 205 | for k,_ := range pods { |
| 206 | if len(pods[k].devIds) != 0 { |
| 207 | return false |
| 208 | } |
| 209 | } |
| 210 | return true |
| 211 | } |
| 212 | |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 213 | func rmPod(pods []*rwPod, idx int) []*rwPod { |
| 214 | return append(pods[:idx],pods[idx+1:]...) |
| 215 | } |
| 216 | |
| 217 | func groupIntersectingPods1(pods []*rwPod, podCt int) ([][]*rwPod,[]*rwPod) { |
| 218 | var rtrn [][]*rwPod |
| 219 | var out []*rwPod |
| 220 | |
| 221 | for { |
| 222 | if len(pods) == 0 { |
| 223 | break |
| 224 | } |
| 225 | if len(pods[0].devIds) == 0 { // Ignore pods with no devices |
| 226 | ////log.Debugf("%s empty pod", pd[k].pod.name) |
| 227 | out = append(out, pods[0]) |
| 228 | pods = rmPod(pods, 0) |
| 229 | continue |
| 230 | } |
| 231 | // Start a pod group with this pod |
| 232 | var grp []*rwPod |
| 233 | grp = append(grp, pods[0]) |
| 234 | pods = rmPod(pods,0) |
| 235 | //log.Debugf("Creating new group %s", pd[k].pod.name) |
| 236 | // Find the peer pod based on device overlap |
| 237 | // It's ok if one isn't found, an empty one will be used instead |
| 238 | for k,_ := range pods { |
| 239 | if len(pods[k].devIds) == 0 { // Skip pods with no devices |
| 240 | //log.Debugf("%s empty pod", pd[k1].pod.name) |
| 241 | continue |
| 242 | } |
| 243 | if intersect(grp[0].devIds, pods[k].devIds) == true { |
| 244 | //log.Debugf("intersection found %s:%s", pd[k].pod.name, pd[k1].pod.name) |
| 245 | if grp[0].node == pods[k].node { |
| 246 | // This should never happen |
| 247 | log.Errorf("pods %s and %s intersect and are on the same server!! Not pairing", |
| 248 | grp[0].name, pods[k].name) |
| 249 | continue |
| 250 | } |
| 251 | grp = append(grp, pods[k]) |
| 252 | pods = rmPod(pods, k) |
| 253 | break |
| 254 | |
| 255 | } |
| 256 | } |
| 257 | rtrn = append(rtrn, grp) |
| 258 | //log.Debugf("Added group %s", grp[0].name) |
| 259 | // Check if the number of groups = half the pods, if so all groups are started. |
| 260 | if len(rtrn) == podCt >> 1 { |
| 261 | // Append any remaining pods to out |
| 262 | out = append(out,pods[0:]...) |
| 263 | break |
| 264 | } |
| 265 | } |
| 266 | return rtrn,out |
| 267 | } |
| 268 | |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 269 | func unallocPodCount(pd []*podTrack) int { |
| 270 | var rtrn int = 0 |
| 271 | for _,v := range pd { |
| 272 | if v.dn == false { |
| 273 | rtrn++ |
| 274 | } |
| 275 | } |
| 276 | return rtrn |
| 277 | } |
| 278 | |
| 279 | |
| 280 | func sameNode(pod *rwPod, grps [][]*rwPod) bool { |
| 281 | for _,v := range grps { |
| 282 | if v[0].node == pod.node { |
| 283 | return true |
| 284 | } |
| 285 | if len(v) == 2 && v[1].node == pod.node { |
| 286 | return true |
| 287 | } |
| 288 | } |
| 289 | return false |
| 290 | } |
| 291 | |
| 292 | func startRemainingGroups1(grps [][]*rwPod, pods []*rwPod, podCt int) ([][]*rwPod, []*rwPod) { |
| 293 | var grp []*rwPod |
| 294 | |
| 295 | for k,_ := range pods { |
| 296 | if sameNode(pods[k], grps) { |
| 297 | continue |
| 298 | } |
| 299 | grp = []*rwPod{} |
| 300 | grp = append(grp, pods[k]) |
| 301 | pods = rmPod(pods, k) |
| 302 | grps = append(grps, grp) |
| 303 | if len(grps) == podCt >> 1 { |
| 304 | break |
| 305 | } |
| 306 | } |
| 307 | return grps, pods |
| 308 | } |
| 309 | |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 310 | func hasSingleSecondNode(grp []*rwPod) bool { |
| 311 | var srvrs map[string]struct{} = make(map[string]struct{}) |
| 312 | for k,_ := range grp { |
| 313 | if k == 0 { |
| 314 | continue // Ignore the first item |
| 315 | } |
| 316 | srvrs[grp[k].node] = struct{}{} |
| 317 | } |
| 318 | if len(srvrs) == 1 { |
| 319 | return true |
| 320 | } |
| 321 | return false |
| 322 | } |
| 323 | |
| 324 | func addNode(grps [][]*rwPod, idx *rwPod, item *rwPod) [][]*rwPod { |
| 325 | for k,_ := range grps { |
| 326 | if grps[k][0].name == idx.name { |
| 327 | grps[k] = append(grps[k], item) |
| 328 | return grps |
| 329 | } |
| 330 | } |
| 331 | // TODO: Error checking required here. |
| 332 | return grps |
| 333 | } |
| 334 | |
| 335 | func removeNode(grps [][]*rwPod, item *rwPod) [][]*rwPod { |
| 336 | for k,_ := range grps { |
| 337 | for k1,_ := range grps[k] { |
| 338 | if grps[k][k1].name == item.name { |
| 339 | grps[k] = append(grps[k][:k1],grps[k][k1+1:]...) |
| 340 | break |
| 341 | } |
| 342 | } |
| 343 | } |
| 344 | return grps |
| 345 | } |
| 346 | |
| 347 | func groupRemainingPods1(grps [][]*rwPod, pods []*rwPod) [][]*rwPod { |
| 348 | var lgrps [][]*rwPod |
| 349 | // All groups must be started when this function is called. |
| 350 | // Copy incomplete groups |
| 351 | for k,_ := range grps { |
| 352 | if len(grps[k]) != 2 { |
| 353 | lgrps = append(lgrps, grps[k]) |
| 354 | } |
| 355 | } |
| 356 | |
| 357 | // Add all pairing candidates to each started group. |
| 358 | for k,_ := range pods { |
| 359 | for k2,_ := range lgrps { |
| 360 | if lgrps[k2][0].node != pods[k].node { |
| 361 | lgrps[k2] = append(lgrps[k2], pods[k]) |
| 362 | } |
| 363 | } |
| 364 | } |
| 365 | |
| 366 | //TODO: If any member of lgrps doesn't have at least 2 |
| 367 | // nodes something is wrong. Check for that here |
| 368 | |
| 369 | for { |
| 370 | for { // Address groups with only a single server choice |
| 371 | var ssn bool = false |
| 372 | |
| 373 | for k,_ := range lgrps { |
| 374 | // Now if any of the groups only have a single |
| 375 | // node as the choice for the second member |
| 376 | // address that one first. |
| 377 | if hasSingleSecondNode(lgrps[k]) == true { |
| 378 | ssn = true |
| 379 | // Add this pairing to the groups |
| 380 | grps = addNode(grps, lgrps[k][0], lgrps[k][1]) |
| 381 | // Since this node is now used, remove it from all |
| 382 | // remaining tenative groups |
| 383 | lgrps = removeNode(lgrps, lgrps[k][1]) |
| 384 | // Now remove this group completely since |
| 385 | // it's been addressed |
| 386 | lgrps = append(lgrps[:k],lgrps[k+1:]...) |
| 387 | break |
| 388 | } |
| 389 | } |
| 390 | if ssn == false { |
| 391 | break |
| 392 | } |
| 393 | } |
| 394 | // Now adress one of the remaining groups |
| 395 | if len(lgrps) == 0 { |
| 396 | break // Nothing left to do, exit the loop |
| 397 | } |
| 398 | grps = addNode(grps, lgrps[0][0], lgrps[0][1]) |
| 399 | lgrps = removeNode(lgrps, lgrps[0][1]) |
| 400 | lgrps = append(lgrps[:0],lgrps[1:]...) |
| 401 | } |
| 402 | return grps |
| 403 | } |
| 404 | |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 405 | func groupPods1(pods []*rwPod) [][]*rwPod { |
| 406 | var rtrn [][]*rwPod |
| 407 | var podCt int = len(pods) |
| 408 | |
| 409 | rtrn,pods = groupIntersectingPods1(pods, podCt) |
| 410 | // There are several outcomes here |
| 411 | // 1) All pods have been paired and we're done |
| 412 | // 2) Some un-allocated pods remain |
| 413 | // 2.a) All groups have been started |
| 414 | // 2.b) Not all groups have been started |
| 415 | if len(pods) == 0 { |
| 416 | return rtrn |
| 417 | } else if len(rtrn) == podCt >> 1 { // All groupings started |
| 418 | // Allocate the remaining (presumably empty) pods to the started groups |
| 419 | return groupRemainingPods1(rtrn, pods) |
| 420 | } else { // Some groupings started |
| 421 | // Start empty groups with remaining pods |
| 422 | // each grouping is on a different server then |
| 423 | // allocate remaining pods. |
| 424 | rtrn, pods = startRemainingGroups1(rtrn, pods, podCt) |
| 425 | return groupRemainingPods1(rtrn, pods) |
| 426 | } |
| 427 | } |
| 428 | |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 429 | func intersect(d1 map[string]struct{}, d2 map[string]struct{}) bool { |
| 430 | for k,_ := range d1 { |
| 431 | if _,ok := d2[k]; ok == true { |
| 432 | return true |
| 433 | } |
| 434 | } |
| 435 | return false |
| 436 | } |
| 437 | |
| 438 | func setConnection(client pb.ConfigurationClient, backend string, connection string, addr string, port uint64) { |
| 439 | log.Debugf("Configuring backend %s : connection %s\n\n", backend, connection) |
| 440 | cnf := &pb.Conn{Server:"grpc_command",Cluster:"vcore",Backend:backend, |
| 441 | Connection:connection,Addr:addr, |
| 442 | Port:port} |
| 443 | if res, err := client.SetConnection(context.Background(), cnf); err != nil { |
| 444 | log.Debugf("failed SetConnection RPC call: %s", err) |
| 445 | } else { |
| 446 | log.Debugf("Result: %v", res) |
| 447 | } |
| 448 | } |
| 449 | |
| 450 | func setAffinity(client pb.ConfigurationClient, ids map[string]struct{}, backend string) { |
| 451 | log.Debugf("Configuring backend %s : affinities \n", backend) |
| 452 | aff := &pb.Affinity{Router:"vcore",Route:"dev_manager",Cluster:"vcore",Backend:backend} |
| 453 | for k,_ := range ids { |
| 454 | log.Debugf("Setting affinity for id %s", k) |
| 455 | aff.Id = k |
| 456 | if res, err := client.SetAffinity(context.Background(), aff); err != nil { |
| 457 | log.Debugf("failed affinity RPC call: %s", err) |
| 458 | } else { |
| 459 | log.Debugf("Result: %v", res) |
| 460 | } |
| 461 | } |
| 462 | } |
| 463 | |
sslobodr | 38afd0d | 2019-01-21 12:31:46 -0500 | [diff] [blame] | 464 | func getBackendForCore(coreId string, coreGroups [][]*rwPod) string { |
sslobodr | 6c1689c | 2019-01-24 07:31:15 -0500 | [diff] [blame] | 465 | for _,v := range coreGroups { |
| 466 | for _,v2 := range v { |
sslobodr | 38afd0d | 2019-01-21 12:31:46 -0500 | [diff] [blame] | 467 | if v2.name == coreId { |
| 468 | return v2.backend |
| 469 | } |
| 470 | } |
| 471 | } |
| 472 | log.Errorf("No backend found for core %s\n", coreId) |
| 473 | return "" |
| 474 | } |
| 475 | |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 476 | func monitorDiscovery(client pb.ConfigurationClient, |
sslobodr | 38afd0d | 2019-01-21 12:31:46 -0500 | [diff] [blame] | 477 | ch <-chan *ic.InterContainerMessage, |
| 478 | coreGroups [][]*rwPod) { |
| 479 | var id map[string]struct{} = make(map[string]struct{}) |
| 480 | |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 481 | select { |
| 482 | case msg := <-ch: |
| 483 | log.Debugf("Received a device discovery notification") |
sslobodr | 38afd0d | 2019-01-21 12:31:46 -0500 | [diff] [blame] | 484 | device := &ic.DeviceDiscovered{} |
| 485 | if err := ptypes.UnmarshalAny(msg.Body, device); err != nil { |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 486 | log.Errorf("Could not unmarshal received notification %v", msg) |
| 487 | } else { |
sslobodr | 38afd0d | 2019-01-21 12:31:46 -0500 | [diff] [blame] | 488 | // Set the affinity of the discovered device. |
| 489 | if be := getBackendForCore(device.Id, coreGroups); be != "" { |
| 490 | id[device.Id]=struct{}{} |
| 491 | setAffinity(client, id, be) |
| 492 | } else { |
| 493 | log.Error("Cant use an empty string as a backend name") |
| 494 | } |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 495 | } |
| 496 | break |
| 497 | } |
| 498 | } |
| 499 | |
sslobodr | 38afd0d | 2019-01-21 12:31:46 -0500 | [diff] [blame] | 500 | func startDiscoveryMonitor(client pb.ConfigurationClient, |
| 501 | coreGroups [][]*rwPod) error { |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 502 | var ch <-chan *ic.InterContainerMessage |
| 503 | // Connect to kafka for discovery events |
| 504 | topic := &kafka.Topic{Name: "AffinityRouter"} |
| 505 | kc,err := newKafkaClient("sarama", "kafka", 9092, "arouterd") |
| 506 | kc.Start() |
| 507 | |
| 508 | if ch, err = kc.Subscribe(topic); err != nil { |
| 509 | log.Error("Could not subscribe to the 'AffinityRouter' channel, discovery disabled") |
| 510 | return err |
| 511 | } |
sslobodr | 38afd0d | 2019-01-21 12:31:46 -0500 | [diff] [blame] | 512 | go monitorDiscovery(client, ch, coreGroups) |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 513 | return nil |
| 514 | } |
| 515 | |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 516 | // Determines which items in core groups |
| 517 | // have changed based on the list provided |
| 518 | // and returns a coreGroup with only the changed |
| 519 | // items and a pod list with the new items |
| 520 | func getAddrDiffs(coreGroups [][]*rwPod, rwPods []*rwPod) ([][]*rwPod, []*rwPod) { |
| 521 | var nList []*rwPod |
| 522 | var rtrn [][]*rwPod = make([][]*rwPod, nPods>>1) |
| 523 | var ipAddrs map[string]struct{} = make(map[string]struct{}) |
| 524 | |
| 525 | log.Debug("Get addr diffs") |
| 526 | |
| 527 | // Start with an empty array |
sslobodr | 6c1689c | 2019-01-24 07:31:15 -0500 | [diff] [blame] | 528 | for k,_ := range rtrn { |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 529 | rtrn[k] = make([]*rwPod, 2) |
| 530 | } |
| 531 | |
| 532 | // Build a list with only the new items |
sslobodr | 6c1689c | 2019-01-24 07:31:15 -0500 | [diff] [blame] | 533 | for _,v := range rwPods { |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 534 | if hasIpAddr(coreGroups, v.ipAddr) == false { |
| 535 | nList = append(nList, v) |
| 536 | } |
| 537 | ipAddrs[v.ipAddr] = struct{}{} // for the search below |
| 538 | } |
| 539 | |
| 540 | // Now build the coreGroups with only the changed items |
sslobodr | 6c1689c | 2019-01-24 07:31:15 -0500 | [diff] [blame] | 541 | for k1,v1 := range coreGroups { |
| 542 | for k2,v2 := range v1 { |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 543 | if _,ok := ipAddrs[v2.ipAddr]; ok == false { |
| 544 | rtrn[k1][k2] = v2 |
| 545 | } |
| 546 | } |
| 547 | } |
| 548 | return rtrn, nList |
| 549 | } |
| 550 | |
| 551 | // Figure out where best to put the new pods |
| 552 | // in the coreGroup array based on the old |
| 553 | // pods being replaced. The criteria is that |
| 554 | // the new pod be on the same server as the |
| 555 | // old pod was. |
| 556 | func reconcileAddrDiffs(coreGroupDiffs [][]*rwPod, rwPodDiffs []*rwPod) ([][]*rwPod) { |
| 557 | var srvrs map[string][]*rwPod = make(map[string][]*rwPod) |
| 558 | |
| 559 | log.Debug("Reconciling diffs") |
| 560 | log.Debug("Building server list") |
sslobodr | 6c1689c | 2019-01-24 07:31:15 -0500 | [diff] [blame] | 561 | for _,v := range rwPodDiffs { |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 562 | log.Debugf("Adding %v to the server list", *v) |
| 563 | srvrs[v.node] = append(srvrs[v.node], v) |
| 564 | } |
| 565 | |
sslobodr | 6c1689c | 2019-01-24 07:31:15 -0500 | [diff] [blame] | 566 | for k1,v1 := range coreGroupDiffs { |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 567 | log.Debugf("k1:%v, v1:%v", k1,v1) |
sslobodr | 6c1689c | 2019-01-24 07:31:15 -0500 | [diff] [blame] | 568 | for k2,v2 := range v1 { |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 569 | log.Debugf("k2:%v, v2:%v", k2,v2) |
| 570 | if v2 == nil { // Nothing to do here |
| 571 | continue |
| 572 | } |
| 573 | if _,ok := srvrs[v2.node]; ok == true { |
| 574 | coreGroupDiffs[k1][k2] = srvrs[v2.node][0] |
| 575 | if len(srvrs[v2.node]) > 1 { // remove one entry from the list |
| 576 | srvrs[v2.node] = append(srvrs[v2.node][:0], srvrs[v2.node][1:]...) |
| 577 | } else { // Delete the endtry from the map |
| 578 | delete(srvrs, v2.node) |
| 579 | } |
| 580 | } else { |
| 581 | log.Error("This should never happen, node appears to have changed names") |
| 582 | // attempt to limp along by keeping this old entry |
| 583 | } |
| 584 | } |
| 585 | } |
| 586 | |
| 587 | return coreGroupDiffs |
| 588 | } |
| 589 | |
| 590 | func applyAddrDiffs(client pb.ConfigurationClient, coreGroups [][]*rwPod, rwPods []*rwPod) { |
| 591 | var newEntries [][]*rwPod |
| 592 | |
| 593 | log.Debug("Applying diffs") |
| 594 | newEntries = reconcileAddrDiffs(getAddrDiffs(coreGroups, rwPods)) |
| 595 | |
| 596 | // Now replace the information in coreGropus with the new |
| 597 | // entries and then reconcile the device ids on the core |
| 598 | // that's in the new entry with the device ids of it's |
| 599 | // active-active peer. |
sslobodr | 6c1689c | 2019-01-24 07:31:15 -0500 | [diff] [blame] | 600 | for k1,v1 := range coreGroups { |
| 601 | for k2,v2 := range v1 { |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 602 | if newEntries[k1][k2] != nil { |
| 603 | // TODO: Missing is the case where bothe the primary |
| 604 | // and the secondary core crash and come back. |
| 605 | // Pull the device ids from the active-active peer |
| 606 | ids := queryPodDeviceIds(coreGroups[k1][k2^1]) |
| 607 | if len(ids) != 0 { |
| 608 | if reconcilePodDeviceIds(newEntries[k1][k2], ids) == false { |
| 609 | log.Errorf("Attempt to reconcile ids on pod %v failed",newEntries[k1][k2]) |
| 610 | } |
| 611 | } |
| 612 | // Send the affininty router new connection information |
| 613 | setConnection(client, v2.backend, v2.connection, newEntries[k1][k2].ipAddr, 50057) |
| 614 | // Copy the new entry information over |
| 615 | coreGroups[k1][k2].ipAddr = newEntries[k1][k2].ipAddr |
| 616 | coreGroups[k1][k2].name = newEntries[k1][k2].name |
| 617 | coreGroups[k1][k2].devIds = ids |
| 618 | } |
| 619 | } |
| 620 | } |
| 621 | } |
| 622 | |
sslobodr | cd37bc5 | 2019-01-24 11:47:16 -0500 | [diff] [blame] | 623 | func updateDeviceIds(coreGroups [][]*rwPod, rwPods []*rwPod) { |
| 624 | var byName map[string]*rwPod = make(map[string]*rwPod) |
| 625 | |
| 626 | // Convinience |
| 627 | for _,v := range rwPods { |
| 628 | byName[v.name] = v |
| 629 | } |
| 630 | |
| 631 | for k1,v1 := range coreGroups { |
| 632 | for k2,_ := range v1 { |
| 633 | coreGroups[k1][k2].devIds = byName[v1[k2].name].devIds |
| 634 | } |
| 635 | } |
| 636 | } |
| 637 | |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 638 | func startCoreMonitor(client pb.ConfigurationClient, |
| 639 | clientset *kubernetes.Clientset, |
| 640 | coreFltr *regexp.Regexp, |
| 641 | coreGroups [][]*rwPod) error { |
| 642 | // Now that initial allocation has been completed, monitor the pods |
| 643 | // for IP changes |
| 644 | // The main loop needs to do the following: |
| 645 | // 1) Periodically query the pods and filter out |
| 646 | // the vcore ones |
| 647 | // 2) Validate that the pods running are the same |
| 648 | // as the previous check |
| 649 | // 3) Validate that the IP addresses are the same |
| 650 | // as the last check. |
| 651 | // If the pod name(s) ha(s/ve) changed then remove |
| 652 | // the unused pod names and add in the new pod names |
| 653 | // maintaining the cluster/backend information. |
| 654 | // If an IP address has changed (which shouldn't |
| 655 | // happen unless a pod is re-started) it should get |
| 656 | // caught by the pod name change. |
| 657 | for { |
| 658 | time.Sleep(10 * time.Second) // Wait a while |
| 659 | // Get the rw core list from k8s |
| 660 | rwPods := getRwPods(clientset, coreFltr) |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 661 | queryDeviceIds(rwPods) |
sslobodr | cd37bc5 | 2019-01-24 11:47:16 -0500 | [diff] [blame] | 662 | updateDeviceIds(coreGroups, rwPods) |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 663 | // If we didn't get 2n+1 pods then wait since |
| 664 | // something is down and will hopefully come |
| 665 | // back up at some point. |
| 666 | // TODO: remove the 6 pod hardcoding |
| 667 | if len(rwPods) != 6 { |
| 668 | continue |
| 669 | } |
| 670 | // We have all pods, check if any IP addresses |
| 671 | // have changed. |
| 672 | for _,v := range rwPods { |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 673 | if hasIpAddr(coreGroups, v.ipAddr) == false { |
| 674 | log.Debug("Address has changed...") |
| 675 | applyAddrDiffs(client, coreGroups, rwPods) |
| 676 | |
| 677 | } |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 678 | } |
| 679 | } |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 680 | } |
| 681 | |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 682 | func hasIpAddr(coreGroups [][]*rwPod, ipAddr string) bool { |
sslobodr | 6c1689c | 2019-01-24 07:31:15 -0500 | [diff] [blame] | 683 | for _,v1 := range coreGroups { |
| 684 | for _,v2 := range v1 { |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 685 | if v2.ipAddr == ipAddr { |
| 686 | return true |
| 687 | } |
| 688 | } |
| 689 | } |
| 690 | return false |
| 691 | } |
| 692 | |
| 693 | |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 694 | func main() { |
| 695 | // This is currently hard coded to a cluster with 3 servers |
| 696 | //var connections map[string]configConn = make(map[string]configConn) |
| 697 | //var rwCorePodsPrev map[string]rwPod = make(map[string]rwPod) |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 698 | var err error |
| 699 | var conn *grpc.ClientConn |
| 700 | |
| 701 | |
| 702 | // Set up the regular expression to identify the voltha cores |
| 703 | coreFltr := regexp.MustCompile(`rw-core[0-9]-`) |
| 704 | |
| 705 | // Set up logging |
| 706 | if _, err := log.SetDefaultLogger(log.JSON, 0, nil); err != nil { |
| 707 | log.With(log.Fields{"error": err}).Fatal("Cannot setup logging") |
| 708 | } |
| 709 | |
| 710 | // Set up kubernetes api |
| 711 | clientset := k8sClientSet() |
| 712 | |
| 713 | // Connect to the affinity router and set up the client |
| 714 | conn, err = connect("localhost:55554") // This is a sidecar container so communicating over localhost |
sslobodr | cd37bc5 | 2019-01-24 11:47:16 -0500 | [diff] [blame] | 715 | defer conn.Close() |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 716 | if err != nil { |
| 717 | panic(err.Error()) |
| 718 | } |
| 719 | client := pb.NewConfigurationClient(conn) |
| 720 | |
| 721 | // Get the voltha rw-core podes |
| 722 | rwPods := getRwPods(clientset, coreFltr) |
| 723 | |
| 724 | // Fetch the devices held by each running core |
sslobodr | e7ce71d | 2019-01-22 16:21:45 -0500 | [diff] [blame] | 725 | queryDeviceIds(rwPods) |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 726 | |
| 727 | // For debugging... comment out l8r |
| 728 | for _,v := range rwPods { |
| 729 | log.Debugf("Pod list %v", *v) |
| 730 | } |
| 731 | |
| 732 | coreGroups := groupPods1(rwPods) |
| 733 | |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 734 | // Assign the groupings to the the backends and connections |
| 735 | for k,_ := range coreGroups { |
| 736 | for k1,_ := range coreGroups[k] { |
| 737 | coreGroups[k][k1].backend = "vcore"+strconv.Itoa(k+1) |
| 738 | coreGroups[k][k1].connection = "vcore"+strconv.Itoa(k+1)+strconv.Itoa(k1+1) |
| 739 | } |
| 740 | } |
sslobodr | cd37bc5 | 2019-01-24 11:47:16 -0500 | [diff] [blame] | 741 | log.Info("Core gouping completed") |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 742 | |
| 743 | // TODO: Debugging code, comment out for production |
| 744 | for k,v := range coreGroups { |
| 745 | for k2,v2 := range v { |
| 746 | log.Debugf("Core group %d,%d: %v", k, k2, v2) |
| 747 | } |
| 748 | } |
sslobodr | cd37bc5 | 2019-01-24 11:47:16 -0500 | [diff] [blame] | 749 | log.Info("Setting affinities") |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 750 | // Now set the affinities for exising devices in the cores |
| 751 | for _,v := range coreGroups { |
| 752 | setAffinity(client, v[0].devIds, v[0].backend) |
| 753 | setAffinity(client, v[1].devIds, v[1].backend) |
| 754 | } |
sslobodr | cd37bc5 | 2019-01-24 11:47:16 -0500 | [diff] [blame] | 755 | log.Info("Setting connections") |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 756 | // Configure the backeds based on the calculated core groups |
| 757 | for _,v := range coreGroups { |
| 758 | setConnection(client, v[0].backend, v[0].connection, v[0].ipAddr, 50057) |
| 759 | setConnection(client, v[1].backend, v[1].connection, v[1].ipAddr, 50057) |
| 760 | } |
| 761 | |
sslobodr | cd37bc5 | 2019-01-24 11:47:16 -0500 | [diff] [blame] | 762 | log.Info("Starting discovery monitoring") |
sslobodr | 38afd0d | 2019-01-21 12:31:46 -0500 | [diff] [blame] | 763 | startDiscoveryMonitor(client, coreGroups) |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 764 | |
sslobodr | cd37bc5 | 2019-01-24 11:47:16 -0500 | [diff] [blame] | 765 | log.Info("Starting core monitoring") |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 766 | startCoreMonitor(client, clientset, coreFltr, coreGroups) // Never returns |
| 767 | return |
sslobodr | 16e41bc | 2019-01-18 16:22:21 -0500 | [diff] [blame] | 768 | } |
| 769 | |