001/* 002 * Licensed to the Apache Software Foundation (ASF) under one 003 * or more contributor license agreements. See the NOTICE file 004 * distributed with this work for additional information 005 * regarding copyright ownership. The ASF licenses this file 006 * to you under the Apache License, Version 2.0 (the 007 * "License"); you may not use this file except in compliance 008 * with the License. You may obtain a copy of the License at 009 * 010 * http://www.apache.org/licenses/LICENSE-2.0 011 * 012 * Unless required by applicable law or agreed to in writing, software 013 * distributed under the License is distributed on an "AS IS" BASIS, 014 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 015 * See the License for the specific language governing permissions and 016 * limitations under the License. 017 */ 018package org.apache.hadoop.hbase.wal; 019 020import java.io.IOException; 021import java.util.ArrayList; 022import java.util.List; 023import java.util.Map; 024import java.util.Set; 025import java.util.TreeSet; 026 027import org.apache.hadoop.hbase.Cell; 028import org.apache.hadoop.hbase.CellUtil; 029import org.apache.hadoop.hbase.HBaseInterfaceAudience; 030import org.apache.hadoop.hbase.PrivateCellUtil; 031import org.apache.hadoop.hbase.KeyValue; 032import org.apache.hadoop.hbase.client.RegionInfo; 033import org.apache.hadoop.hbase.codec.Codec; 034import org.apache.hadoop.hbase.io.HeapSize; 035import org.apache.hadoop.hbase.util.Bytes; 036import org.apache.hadoop.hbase.util.ClassSize; 037import org.apache.hadoop.hbase.util.EnvironmentEdgeManager; 038import org.apache.yetus.audience.InterfaceAudience; 039 040import org.apache.hadoop.hbase.shaded.protobuf.generated.WALProtos; 041import org.apache.hadoop.hbase.shaded.protobuf.generated.WALProtos.CompactionDescriptor; 042import org.apache.hadoop.hbase.shaded.protobuf.generated.WALProtos.FlushDescriptor; 043import org.apache.hadoop.hbase.shaded.protobuf.generated.WALProtos.RegionEventDescriptor; 044 045/** 046 * Used in HBase's transaction log (WAL) to represent a collection of edits (Cell/KeyValue objects) 047 * that came in as a single transaction. All the edits for a given transaction are written out as a 048 * single record, in PB format, followed (optionally) by Cells written via the WALCellEncoder. 049 * <p>This class is LimitedPrivate for CPs to read-only. The {@link #add} methods are 050 * classified as private methods, not for use by CPs.</p> 051 * 052 * <p>A particular WALEdit 'type' is the 'meta' type used to mark key operational 053 * events in the WAL such as compaction, flush, or region open. These meta types do not traverse 054 * hbase memstores. They are edits made by the hbase system rather than edit data submitted by 055 * clients. They only show in the WAL. These 'Meta' types have not been formally specified 056 * (or made into an explicit class type). They evolved organically. HBASE-8457 suggests codifying 057 * a WALEdit 'type' by adding a type field to WALEdit that gets serialized into the WAL. TODO. 058 * Would have to work on the consumption-side. Reading WALs on replay we seem to consume 059 * a Cell-at-a-time rather than by WALEdit. We are already in the below going out of our 060 * way to figure particular types -- e.g. if a compaction, replay, or close meta Marker -- during 061 * normal processing so would make sense to do this. Current system is an awkward marking of Cell 062 * columnfamily as {@link #METAFAMILY} and then setting qualifier based off meta edit type. For 063 * replay-time where we read Cell-at-a-time, there are utility methods below for figuring 064 * meta type. See also 065 * {@link #createBulkLoadEvent(RegionInfo, WALProtos.BulkLoadDescriptor)}, etc., for where we 066 * create meta WALEdit instances.</p> 067 * 068 * <p>WALEdit will accumulate a Set of all column family names referenced by the Cells 069 * {@link #add(Cell)}'d. This is an optimization. Usually when loading a WALEdit, we have the 070 * column family name to-hand.. just shove it into the WALEdit if available. Doing this, we can 071 * save on a parse of each Cell to figure column family down the line when we go to add the 072 * WALEdit to the WAL file. See the hand-off in FSWALEntry Constructor. 073 * @see WALKey 074 */ 075// TODO: Do not expose this class to Coprocessors. It has set methods. A CP might meddle. 076@InterfaceAudience.LimitedPrivate({ HBaseInterfaceAudience.REPLICATION, 077 HBaseInterfaceAudience.COPROC }) 078public class WALEdit implements HeapSize { 079 // Below defines are for writing WALEdit 'meta' Cells.. 080 // TODO: Get rid of this system of special 'meta' Cells. See HBASE-8457. It suggests 081 // adding a type to WALEdit itself for use denoting meta Edits and their types. 082 public static final byte [] METAFAMILY = Bytes.toBytes("METAFAMILY"); 083 084 /** 085 * @deprecated Since 2.3.0. Not used. 086 */ 087 @Deprecated 088 public static final byte [] METAROW = Bytes.toBytes("METAROW"); 089 090 /** 091 * @deprecated Since 2.3.0. Make it protected, internal-use only. Use 092 * {@link #isCompactionMarker(Cell)} 093 */ 094 @Deprecated 095 @InterfaceAudience.Private 096 public static final byte[] COMPACTION = Bytes.toBytes("HBASE::COMPACTION"); 097 098 /** 099 * @deprecated Since 2.3.0. Make it protected, internal-use only. 100 */ 101 @Deprecated 102 @InterfaceAudience.Private 103 public static final byte [] FLUSH = Bytes.toBytes("HBASE::FLUSH"); 104 105 /** 106 * Qualifier for region event meta 'Marker' WALEdits start with the 107 * {@link #REGION_EVENT_PREFIX} prefix ('HBASE::REGION_EVENT::'). After the prefix, 108 * we note the type of the event which we get from the RegionEventDescriptor protobuf 109 * instance type (A RegionEventDescriptor protobuf instance is written as the meta Marker 110 * Cell value). Adding a type suffix means we do not have to deserialize the protobuf to 111 * figure out what type of event this is.. .just read the qualifier suffix. For example, 112 * a close region event descriptor will have a qualifier of HBASE::REGION_EVENT::REGION_CLOSE. 113 * See WAL.proto and the EventType in RegionEventDescriptor protos for all possible 114 * event types. 115 */ 116 private static final String REGION_EVENT_STR = "HBASE::REGION_EVENT"; 117 private static final String REGION_EVENT_PREFIX_STR = REGION_EVENT_STR + "::"; 118 private static final byte [] REGION_EVENT_PREFIX = Bytes.toBytes(REGION_EVENT_PREFIX_STR); 119 120 /** 121 * @deprecated Since 2.3.0. Remove. Not for external use. Not used. 122 */ 123 @Deprecated 124 public static final byte [] REGION_EVENT = Bytes.toBytes(REGION_EVENT_STR); 125 126 /** 127 * We use this define figuring if we are carrying a close event. 128 */ 129 private static final byte [] REGION_EVENT_CLOSE = 130 createRegionEventDescriptorQualifier(RegionEventDescriptor.EventType.REGION_CLOSE); 131 132 @InterfaceAudience.Private 133 public static final byte [] BULK_LOAD = Bytes.toBytes("HBASE::BULK_LOAD"); 134 135 private final transient boolean replay; 136 137 private ArrayList<Cell> cells; 138 139 /** 140 * All the Cell families in <code>cells</code>. Updated by {@link #add(Cell)} and 141 * {@link #add(Map)}. This Set is passed to the FSWALEntry so it does not have 142 * to recalculate the Set of families in a transaction; makes for a bunch of CPU savings. 143 */ 144 private Set<byte []> families = null; 145 146 public WALEdit() { 147 this(1, false); 148 } 149 150 /** 151 * @deprecated since 2.0.1 and will be removed in 4.0.0. Use {@link #WALEdit(int, boolean)} 152 * instead. 153 * @see #WALEdit(int, boolean) 154 * @see <a href="https://issues.apache.org/jira/browse/HBASE-20781">HBASE-20781</a> 155 */ 156 @Deprecated 157 public WALEdit(boolean replay) { 158 this(1, replay); 159 } 160 161 /** 162 * @deprecated since 2.0.1 and will be removed in 4.0.0. Use {@link #WALEdit(int, boolean)} 163 * instead. 164 * @see #WALEdit(int, boolean) 165 * @see <a href="https://issues.apache.org/jira/browse/HBASE-20781">HBASE-20781</a> 166 */ 167 @Deprecated 168 public WALEdit(int cellCount) { 169 this(cellCount, false); 170 } 171 172 /** 173 * @param cellCount Pass so can pre-size the WALEdit. Optimization. 174 */ 175 public WALEdit(int cellCount, boolean isReplay) { 176 this.replay = isReplay; 177 cells = new ArrayList<>(cellCount); 178 } 179 180 private Set<byte[]> getOrCreateFamilies() { 181 if (this.families == null) { 182 this.families = new TreeSet<>(Bytes.BYTES_COMPARATOR); 183 } 184 return this.families; 185 } 186 187 /** 188 * For use by FSWALEntry ONLY. An optimization. 189 * @return All families in {@link #getCells()}; may be null. 190 */ 191 public Set<byte []> getFamilies() { 192 return this.families; 193 } 194 195 /** 196 * @return True is <code>f</code> is {@link #METAFAMILY} 197 * @deprecated Since 2.3.0. Do not expose. Make protected. 198 */ 199 @Deprecated 200 public static boolean isMetaEditFamily(final byte [] f) { 201 return Bytes.equals(METAFAMILY, f); 202 } 203 204 /** 205 * Replaying WALs can read Cell-at-a-time so need this method in those cases. 206 */ 207 public static boolean isMetaEditFamily(Cell cell) { 208 return CellUtil.matchingFamily(cell, METAFAMILY); 209 } 210 211 /** 212 * @return True if this is a meta edit; has one edit only and its columnfamily 213 * is {@link #METAFAMILY}. 214 */ 215 public boolean isMetaEdit() { 216 return this.families != null && this.families.size() == 1 && this.families.contains(METAFAMILY); 217 } 218 219 /** 220 * @return True when current WALEdit is created by log replay. Replication skips WALEdits from 221 * replay. 222 */ 223 public boolean isReplay() { 224 return this.replay; 225 } 226 227 @InterfaceAudience.Private 228 public WALEdit add(Cell cell, byte [] family) { 229 getOrCreateFamilies().add(family); 230 return addCell(cell); 231 } 232 233 @InterfaceAudience.Private 234 public WALEdit add(Cell cell) { 235 // We clone Family each time we add a Cell. Expensive but safe. For CPU savings, use 236 // add(Map) or add(Cell, family). 237 return add(cell, CellUtil.cloneFamily(cell)); 238 } 239 240 public boolean isEmpty() { 241 return cells.isEmpty(); 242 } 243 244 public int size() { 245 return cells.size(); 246 } 247 248 public ArrayList<Cell> getCells() { 249 return cells; 250 } 251 252 /** 253 * This is not thread safe. 254 * This will change the WALEdit and shouldn't be used unless you are sure that nothing 255 * else depends on the contents being immutable. 256 * 257 * @param cells the list of cells that this WALEdit now contains. 258 */ 259 @InterfaceAudience.Private 260 // Used by replay. 261 public void setCells(ArrayList<Cell> cells) { 262 this.cells = cells; 263 this.families = null; 264 } 265 266 /** 267 * Reads WALEdit from cells. 268 * @param cellDecoder Cell decoder. 269 * @param expectedCount Expected cell count. 270 * @return Number of KVs read. 271 */ 272 public int readFromCells(Codec.Decoder cellDecoder, int expectedCount) throws IOException { 273 cells.clear(); 274 cells.ensureCapacity(expectedCount); 275 while (cells.size() < expectedCount && cellDecoder.advance()) { 276 add(cellDecoder.current()); 277 } 278 return cells.size(); 279 } 280 281 @Override 282 public long heapSize() { 283 long ret = ClassSize.ARRAYLIST; 284 for (Cell cell : cells) { 285 ret += cell.heapSize(); 286 } 287 return ret; 288 } 289 290 public long estimatedSerializedSizeOf() { 291 long ret = 0; 292 for (Cell cell: cells) { 293 ret += PrivateCellUtil.estimatedSerializedSizeOf(cell); 294 } 295 return ret; 296 } 297 298 @Override 299 public String toString() { 300 StringBuilder sb = new StringBuilder(); 301 302 sb.append("[#edits: ").append(cells.size()).append(" = <"); 303 for (Cell cell : cells) { 304 sb.append(cell); 305 sb.append("; "); 306 } 307 sb.append(">]"); 308 return sb.toString(); 309 } 310 311 public static WALEdit createFlushWALEdit(RegionInfo hri, FlushDescriptor f) { 312 KeyValue kv = new KeyValue(getRowForRegion(hri), METAFAMILY, FLUSH, 313 EnvironmentEdgeManager.currentTime(), f.toByteArray()); 314 return new WALEdit().add(kv, METAFAMILY); 315 } 316 317 public static FlushDescriptor getFlushDescriptor(Cell cell) throws IOException { 318 return CellUtil.matchingColumn(cell, METAFAMILY, FLUSH)? 319 FlushDescriptor.parseFrom(CellUtil.cloneValue(cell)): null; 320 } 321 322 /** 323 * @return A meta Marker WALEdit that has a single Cell whose value is the passed in 324 * <code>regionEventDesc</code> serialized and whose row is this region, 325 * columnfamily is {@link #METAFAMILY} and qualifier is 326 * {@link #REGION_EVENT_PREFIX} + {@link RegionEventDescriptor#getEventType()}; 327 * for example HBASE::REGION_EVENT::REGION_CLOSE. 328 */ 329 public static WALEdit createRegionEventWALEdit(RegionInfo hri, 330 RegionEventDescriptor regionEventDesc) { 331 return createRegionEventWALEdit(getRowForRegion(hri), regionEventDesc); 332 } 333 334 @InterfaceAudience.Private 335 public static WALEdit createRegionEventWALEdit(byte [] rowForRegion, 336 RegionEventDescriptor regionEventDesc) { 337 KeyValue kv = new KeyValue(rowForRegion, METAFAMILY, 338 createRegionEventDescriptorQualifier(regionEventDesc.getEventType()), 339 EnvironmentEdgeManager.currentTime(), regionEventDesc.toByteArray()); 340 return new WALEdit().add(kv, METAFAMILY); 341 } 342 343 /** 344 * @return Cell qualifier for the passed in RegionEventDescriptor Type; e.g. we'll 345 * return something like a byte array with HBASE::REGION_EVENT::REGION_OPEN in it. 346 */ 347 @InterfaceAudience.Private 348 public static byte [] createRegionEventDescriptorQualifier(RegionEventDescriptor.EventType t) { 349 return Bytes.toBytes(REGION_EVENT_PREFIX_STR + t.toString()); 350 } 351 352 /** 353 * Public so can be accessed from regionserver.wal package. 354 * @return True if this is a Marker Edit and it is a RegionClose type. 355 */ 356 public boolean isRegionCloseMarker() { 357 return isMetaEdit() && PrivateCellUtil.matchingQualifier(this.cells.get(0), 358 REGION_EVENT_CLOSE, 0, REGION_EVENT_CLOSE.length); 359 } 360 361 /** 362 * @return Returns a RegionEventDescriptor made by deserializing the content of the 363 * passed in <code>cell</code>, IFF the <code>cell</code> is a RegionEventDescriptor 364 * type WALEdit. 365 */ 366 public static RegionEventDescriptor getRegionEventDescriptor(Cell cell) throws IOException { 367 return CellUtil.matchingColumnFamilyAndQualifierPrefix(cell, METAFAMILY, REGION_EVENT_PREFIX)? 368 RegionEventDescriptor.parseFrom(CellUtil.cloneValue(cell)): null; 369 } 370 371 /** 372 * @return A Marker WALEdit that has <code>c</code> serialized as its value 373 */ 374 public static WALEdit createCompaction(final RegionInfo hri, final CompactionDescriptor c) { 375 byte [] pbbytes = c.toByteArray(); 376 KeyValue kv = new KeyValue(getRowForRegion(hri), METAFAMILY, COMPACTION, 377 EnvironmentEdgeManager.currentTime(), pbbytes); 378 return new WALEdit().add(kv, METAFAMILY); //replication scope null so this won't be replicated 379 } 380 381 public static byte[] getRowForRegion(RegionInfo hri) { 382 byte[] startKey = hri.getStartKey(); 383 if (startKey.length == 0) { 384 // empty row key is not allowed in mutations because it is both the start key and the end key 385 // we return the smallest byte[] that is bigger (in lex comparison) than byte[0]. 386 return new byte[] {0}; 387 } 388 return startKey; 389 } 390 391 /** 392 * Deserialized and returns a CompactionDescriptor is the KeyValue contains one. 393 * @param kv the key value 394 * @return deserialized CompactionDescriptor or null. 395 */ 396 public static CompactionDescriptor getCompaction(Cell kv) throws IOException { 397 return isCompactionMarker(kv)? CompactionDescriptor.parseFrom(CellUtil.cloneValue(kv)): null; 398 } 399 400 /** 401 * Returns true if the given cell is a serialized {@link CompactionDescriptor} 402 * 403 * @see #getCompaction(Cell) 404 */ 405 public static boolean isCompactionMarker(Cell cell) { 406 return CellUtil.matchingColumn(cell, METAFAMILY, COMPACTION); 407 } 408 409 /** 410 * Create a bulk loader WALEdit 411 * 412 * @param hri The RegionInfo for the region in which we are bulk loading 413 * @param bulkLoadDescriptor The descriptor for the Bulk Loader 414 * @return The WALEdit for the BulkLoad 415 */ 416 public static WALEdit createBulkLoadEvent(RegionInfo hri, 417 WALProtos.BulkLoadDescriptor bulkLoadDescriptor) { 418 KeyValue kv = new KeyValue(getRowForRegion(hri), METAFAMILY, BULK_LOAD, 419 EnvironmentEdgeManager.currentTime(), bulkLoadDescriptor.toByteArray()); 420 return new WALEdit().add(kv, METAFAMILY); 421 } 422 423 /** 424 * Deserialized and returns a BulkLoadDescriptor from the passed in Cell 425 * @param cell the key value 426 * @return deserialized BulkLoadDescriptor or null. 427 */ 428 public static WALProtos.BulkLoadDescriptor getBulkLoadDescriptor(Cell cell) throws IOException { 429 return CellUtil.matchingColumn(cell, METAFAMILY, BULK_LOAD)? 430 WALProtos.BulkLoadDescriptor.parseFrom(CellUtil.cloneValue(cell)): null; 431 } 432 433 /** 434 * Append the given map of family->edits to a WALEdit data structure. 435 * This does not write to the WAL itself. 436 * Note that as an optimization, we will stamp the Set of column families into the WALEdit 437 * to save on our having to calculate column families subsequently down in the actual WAL 438 * writing. 439 * 440 * @param familyMap map of family->edits 441 */ 442 public void add(Map<byte[], List<Cell>> familyMap) { 443 for (Map.Entry<byte [], List<Cell>> e: familyMap.entrySet()) { 444 // 'foreach' loop NOT used. See HBASE-12023 "...creates too many iterator objects." 445 int listSize = e.getValue().size(); 446 // Add all Cells first and then at end, add the family rather than call {@link #add(Cell)} 447 // and have it clone family each time. Optimization! 448 for (int i = 0; i < listSize; i++) { 449 addCell(e.getValue().get(i)); 450 } 451 addFamily(e.getKey()); 452 } 453 } 454 455 private void addFamily(byte [] family) { 456 getOrCreateFamilies().add(family); 457 } 458 459 private WALEdit addCell(Cell cell) { 460 this.cells.add(cell); 461 return this; 462 } 463}