|
| 1 | +// Copyright 2025 The Cockroach Authors. |
| 2 | +// |
| 3 | +// Use of this software is governed by the CockroachDB Software License |
| 4 | +// included in the /LICENSE file. |
| 5 | + |
| 6 | +package logical |
| 7 | + |
| 8 | +import ( |
| 9 | + "context" |
| 10 | + |
| 11 | + "github.com/cockroachdb/cockroach/pkg/keys" |
| 12 | + "github.com/cockroachdb/cockroach/pkg/settings/cluster" |
| 13 | + "github.com/cockroachdb/cockroach/pkg/sql/catalog" |
| 14 | + "github.com/cockroachdb/cockroach/pkg/sql/catalog/descpb" |
| 15 | + "github.com/cockroachdb/cockroach/pkg/sql/catalog/descs" |
| 16 | + "github.com/cockroachdb/cockroach/pkg/sql/catalog/lease" |
| 17 | + "github.com/cockroachdb/cockroach/pkg/sql/isql" |
| 18 | + "github.com/cockroachdb/cockroach/pkg/sql/sem/tree" |
| 19 | + "github.com/cockroachdb/cockroach/pkg/sql/sessiondata" |
| 20 | + "github.com/cockroachdb/errors" |
| 21 | +) |
| 22 | + |
| 23 | +// tableHandler applies batches of replication events that are destined for a |
| 24 | +// sinlgle table. |
| 25 | +type tableHandler struct { |
| 26 | + sqlReader *sqlRowReader |
| 27 | + sqlWriter *sqlRowWriter |
| 28 | + db descs.DB |
| 29 | + tombstoneUpdater *tombstoneUpdater |
| 30 | +} |
| 31 | + |
| 32 | +type tableBatchStats struct { |
| 33 | + // refreshedRows are the number of rows that were re-read from the local |
| 34 | + // database. |
| 35 | + refreshedRows int64 |
| 36 | + // inserts is the number of rows that were inserted. |
| 37 | + inserts int64 |
| 38 | + // updates is the number of rows that were updated. |
| 39 | + updates int64 |
| 40 | + // deletes is the number of rows that were deleted. |
| 41 | + deletes int64 |
| 42 | + // tombstoneUpdates is the number of tombstones that were updated. This case |
| 43 | + // only occurs if the event is a replicated delete and the row did not exist |
| 44 | + // locally. |
| 45 | + tombstoneUpdates int64 |
| 46 | + // refreshLwwLosers is the number of rows that were dropped as lww losers |
| 47 | + // after reading them locally. |
| 48 | + refreshLwwLosers int64 |
| 49 | + // kvLwwLosers is the number of rows that were dropped as lww losers after |
| 50 | + // attempting to write to the KV layer. This case only occurs if there is a |
| 51 | + // tombstone that is more recent than the replicated action. |
| 52 | + kvLwwLosers int64 |
| 53 | +} |
| 54 | + |
| 55 | +func (t *tableBatchStats) Add(o tableBatchStats) { |
| 56 | + t.refreshedRows += o.refreshedRows |
| 57 | + t.inserts += o.inserts |
| 58 | + t.updates += o.updates |
| 59 | + t.deletes += o.deletes |
| 60 | + t.tombstoneUpdates += o.tombstoneUpdates |
| 61 | + t.refreshLwwLosers += o.refreshLwwLosers |
| 62 | + t.kvLwwLosers += o.kvLwwLosers |
| 63 | +} |
| 64 | + |
| 65 | +// newTableHandler creates a new tableHandler for the given table descriptor ID. |
| 66 | +// It internally constructs the sqlReader and sqlWriter components. |
| 67 | +func newTableHandler( |
| 68 | + tableID descpb.ID, |
| 69 | + db descs.DB, |
| 70 | + codec keys.SQLCodec, |
| 71 | + sd *sessiondata.SessionData, |
| 72 | + leaseMgr *lease.Manager, |
| 73 | + settings *cluster.Settings, |
| 74 | +) (*tableHandler, error) { |
| 75 | + var table catalog.TableDescriptor |
| 76 | + |
| 77 | + // NOTE: we don't hold a lease on the table descriptor, but validation |
| 78 | + // prevents users from changing the set of columns or the primary key of an |
| 79 | + // LDR replicated table. |
| 80 | + err := db.DescsTxn(context.Background(), func(ctx context.Context, txn descs.Txn) error { |
| 81 | + var err error |
| 82 | + table, err = txn.Descriptors().GetLeasedImmutableTableByID(ctx, txn.KV(), tableID) |
| 83 | + return err |
| 84 | + }) |
| 85 | + if err != nil { |
| 86 | + return nil, err |
| 87 | + } |
| 88 | + |
| 89 | + reader, err := newSQLRowReader(table) |
| 90 | + if err != nil { |
| 91 | + return nil, err |
| 92 | + } |
| 93 | + |
| 94 | + writer, err := newSQLRowWriter(table) |
| 95 | + if err != nil { |
| 96 | + return nil, err |
| 97 | + } |
| 98 | + |
| 99 | + tombstoneUpdater := newTombstoneUpdater(codec, db.KV(), leaseMgr, tableID, sd, settings) |
| 100 | + |
| 101 | + return &tableHandler{ |
| 102 | + sqlReader: reader, |
| 103 | + sqlWriter: writer, |
| 104 | + db: db, |
| 105 | + tombstoneUpdater: tombstoneUpdater, |
| 106 | + }, nil |
| 107 | +} |
| 108 | + |
| 109 | +func (t *tableHandler) handleDecodedBatch( |
| 110 | + ctx context.Context, batch []decodedEvent, |
| 111 | +) (tableBatchStats, error) { |
| 112 | + stats, err := t.attemptBatch(ctx, batch) |
| 113 | + if err == nil { |
| 114 | + return stats, nil |
| 115 | + } |
| 116 | + |
| 117 | + refreshedBatch, refreshStats, err := t.refreshPrevRows(ctx, batch) |
| 118 | + if err != nil { |
| 119 | + return stats, err |
| 120 | + } |
| 121 | + |
| 122 | + stats, err = t.attemptBatch(ctx, refreshedBatch) |
| 123 | + if err != nil { |
| 124 | + return tableBatchStats{}, err |
| 125 | + } |
| 126 | + |
| 127 | + stats.Add(refreshStats) |
| 128 | + |
| 129 | + return stats, nil |
| 130 | +} |
| 131 | + |
| 132 | +func (t *tableHandler) attemptBatch( |
| 133 | + ctx context.Context, batch []decodedEvent, |
| 134 | +) (tableBatchStats, error) { |
| 135 | + var stats tableBatchStats |
| 136 | + err := t.db.Txn(ctx, func(ctx context.Context, txn isql.Txn) error { |
| 137 | + for _, event := range batch { |
| 138 | + switch { |
| 139 | + case event.isDelete && len(event.prevRow) != 0: |
| 140 | + stats.deletes++ |
| 141 | + err := t.sqlWriter.DeleteRow(ctx, txn, event.originTimestamp, event.prevRow) |
| 142 | + if err != nil { |
| 143 | + return err |
| 144 | + } |
| 145 | + case event.isDelete && len(event.prevRow) == 0: |
| 146 | + stats.tombstoneUpdates++ |
| 147 | + tombstoneUpdateStats, err := t.tombstoneUpdater.updateTombstone(ctx, txn, event.originTimestamp, event.row) |
| 148 | + if err != nil { |
| 149 | + return err |
| 150 | + } |
| 151 | + stats.kvLwwLosers += tombstoneUpdateStats.kvWriteTooOld |
| 152 | + case event.prevRow == nil: |
| 153 | + stats.inserts++ |
| 154 | + err := t.sqlWriter.InsertRow(ctx, txn, event.originTimestamp, event.row) |
| 155 | + if isLwwLoser(err) { |
| 156 | + // Insert may observe a LWW failure if it attempts to write over a tombstone. |
| 157 | + stats.kvLwwLosers++ |
| 158 | + continue |
| 159 | + } |
| 160 | + if err != nil { |
| 161 | + return err |
| 162 | + } |
| 163 | + case event.prevRow != nil: |
| 164 | + stats.updates++ |
| 165 | + err := t.sqlWriter.UpdateRow(ctx, txn, event.originTimestamp, event.prevRow, event.row) |
| 166 | + if err != nil { |
| 167 | + return err |
| 168 | + } |
| 169 | + default: |
| 170 | + return errors.AssertionFailedf("unhandled event type: %v", event) |
| 171 | + } |
| 172 | + } |
| 173 | + return nil |
| 174 | + }) |
| 175 | + if err != nil { |
| 176 | + return tableBatchStats{}, err |
| 177 | + } |
| 178 | + return stats, nil |
| 179 | +} |
| 180 | + |
| 181 | +// refreshPrevRows refreshes the prevRow field for each event in the batch. If |
| 182 | +// any event is known to be a lww loser based on the read, its dropped from the |
| 183 | +// batch. |
| 184 | +func (t *tableHandler) refreshPrevRows( |
| 185 | + ctx context.Context, batch []decodedEvent, |
| 186 | +) ([]decodedEvent, tableBatchStats, error) { |
| 187 | + var stats tableBatchStats |
| 188 | + stats.refreshedRows = int64(len(batch)) |
| 189 | + |
| 190 | + rows := make([]tree.Datums, 0, len(batch)) |
| 191 | + for _, event := range batch { |
| 192 | + rows = append(rows, event.row) |
| 193 | + } |
| 194 | + |
| 195 | + var refreshedRows map[int]priorRow |
| 196 | + err := t.db.Txn(ctx, func(ctx context.Context, txn isql.Txn) error { |
| 197 | + var err error |
| 198 | + // TODO(jeffswenson): should we apply the batch in the same transaction |
| 199 | + // that we perform the read refresh? We could maybe even use locking reads. |
| 200 | + refreshedRows, err = t.sqlReader.ReadRows(ctx, txn, rows) |
| 201 | + return err |
| 202 | + }) |
| 203 | + if err != nil { |
| 204 | + return nil, tableBatchStats{}, err |
| 205 | + } |
| 206 | + |
| 207 | + refreshedBatch := make([]decodedEvent, 0, len(batch)) |
| 208 | + for i, event := range batch { |
| 209 | + var prevRow tree.Datums |
| 210 | + if refreshed, found := refreshedRows[i]; found { |
| 211 | + if !refreshed.logicalTimestamp.Less(event.originTimestamp) { |
| 212 | + // TODO(jeffswenson): update this logic when its time to handle |
| 213 | + // ties. |
| 214 | + // Skip the row because it is a lww loser. Note: we can only identify LWW |
| 215 | + // losers during the read refresh if the row exists. If its a tombstone, |
| 216 | + // the local value may win LWW, but we have to attempt the |
| 217 | + // insert/tombstone update to find out. We even have to do this if the |
| 218 | + // replicated event is a delete because the local tombstone may have an |
| 219 | + // older logical timestamp. |
| 220 | + stats.refreshLwwLosers++ |
| 221 | + continue |
| 222 | + } |
| 223 | + prevRow = refreshed.row |
| 224 | + } |
| 225 | + refreshedEvent := decodedEvent{ |
| 226 | + dstDescID: event.dstDescID, |
| 227 | + isDelete: event.isDelete, |
| 228 | + row: event.row, |
| 229 | + originTimestamp: event.originTimestamp, |
| 230 | + prevRow: prevRow, |
| 231 | + } |
| 232 | + refreshedBatch = append(refreshedBatch, refreshedEvent) |
| 233 | + } |
| 234 | + |
| 235 | + return refreshedBatch, stats, nil |
| 236 | +} |
| 237 | + |
| 238 | +func (t *tableHandler) ReleaseLeases(ctx context.Context) { |
| 239 | + t.tombstoneUpdater.ReleaseLeases(ctx) |
| 240 | +} |
![craig[bot]](https://github.githubassets.com/images/gravatars/gravatar-user-420.png?size=40){kind=avatar}
0 commit comments