Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 1 | /* |
Joey Armstrong | 5f51f2e | 2023-01-17 17:06:26 -0500 | [diff] [blame^] | 2 | * Copyright 2018-2023 Open Networking Foundation (ONF) and the ONF Contributors |
Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 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 core |
| 18 | |
| 19 | import ( |
| 20 | "context" |
| 21 | "errors" |
| 22 | "time" |
| 23 | |
khenaidoo | d948f77 | 2021-08-11 17:49:24 -0400 | [diff] [blame] | 24 | "github.com/opencord/voltha-lib-go/v7/pkg/db" |
| 25 | "github.com/opencord/voltha-lib-go/v7/pkg/db/kvstore" |
| 26 | "github.com/opencord/voltha-lib-go/v7/pkg/log" |
| 27 | "github.com/opencord/voltha-lib-go/v7/pkg/probe" |
Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 28 | "google.golang.org/grpc/codes" |
| 29 | "google.golang.org/grpc/status" |
| 30 | ) |
| 31 | |
Rohan Agrawal | 31f2180 | 2020-06-12 05:38:46 +0000 | [diff] [blame] | 32 | func newKVClient(ctx context.Context, storeType string, address string, timeout time.Duration) (kvstore.Client, error) { |
| 33 | logger.Infow(ctx, "kv-store-type", log.Fields{"store": storeType}) |
Andrey Pozolotin | 34dd63f | 2021-05-31 21:26:40 +0300 | [diff] [blame] | 34 | if storeType == "etcd" { |
Rohan Agrawal | 31f2180 | 2020-06-12 05:38:46 +0000 | [diff] [blame] | 35 | return kvstore.NewEtcdClient(ctx, address, timeout, log.FatalLevel) |
Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 36 | } |
| 37 | return nil, errors.New("unsupported-kv-store") |
| 38 | } |
| 39 | |
| 40 | func stopKVClient(ctx context.Context, kvClient kvstore.Client) { |
| 41 | // Release all reservations |
| 42 | if err := kvClient.ReleaseAllReservations(ctx); err != nil { |
Rohan Agrawal | 31f2180 | 2020-06-12 05:38:46 +0000 | [diff] [blame] | 43 | logger.Infow(ctx, "fail-to-release-all-reservations", log.Fields{"error": err}) |
Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 44 | } |
| 45 | // Close the DB connection |
Rohan Agrawal | 31f2180 | 2020-06-12 05:38:46 +0000 | [diff] [blame] | 46 | kvClient.Close(ctx) |
Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 47 | } |
| 48 | |
| 49 | // waitUntilKVStoreReachableOrMaxTries will wait until it can connect to a KV store or until maxtries has been reached |
khenaidoo | d948f77 | 2021-08-11 17:49:24 -0400 | [diff] [blame] | 50 | func waitUntilKVStoreReachableOrMaxTries(ctx context.Context, kvClient kvstore.Client, maxRetries int, retryInterval time.Duration, serviceName string) error { |
Rohan Agrawal | 31f2180 | 2020-06-12 05:38:46 +0000 | [diff] [blame] | 51 | logger.Infow(ctx, "verifying-KV-store-connectivity", log.Fields{"retries": maxRetries, "retryInterval": retryInterval}) |
Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 52 | count := 0 |
| 53 | for { |
| 54 | if !kvClient.IsConnectionUp(ctx) { |
Rohan Agrawal | 31f2180 | 2020-06-12 05:38:46 +0000 | [diff] [blame] | 55 | logger.Info(ctx, "KV-store-unreachable") |
Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 56 | if maxRetries != -1 { |
| 57 | if count >= maxRetries { |
| 58 | return status.Error(codes.Unavailable, "kv store unreachable") |
| 59 | } |
| 60 | } |
| 61 | count++ |
| 62 | |
| 63 | // Take a nap before retrying |
| 64 | select { |
| 65 | case <-ctx.Done(): |
| 66 | //ctx canceled |
| 67 | return ctx.Err() |
| 68 | case <-time.After(retryInterval): |
| 69 | } |
Rohan Agrawal | 31f2180 | 2020-06-12 05:38:46 +0000 | [diff] [blame] | 70 | logger.Infow(ctx, "retry-KV-store-connectivity", log.Fields{"retryCount": count, "maxRetries": maxRetries, "retryInterval": retryInterval}) |
Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 71 | } else { |
| 72 | break |
| 73 | } |
| 74 | } |
khenaidoo | d948f77 | 2021-08-11 17:49:24 -0400 | [diff] [blame] | 75 | probe.UpdateStatusFromContext(ctx, serviceName, probe.ServiceStatusRunning) |
Rohan Agrawal | 31f2180 | 2020-06-12 05:38:46 +0000 | [diff] [blame] | 76 | logger.Info(ctx, "KV-store-reachable") |
Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 77 | return nil |
| 78 | } |
| 79 | |
| 80 | /* |
| 81 | * Thread to monitor kvstore Liveness (connection status) |
| 82 | * |
| 83 | * This function constantly monitors Liveness State of kvstore as reported |
| 84 | * periodically by backend and updates the Status of kv-store service registered |
| 85 | * with rw_core probe. |
| 86 | * |
| 87 | * If no liveness event has been seen within a timeout, then the thread will |
| 88 | * perform a "liveness" check attempt, which will in turn trigger a liveness event on |
| 89 | * the liveness channel, true or false depending on whether the attempt succeeded. |
| 90 | * |
| 91 | * The gRPC server in turn monitors the state of the readiness probe and will |
| 92 | * start issuing UNAVAILABLE response while the probe is not ready. |
| 93 | */ |
khenaidoo | d948f77 | 2021-08-11 17:49:24 -0400 | [diff] [blame] | 94 | func monitorKVStoreLiveness(ctx context.Context, backend *db.Backend, serviceName string, liveProbeInterval, notLiveProbeInterval time.Duration) { |
Rohan Agrawal | 31f2180 | 2020-06-12 05:38:46 +0000 | [diff] [blame] | 95 | logger.Info(ctx, "start-monitoring-kvstore-liveness") |
Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 96 | |
| 97 | // Instruct backend to create Liveness channel for transporting state updates |
Rohan Agrawal | 31f2180 | 2020-06-12 05:38:46 +0000 | [diff] [blame] | 98 | livenessChannel := backend.EnableLivenessChannel(ctx) |
Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 99 | |
khenaidoo | d948f77 | 2021-08-11 17:49:24 -0400 | [diff] [blame] | 100 | logger.Infow(ctx, "enabled-liveness-channel", log.Fields{"service-name": serviceName}) |
Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 101 | |
| 102 | // Default state for kvstore is alive for rw_core |
| 103 | timeout := liveProbeInterval |
| 104 | loop: |
| 105 | for { |
| 106 | timeoutTimer := time.NewTimer(timeout) |
| 107 | select { |
| 108 | |
| 109 | case liveness := <-livenessChannel: |
khenaidoo | d948f77 | 2021-08-11 17:49:24 -0400 | [diff] [blame] | 110 | logger.Debugw(ctx, "received-liveness-change-notification", log.Fields{"liveness": liveness, "service-name": serviceName}) |
Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 111 | |
| 112 | if !liveness { |
khenaidoo | d948f77 | 2021-08-11 17:49:24 -0400 | [diff] [blame] | 113 | probe.UpdateStatusFromContext(ctx, serviceName, probe.ServiceStatusNotReady) |
| 114 | logger.Infow(ctx, "service-not-ready", log.Fields{"service-name": serviceName}) |
Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 115 | |
| 116 | timeout = notLiveProbeInterval |
| 117 | |
| 118 | } else { |
khenaidoo | d948f77 | 2021-08-11 17:49:24 -0400 | [diff] [blame] | 119 | probe.UpdateStatusFromContext(ctx, serviceName, probe.ServiceStatusRunning) |
| 120 | logger.Infow(ctx, "service-ready", log.Fields{"service-name": serviceName}) |
Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 121 | |
| 122 | timeout = liveProbeInterval |
| 123 | } |
| 124 | |
| 125 | if !timeoutTimer.Stop() { |
| 126 | <-timeoutTimer.C |
| 127 | } |
| 128 | |
| 129 | case <-ctx.Done(): |
| 130 | break loop |
| 131 | |
| 132 | case <-timeoutTimer.C: |
khenaidoo | d948f77 | 2021-08-11 17:49:24 -0400 | [diff] [blame] | 133 | logger.Infow(ctx, "perform-liveness-check-on-timeout", log.Fields{"service-name": serviceName}) |
Kent Hagerman | 2f0d055 | 2020-04-23 17:28:52 -0400 | [diff] [blame] | 134 | |
| 135 | // Trigger Liveness check if no liveness update received within the timeout period. |
| 136 | // The Liveness check will push Live state to same channel which this routine is |
| 137 | // reading and processing. This, do it asynchronously to avoid blocking for |
| 138 | // backend response and avoid any possibility of deadlock |
| 139 | go backend.PerformLivenessCheck(ctx) |
| 140 | } |
| 141 | } |
| 142 | } |