| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 1 | /* |
| 2 | * |
| 3 | * Copyright 2017 gRPC authors. |
| 4 | * |
| 5 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 6 | * you may not use this file except in compliance with the License. |
| 7 | * You may obtain a copy of the License at |
| 8 | * |
| 9 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 10 | * |
| 11 | * Unless required by applicable law or agreed to in writing, software |
| 12 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 13 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 14 | * See the License for the specific language governing permissions and |
| 15 | * limitations under the License. |
| 16 | * |
| 17 | */ |
| 18 | |
| 19 | package base |
| 20 | |
| 21 | import ( |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 22 | "errors" |
| 23 | "fmt" |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 24 | |
| 25 | "google.golang.org/grpc/balancer" |
| 26 | "google.golang.org/grpc/connectivity" |
| 27 | "google.golang.org/grpc/grpclog" |
| 28 | "google.golang.org/grpc/resolver" |
| 29 | ) |
| 30 | |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 31 | var logger = grpclog.Component("balancer") |
| 32 | |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 33 | type baseBuilder struct { |
| 34 | name string |
| 35 | pickerBuilder PickerBuilder |
| 36 | config Config |
| 37 | } |
| 38 | |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 39 | func (bb *baseBuilder) Build(cc balancer.ClientConn, _ balancer.BuildOptions) balancer.Balancer { |
| 40 | bal := &baseBalancer{ |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 41 | cc: cc, |
| 42 | pickerBuilder: bb.pickerBuilder, |
| 43 | |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 44 | subConns: resolver.NewAddressMapV2[balancer.SubConn](), |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 45 | scStates: make(map[balancer.SubConn]connectivity.State), |
| Stephane Barbarie | 260a563 | 2019-02-26 16:12:49 -0500 | [diff] [blame] | 46 | csEvltr: &balancer.ConnectivityStateEvaluator{}, |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 47 | config: bb.config, |
| 48 | state: connectivity.Connecting, |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 49 | } |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 50 | // Initialize picker to a picker that always returns |
| 51 | // ErrNoSubConnAvailable, because when state of a SubConn changes, we |
| 52 | // may call UpdateState with this picker. |
| 53 | bal.picker = NewErrPicker(balancer.ErrNoSubConnAvailable) |
| 54 | return bal |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 55 | } |
| 56 | |
| 57 | func (bb *baseBuilder) Name() string { |
| 58 | return bb.name |
| 59 | } |
| 60 | |
| 61 | type baseBalancer struct { |
| 62 | cc balancer.ClientConn |
| 63 | pickerBuilder PickerBuilder |
| 64 | |
| Stephane Barbarie | 260a563 | 2019-02-26 16:12:49 -0500 | [diff] [blame] | 65 | csEvltr *balancer.ConnectivityStateEvaluator |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 66 | state connectivity.State |
| 67 | |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 68 | subConns *resolver.AddressMapV2[balancer.SubConn] |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 69 | scStates map[balancer.SubConn]connectivity.State |
| 70 | picker balancer.Picker |
| 71 | config Config |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 72 | |
| 73 | resolverErr error // the last error reported by the resolver; cleared on successful resolution |
| 74 | connErr error // the last connection error; cleared upon leaving TransientFailure |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 75 | } |
| 76 | |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 77 | func (b *baseBalancer) ResolverError(err error) { |
| 78 | b.resolverErr = err |
| 79 | if b.subConns.Len() == 0 { |
| 80 | b.state = connectivity.TransientFailure |
| 81 | } |
| Scott Baker | 8461e15 | 2019-10-01 14:44:30 -0700 | [diff] [blame] | 82 | |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 83 | if b.state != connectivity.TransientFailure { |
| 84 | // The picker will not change since the balancer does not currently |
| 85 | // report an error. |
| 86 | return |
| 87 | } |
| 88 | b.regeneratePicker() |
| 89 | b.cc.UpdateState(balancer.State{ |
| 90 | ConnectivityState: b.state, |
| 91 | Picker: b.picker, |
| 92 | }) |
| Andrea Campanella | 3614a92 | 2021-02-25 12:40:42 +0100 | [diff] [blame] | 93 | } |
| 94 | |
| 95 | func (b *baseBalancer) UpdateClientConnState(s balancer.ClientConnState) error { |
| Scott Baker | 8461e15 | 2019-10-01 14:44:30 -0700 | [diff] [blame] | 96 | // TODO: handle s.ResolverState.ServiceConfig? |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 97 | if logger.V(2) { |
| 98 | logger.Info("base.baseBalancer: got new ClientConn state: ", s) |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 99 | } |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 100 | // Successful resolution; clear resolver error and ensure we return nil. |
| 101 | b.resolverErr = nil |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 102 | // addrsSet is the set converted from addrs, it's used for quick lookup of an address. |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 103 | addrsSet := resolver.NewAddressMapV2[any]() |
| Scott Baker | 8461e15 | 2019-10-01 14:44:30 -0700 | [diff] [blame] | 104 | for _, a := range s.ResolverState.Addresses { |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 105 | addrsSet.Set(a, nil) |
| 106 | if _, ok := b.subConns.Get(a); !ok { |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 107 | // a is a new address (not existing in b.subConns). |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 108 | var sc balancer.SubConn |
| 109 | opts := balancer.NewSubConnOptions{ |
| 110 | HealthCheckEnabled: b.config.HealthCheck, |
| 111 | StateListener: func(scs balancer.SubConnState) { b.updateSubConnState(sc, scs) }, |
| 112 | } |
| 113 | sc, err := b.cc.NewSubConn([]resolver.Address{a}, opts) |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 114 | if err != nil { |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 115 | logger.Warningf("base.baseBalancer: failed to create new SubConn: %v", err) |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 116 | continue |
| 117 | } |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 118 | b.subConns.Set(a, sc) |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 119 | b.scStates[sc] = connectivity.Idle |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 120 | b.csEvltr.RecordTransition(connectivity.Shutdown, connectivity.Idle) |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 121 | sc.Connect() |
| 122 | } |
| 123 | } |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 124 | for _, a := range b.subConns.Keys() { |
| 125 | sc, _ := b.subConns.Get(a) |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 126 | // a was removed by resolver. |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 127 | if _, ok := addrsSet.Get(a); !ok { |
| 128 | sc.Shutdown() |
| 129 | b.subConns.Delete(a) |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 130 | // Keep the state of this sc in b.scStates until sc's state becomes Shutdown. |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 131 | // The entry will be deleted in updateSubConnState. |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 132 | } |
| 133 | } |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 134 | // If resolver state contains no addresses, return an error so ClientConn |
| 135 | // will trigger re-resolve. Also records this as a resolver error, so when |
| 136 | // the overall state turns transient failure, the error message will have |
| 137 | // the zero address information. |
| 138 | if len(s.ResolverState.Addresses) == 0 { |
| 139 | b.ResolverError(errors.New("produced zero addresses")) |
| 140 | return balancer.ErrBadResolverState |
| 141 | } |
| 142 | |
| 143 | b.regeneratePicker() |
| 144 | b.cc.UpdateState(balancer.State{ConnectivityState: b.state, Picker: b.picker}) |
| Andrea Campanella | 3614a92 | 2021-02-25 12:40:42 +0100 | [diff] [blame] | 145 | return nil |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 146 | } |
| 147 | |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 148 | // mergeErrors builds an error from the last connection error and the last |
| 149 | // resolver error. Must only be called if b.state is TransientFailure. |
| 150 | func (b *baseBalancer) mergeErrors() error { |
| 151 | // connErr must always be non-nil unless there are no SubConns, in which |
| 152 | // case resolverErr must be non-nil. |
| 153 | if b.connErr == nil { |
| 154 | return fmt.Errorf("last resolver error: %v", b.resolverErr) |
| 155 | } |
| 156 | if b.resolverErr == nil { |
| 157 | return fmt.Errorf("last connection error: %v", b.connErr) |
| 158 | } |
| 159 | return fmt.Errorf("last connection error: %v; last resolver error: %v", b.connErr, b.resolverErr) |
| 160 | } |
| 161 | |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 162 | // regeneratePicker takes a snapshot of the balancer, and generates a picker |
| 163 | // from it. The picker is |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 164 | // - errPicker if the balancer is in TransientFailure, |
| 165 | // - built by the pickerBuilder with all READY SubConns otherwise. |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 166 | func (b *baseBalancer) regeneratePicker() { |
| 167 | if b.state == connectivity.TransientFailure { |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 168 | b.picker = NewErrPicker(b.mergeErrors()) |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 169 | return |
| 170 | } |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 171 | readySCs := make(map[balancer.SubConn]SubConnInfo) |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 172 | |
| 173 | // Filter out all ready SCs from full subConn map. |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 174 | for _, addr := range b.subConns.Keys() { |
| 175 | sc, _ := b.subConns.Get(addr) |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 176 | if st, ok := b.scStates[sc]; ok && st == connectivity.Ready { |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 177 | readySCs[sc] = SubConnInfo{Address: addr} |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 178 | } |
| 179 | } |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 180 | b.picker = b.pickerBuilder.Build(PickerBuildInfo{ReadySCs: readySCs}) |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 181 | } |
| 182 | |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 183 | // UpdateSubConnState is a nop because a StateListener is always set in NewSubConn. |
| Scott Baker | 8461e15 | 2019-10-01 14:44:30 -0700 | [diff] [blame] | 184 | func (b *baseBalancer) UpdateSubConnState(sc balancer.SubConn, state balancer.SubConnState) { |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 185 | logger.Errorf("base.baseBalancer: UpdateSubConnState(%v, %+v) called unexpectedly", sc, state) |
| 186 | } |
| 187 | |
| 188 | func (b *baseBalancer) updateSubConnState(sc balancer.SubConn, state balancer.SubConnState) { |
| Scott Baker | 8461e15 | 2019-10-01 14:44:30 -0700 | [diff] [blame] | 189 | s := state.ConnectivityState |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 190 | if logger.V(2) { |
| 191 | logger.Infof("base.baseBalancer: handle SubConn state change: %p, %v", sc, s) |
| Scott Baker | 8461e15 | 2019-10-01 14:44:30 -0700 | [diff] [blame] | 192 | } |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 193 | oldS, ok := b.scStates[sc] |
| 194 | if !ok { |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 195 | if logger.V(2) { |
| 196 | logger.Infof("base.baseBalancer: got state changes for an unknown SubConn: %p, %v", sc, s) |
| 197 | } |
| 198 | return |
| 199 | } |
| 200 | if oldS == connectivity.TransientFailure && |
| 201 | (s == connectivity.Connecting || s == connectivity.Idle) { |
| 202 | // Once a subconn enters TRANSIENT_FAILURE, ignore subsequent IDLE or |
| 203 | // CONNECTING transitions to prevent the aggregated state from being |
| 204 | // always CONNECTING when many backends exist but are all down. |
| 205 | if s == connectivity.Idle { |
| 206 | sc.Connect() |
| Scott Baker | 8461e15 | 2019-10-01 14:44:30 -0700 | [diff] [blame] | 207 | } |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 208 | return |
| 209 | } |
| 210 | b.scStates[sc] = s |
| 211 | switch s { |
| 212 | case connectivity.Idle: |
| 213 | sc.Connect() |
| 214 | case connectivity.Shutdown: |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 215 | // When an address was removed by resolver, b called Shutdown but kept |
| 216 | // the sc's state in scStates. Remove state for this sc here. |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 217 | delete(b.scStates, sc) |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 218 | case connectivity.TransientFailure: |
| 219 | // Save error to be reported via picker. |
| 220 | b.connErr = state.ConnectionError |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 221 | } |
| 222 | |
| Stephane Barbarie | 260a563 | 2019-02-26 16:12:49 -0500 | [diff] [blame] | 223 | b.state = b.csEvltr.RecordTransition(oldS, s) |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 224 | |
| 225 | // Regenerate picker when one of the following happens: |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 226 | // - this sc entered or left ready |
| 227 | // - the aggregated state of balancer is TransientFailure |
| 228 | // (may need to update error message) |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 229 | if (s == connectivity.Ready) != (oldS == connectivity.Ready) || |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 230 | b.state == connectivity.TransientFailure { |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 231 | b.regeneratePicker() |
| 232 | } |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 233 | b.cc.UpdateState(balancer.State{ConnectivityState: b.state, Picker: b.picker}) |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 234 | } |
| 235 | |
| 236 | // Close is a nop because base balancer doesn't have internal state to clean up, |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 237 | // and it doesn't need to call Shutdown for the SubConns. |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 238 | func (b *baseBalancer) Close() { |
| 239 | } |
| 240 | |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 241 | // ExitIdle is a nop because the base balancer attempts to stay connected to |
| 242 | // all SubConns at all times. |
| 243 | func (b *baseBalancer) ExitIdle() { |
| 244 | } |
| 245 | |
| 246 | // NewErrPicker returns a Picker that always returns err on Pick(). |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 247 | func NewErrPicker(err error) balancer.Picker { |
| 248 | return &errPicker{err: err} |
| 249 | } |
| 250 | |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 251 | // NewErrPickerV2 is temporarily defined for backward compatibility reasons. |
| 252 | // |
| 253 | // Deprecated: use NewErrPicker instead. |
| 254 | var NewErrPickerV2 = NewErrPicker |
| 255 | |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 256 | type errPicker struct { |
| 257 | err error // Pick() always returns this err. |
| 258 | } |
| 259 | |
| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 260 | func (p *errPicker) Pick(balancer.PickInfo) (balancer.PickResult, error) { |
| 261 | return balancer.PickResult{}, p.err |
| khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 262 | } |