Scott Baker | 611f6bd | 2019-10-18 13:45:19 -0700 | [diff] [blame^] | 1 | // Copyright 2019 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 tracker |
| 16 | |
| 17 | import ( |
| 18 | "fmt" |
| 19 | "sort" |
| 20 | "strings" |
| 21 | ) |
| 22 | |
| 23 | // Progress represents a follower’s progress in the view of the leader. Leader |
| 24 | // maintains progresses of all followers, and sends entries to the follower |
| 25 | // based on its progress. |
| 26 | // |
| 27 | // NB(tbg): Progress is basically a state machine whose transitions are mostly |
| 28 | // strewn around `*raft.raft`. Additionally, some fields are only used when in a |
| 29 | // certain State. All of this isn't ideal. |
| 30 | type Progress struct { |
| 31 | Match, Next uint64 |
| 32 | // State defines how the leader should interact with the follower. |
| 33 | // |
| 34 | // When in StateProbe, leader sends at most one replication message |
| 35 | // per heartbeat interval. It also probes actual progress of the follower. |
| 36 | // |
| 37 | // When in StateReplicate, leader optimistically increases next |
| 38 | // to the latest entry sent after sending replication message. This is |
| 39 | // an optimized state for fast replicating log entries to the follower. |
| 40 | // |
| 41 | // When in StateSnapshot, leader should have sent out snapshot |
| 42 | // before and stops sending any replication message. |
| 43 | State StateType |
| 44 | |
| 45 | // PendingSnapshot is used in StateSnapshot. |
| 46 | // If there is a pending snapshot, the pendingSnapshot will be set to the |
| 47 | // index of the snapshot. If pendingSnapshot is set, the replication process of |
| 48 | // this Progress will be paused. raft will not resend snapshot until the pending one |
| 49 | // is reported to be failed. |
| 50 | PendingSnapshot uint64 |
| 51 | |
| 52 | // RecentActive is true if the progress is recently active. Receiving any messages |
| 53 | // from the corresponding follower indicates the progress is active. |
| 54 | // RecentActive can be reset to false after an election timeout. |
| 55 | // |
| 56 | // TODO(tbg): the leader should always have this set to true. |
| 57 | RecentActive bool |
| 58 | |
| 59 | // ProbeSent is used while this follower is in StateProbe. When ProbeSent is |
| 60 | // true, raft should pause sending replication message to this peer until |
| 61 | // ProbeSent is reset. See ProbeAcked() and IsPaused(). |
| 62 | ProbeSent bool |
| 63 | |
| 64 | // Inflights is a sliding window for the inflight messages. |
| 65 | // Each inflight message contains one or more log entries. |
| 66 | // The max number of entries per message is defined in raft config as MaxSizePerMsg. |
| 67 | // Thus inflight effectively limits both the number of inflight messages |
| 68 | // and the bandwidth each Progress can use. |
| 69 | // When inflights is Full, no more message should be sent. |
| 70 | // When a leader sends out a message, the index of the last |
| 71 | // entry should be added to inflights. The index MUST be added |
| 72 | // into inflights in order. |
| 73 | // When a leader receives a reply, the previous inflights should |
| 74 | // be freed by calling inflights.FreeLE with the index of the last |
| 75 | // received entry. |
| 76 | Inflights *Inflights |
| 77 | |
| 78 | // IsLearner is true if this progress is tracked for a learner. |
| 79 | IsLearner bool |
| 80 | } |
| 81 | |
| 82 | // ResetState moves the Progress into the specified State, resetting ProbeSent, |
| 83 | // PendingSnapshot, and Inflights. |
| 84 | func (pr *Progress) ResetState(state StateType) { |
| 85 | pr.ProbeSent = false |
| 86 | pr.PendingSnapshot = 0 |
| 87 | pr.State = state |
| 88 | pr.Inflights.reset() |
| 89 | } |
| 90 | |
| 91 | func max(a, b uint64) uint64 { |
| 92 | if a > b { |
| 93 | return a |
| 94 | } |
| 95 | return b |
| 96 | } |
| 97 | |
| 98 | func min(a, b uint64) uint64 { |
| 99 | if a > b { |
| 100 | return b |
| 101 | } |
| 102 | return a |
| 103 | } |
| 104 | |
| 105 | // ProbeAcked is called when this peer has accepted an append. It resets |
| 106 | // ProbeSent to signal that additional append messages should be sent without |
| 107 | // further delay. |
| 108 | func (pr *Progress) ProbeAcked() { |
| 109 | pr.ProbeSent = false |
| 110 | } |
| 111 | |
| 112 | // BecomeProbe transitions into StateProbe. Next is reset to Match+1 or, |
| 113 | // optionally and if larger, the index of the pending snapshot. |
| 114 | func (pr *Progress) BecomeProbe() { |
| 115 | // If the original state is StateSnapshot, progress knows that |
| 116 | // the pending snapshot has been sent to this peer successfully, then |
| 117 | // probes from pendingSnapshot + 1. |
| 118 | if pr.State == StateSnapshot { |
| 119 | pendingSnapshot := pr.PendingSnapshot |
| 120 | pr.ResetState(StateProbe) |
| 121 | pr.Next = max(pr.Match+1, pendingSnapshot+1) |
| 122 | } else { |
| 123 | pr.ResetState(StateProbe) |
| 124 | pr.Next = pr.Match + 1 |
| 125 | } |
| 126 | } |
| 127 | |
| 128 | // BecomeReplicate transitions into StateReplicate, resetting Next to Match+1. |
| 129 | func (pr *Progress) BecomeReplicate() { |
| 130 | pr.ResetState(StateReplicate) |
| 131 | pr.Next = pr.Match + 1 |
| 132 | } |
| 133 | |
| 134 | // BecomeSnapshot moves the Progress to StateSnapshot with the specified pending |
| 135 | // snapshot index. |
| 136 | func (pr *Progress) BecomeSnapshot(snapshoti uint64) { |
| 137 | pr.ResetState(StateSnapshot) |
| 138 | pr.PendingSnapshot = snapshoti |
| 139 | } |
| 140 | |
| 141 | // MaybeUpdate is called when an MsgAppResp arrives from the follower, with the |
| 142 | // index acked by it. The method returns false if the given n index comes from |
| 143 | // an outdated message. Otherwise it updates the progress and returns true. |
| 144 | func (pr *Progress) MaybeUpdate(n uint64) bool { |
| 145 | var updated bool |
| 146 | if pr.Match < n { |
| 147 | pr.Match = n |
| 148 | updated = true |
| 149 | pr.ProbeAcked() |
| 150 | } |
| 151 | if pr.Next < n+1 { |
| 152 | pr.Next = n + 1 |
| 153 | } |
| 154 | return updated |
| 155 | } |
| 156 | |
| 157 | // OptimisticUpdate signals that appends all the way up to and including index n |
| 158 | // are in-flight. As a result, Next is increased to n+1. |
| 159 | func (pr *Progress) OptimisticUpdate(n uint64) { pr.Next = n + 1 } |
| 160 | |
| 161 | // MaybeDecrTo adjusts the Progress to the receipt of a MsgApp rejection. The |
| 162 | // arguments are the index the follower rejected to append to its log, and its |
| 163 | // last index. |
| 164 | // |
| 165 | // Rejections can happen spuriously as messages are sent out of order or |
| 166 | // duplicated. In such cases, the rejection pertains to an index that the |
| 167 | // Progress already knows were previously acknowledged, and false is returned |
| 168 | // without changing the Progress. |
| 169 | // |
| 170 | // If the rejection is genuine, Next is lowered sensibly, and the Progress is |
| 171 | // cleared for sending log entries. |
| 172 | func (pr *Progress) MaybeDecrTo(rejected, last uint64) bool { |
| 173 | if pr.State == StateReplicate { |
| 174 | // The rejection must be stale if the progress has matched and "rejected" |
| 175 | // is smaller than "match". |
| 176 | if rejected <= pr.Match { |
| 177 | return false |
| 178 | } |
| 179 | // Directly decrease next to match + 1. |
| 180 | // |
| 181 | // TODO(tbg): why not use last if it's larger? |
| 182 | pr.Next = pr.Match + 1 |
| 183 | return true |
| 184 | } |
| 185 | |
| 186 | // The rejection must be stale if "rejected" does not match next - 1. This |
| 187 | // is because non-replicating followers are probed one entry at a time. |
| 188 | if pr.Next-1 != rejected { |
| 189 | return false |
| 190 | } |
| 191 | |
| 192 | if pr.Next = min(rejected, last+1); pr.Next < 1 { |
| 193 | pr.Next = 1 |
| 194 | } |
| 195 | pr.ProbeSent = false |
| 196 | return true |
| 197 | } |
| 198 | |
| 199 | // IsPaused returns whether sending log entries to this node has been throttled. |
| 200 | // This is done when a node has rejected recent MsgApps, is currently waiting |
| 201 | // for a snapshot, or has reached the MaxInflightMsgs limit. In normal |
| 202 | // operation, this is false. A throttled node will be contacted less frequently |
| 203 | // until it has reached a state in which it's able to accept a steady stream of |
| 204 | // log entries again. |
| 205 | func (pr *Progress) IsPaused() bool { |
| 206 | switch pr.State { |
| 207 | case StateProbe: |
| 208 | return pr.ProbeSent |
| 209 | case StateReplicate: |
| 210 | return pr.Inflights.Full() |
| 211 | case StateSnapshot: |
| 212 | return true |
| 213 | default: |
| 214 | panic("unexpected state") |
| 215 | } |
| 216 | } |
| 217 | |
| 218 | func (pr *Progress) String() string { |
| 219 | var buf strings.Builder |
| 220 | fmt.Fprintf(&buf, "%s match=%d next=%d", pr.State, pr.Match, pr.Next) |
| 221 | if pr.IsLearner { |
| 222 | fmt.Fprint(&buf, " learner") |
| 223 | } |
| 224 | if pr.IsPaused() { |
| 225 | fmt.Fprint(&buf, " paused") |
| 226 | } |
| 227 | if pr.PendingSnapshot > 0 { |
| 228 | fmt.Fprintf(&buf, " pendingSnap=%d", pr.PendingSnapshot) |
| 229 | } |
| 230 | if !pr.RecentActive { |
| 231 | fmt.Fprintf(&buf, " inactive") |
| 232 | } |
| 233 | if n := pr.Inflights.Count(); n > 0 { |
| 234 | fmt.Fprintf(&buf, " inflight=%d", n) |
| 235 | if pr.Inflights.Full() { |
| 236 | fmt.Fprint(&buf, "[full]") |
| 237 | } |
| 238 | } |
| 239 | return buf.String() |
| 240 | } |
| 241 | |
| 242 | // ProgressMap is a map of *Progress. |
| 243 | type ProgressMap map[uint64]*Progress |
| 244 | |
| 245 | // String prints the ProgressMap in sorted key order, one Progress per line. |
| 246 | func (m ProgressMap) String() string { |
| 247 | ids := make([]uint64, 0, len(m)) |
| 248 | for k := range m { |
| 249 | ids = append(ids, k) |
| 250 | } |
| 251 | sort.Slice(ids, func(i, j int) bool { |
| 252 | return ids[i] < ids[j] |
| 253 | }) |
| 254 | var buf strings.Builder |
| 255 | for _, id := range ids { |
| 256 | fmt.Fprintf(&buf, "%d: %s\n", id, m[id]) |
| 257 | } |
| 258 | return buf.String() |
| 259 | } |