khenaidoo | ab1f7bd | 2019-11-14 14:00:27 -0500 | [diff] [blame] | 1 | // Copyright 2017 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 backend |
| 16 | |
| 17 | import ( |
| 18 | "bytes" |
| 19 | "math" |
| 20 | "sync" |
| 21 | |
| 22 | bolt "go.etcd.io/bbolt" |
| 23 | ) |
| 24 | |
| 25 | // safeRangeBucket is a hack to avoid inadvertently reading duplicate keys; |
| 26 | // overwrites on a bucket should only fetch with limit=1, but safeRangeBucket |
| 27 | // is known to never overwrite any key so range is safe. |
| 28 | var safeRangeBucket = []byte("key") |
| 29 | |
| 30 | type ReadTx interface { |
| 31 | Lock() |
| 32 | Unlock() |
| 33 | RLock() |
| 34 | RUnlock() |
| 35 | |
| 36 | UnsafeRange(bucketName []byte, key, endKey []byte, limit int64) (keys [][]byte, vals [][]byte) |
| 37 | UnsafeForEach(bucketName []byte, visitor func(k, v []byte) error) error |
| 38 | } |
| 39 | |
| 40 | type readTx struct { |
| 41 | // mu protects accesses to the txReadBuffer |
| 42 | mu sync.RWMutex |
| 43 | buf txReadBuffer |
| 44 | |
| 45 | // TODO: group and encapsulate {txMu, tx, buckets, txWg}, as they share the same lifecycle. |
| 46 | // txMu protects accesses to buckets and tx on Range requests. |
| 47 | txMu sync.RWMutex |
| 48 | tx *bolt.Tx |
| 49 | buckets map[string]*bolt.Bucket |
| 50 | // txWg protects tx from being rolled back at the end of a batch interval until all reads using this tx are done. |
| 51 | txWg *sync.WaitGroup |
| 52 | } |
| 53 | |
| 54 | func (rt *readTx) Lock() { rt.mu.Lock() } |
| 55 | func (rt *readTx) Unlock() { rt.mu.Unlock() } |
| 56 | func (rt *readTx) RLock() { rt.mu.RLock() } |
| 57 | func (rt *readTx) RUnlock() { rt.mu.RUnlock() } |
| 58 | |
| 59 | func (rt *readTx) UnsafeRange(bucketName, key, endKey []byte, limit int64) ([][]byte, [][]byte) { |
| 60 | if endKey == nil { |
| 61 | // forbid duplicates for single keys |
| 62 | limit = 1 |
| 63 | } |
| 64 | if limit <= 0 { |
| 65 | limit = math.MaxInt64 |
| 66 | } |
| 67 | if limit > 1 && !bytes.Equal(bucketName, safeRangeBucket) { |
| 68 | panic("do not use unsafeRange on non-keys bucket") |
| 69 | } |
| 70 | keys, vals := rt.buf.Range(bucketName, key, endKey, limit) |
| 71 | if int64(len(keys)) == limit { |
| 72 | return keys, vals |
| 73 | } |
| 74 | |
| 75 | // find/cache bucket |
| 76 | bn := string(bucketName) |
| 77 | rt.txMu.RLock() |
| 78 | bucket, ok := rt.buckets[bn] |
| 79 | rt.txMu.RUnlock() |
| 80 | if !ok { |
| 81 | rt.txMu.Lock() |
| 82 | bucket = rt.tx.Bucket(bucketName) |
| 83 | rt.buckets[bn] = bucket |
| 84 | rt.txMu.Unlock() |
| 85 | } |
| 86 | |
| 87 | // ignore missing bucket since may have been created in this batch |
| 88 | if bucket == nil { |
| 89 | return keys, vals |
| 90 | } |
| 91 | rt.txMu.Lock() |
| 92 | c := bucket.Cursor() |
| 93 | rt.txMu.Unlock() |
| 94 | |
| 95 | k2, v2 := unsafeRange(c, key, endKey, limit-int64(len(keys))) |
| 96 | return append(k2, keys...), append(v2, vals...) |
| 97 | } |
| 98 | |
| 99 | func (rt *readTx) UnsafeForEach(bucketName []byte, visitor func(k, v []byte) error) error { |
| 100 | dups := make(map[string]struct{}) |
| 101 | getDups := func(k, v []byte) error { |
| 102 | dups[string(k)] = struct{}{} |
| 103 | return nil |
| 104 | } |
| 105 | visitNoDup := func(k, v []byte) error { |
| 106 | if _, ok := dups[string(k)]; ok { |
| 107 | return nil |
| 108 | } |
| 109 | return visitor(k, v) |
| 110 | } |
| 111 | if err := rt.buf.ForEach(bucketName, getDups); err != nil { |
| 112 | return err |
| 113 | } |
| 114 | rt.txMu.Lock() |
| 115 | err := unsafeForEach(rt.tx, bucketName, visitNoDup) |
| 116 | rt.txMu.Unlock() |
| 117 | if err != nil { |
| 118 | return err |
| 119 | } |
| 120 | return rt.buf.ForEach(bucketName, visitor) |
| 121 | } |
| 122 | |
| 123 | func (rt *readTx) reset() { |
| 124 | rt.buf.reset() |
| 125 | rt.buckets = make(map[string]*bolt.Bucket) |
| 126 | rt.tx = nil |
| 127 | rt.txWg = new(sync.WaitGroup) |
| 128 | } |
| 129 | |
| 130 | // TODO: create a base type for readTx and concurrentReadTx to avoid duplicated function implementation? |
| 131 | type concurrentReadTx struct { |
| 132 | buf txReadBuffer |
| 133 | txMu *sync.RWMutex |
| 134 | tx *bolt.Tx |
| 135 | buckets map[string]*bolt.Bucket |
| 136 | txWg *sync.WaitGroup |
| 137 | } |
| 138 | |
| 139 | func (rt *concurrentReadTx) Lock() {} |
| 140 | func (rt *concurrentReadTx) Unlock() {} |
| 141 | |
| 142 | // RLock is no-op. concurrentReadTx does not need to be locked after it is created. |
| 143 | func (rt *concurrentReadTx) RLock() {} |
| 144 | |
| 145 | // RUnlock signals the end of concurrentReadTx. |
| 146 | func (rt *concurrentReadTx) RUnlock() { rt.txWg.Done() } |
| 147 | |
| 148 | func (rt *concurrentReadTx) UnsafeForEach(bucketName []byte, visitor func(k, v []byte) error) error { |
| 149 | dups := make(map[string]struct{}) |
| 150 | getDups := func(k, v []byte) error { |
| 151 | dups[string(k)] = struct{}{} |
| 152 | return nil |
| 153 | } |
| 154 | visitNoDup := func(k, v []byte) error { |
| 155 | if _, ok := dups[string(k)]; ok { |
| 156 | return nil |
| 157 | } |
| 158 | return visitor(k, v) |
| 159 | } |
| 160 | if err := rt.buf.ForEach(bucketName, getDups); err != nil { |
| 161 | return err |
| 162 | } |
| 163 | rt.txMu.Lock() |
| 164 | err := unsafeForEach(rt.tx, bucketName, visitNoDup) |
| 165 | rt.txMu.Unlock() |
| 166 | if err != nil { |
| 167 | return err |
| 168 | } |
| 169 | return rt.buf.ForEach(bucketName, visitor) |
| 170 | } |
| 171 | |
| 172 | func (rt *concurrentReadTx) UnsafeRange(bucketName, key, endKey []byte, limit int64) ([][]byte, [][]byte) { |
| 173 | if endKey == nil { |
| 174 | // forbid duplicates for single keys |
| 175 | limit = 1 |
| 176 | } |
| 177 | if limit <= 0 { |
| 178 | limit = math.MaxInt64 |
| 179 | } |
| 180 | if limit > 1 && !bytes.Equal(bucketName, safeRangeBucket) { |
| 181 | panic("do not use unsafeRange on non-keys bucket") |
| 182 | } |
| 183 | keys, vals := rt.buf.Range(bucketName, key, endKey, limit) |
| 184 | if int64(len(keys)) == limit { |
| 185 | return keys, vals |
| 186 | } |
| 187 | |
| 188 | // find/cache bucket |
| 189 | bn := string(bucketName) |
| 190 | rt.txMu.RLock() |
| 191 | bucket, ok := rt.buckets[bn] |
| 192 | rt.txMu.RUnlock() |
| 193 | if !ok { |
| 194 | rt.txMu.Lock() |
| 195 | bucket = rt.tx.Bucket(bucketName) |
| 196 | rt.buckets[bn] = bucket |
| 197 | rt.txMu.Unlock() |
| 198 | } |
| 199 | |
| 200 | // ignore missing bucket since may have been created in this batch |
| 201 | if bucket == nil { |
| 202 | return keys, vals |
| 203 | } |
| 204 | rt.txMu.Lock() |
| 205 | c := bucket.Cursor() |
| 206 | rt.txMu.Unlock() |
| 207 | |
| 208 | k2, v2 := unsafeRange(c, key, endKey, limit-int64(len(keys))) |
| 209 | return append(k2, keys...), append(v2, vals...) |
| 210 | } |