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.client; 019 020import static org.apache.hadoop.hbase.HConstants.DEFAULT_USE_META_REPLICAS; 021import static org.apache.hadoop.hbase.HConstants.EMPTY_END_ROW; 022import static org.apache.hadoop.hbase.HConstants.NINES; 023import static org.apache.hadoop.hbase.HConstants.USE_META_REPLICAS; 024import static org.apache.hadoop.hbase.HConstants.ZEROES; 025import static org.apache.hadoop.hbase.TableName.META_TABLE_NAME; 026import static org.apache.hadoop.hbase.client.AsyncRegionLocatorHelper.canUpdateOnError; 027import static org.apache.hadoop.hbase.client.AsyncRegionLocatorHelper.createRegionLocations; 028import static org.apache.hadoop.hbase.client.AsyncRegionLocatorHelper.isGood; 029import static org.apache.hadoop.hbase.client.AsyncRegionLocatorHelper.removeRegionLocation; 030import static org.apache.hadoop.hbase.client.ConnectionUtils.createClosestRowAfter; 031import static org.apache.hadoop.hbase.client.ConnectionUtils.isEmptyStopRow; 032import static org.apache.hadoop.hbase.client.RegionInfo.createRegionName; 033import static org.apache.hadoop.hbase.client.RegionLocator.LOCATOR_META_REPLICAS_MODE; 034import static org.apache.hadoop.hbase.util.Bytes.BYTES_COMPARATOR; 035import static org.apache.hadoop.hbase.util.ConcurrentMapUtils.computeIfAbsent; 036 037import java.io.IOException; 038import java.util.Arrays; 039import java.util.HashSet; 040import java.util.Iterator; 041import java.util.LinkedHashMap; 042import java.util.Map; 043import java.util.Optional; 044import java.util.Set; 045import java.util.concurrent.CompletableFuture; 046import java.util.concurrent.ConcurrentHashMap; 047import java.util.concurrent.ConcurrentMap; 048import java.util.concurrent.ConcurrentNavigableMap; 049import java.util.concurrent.ConcurrentSkipListMap; 050import java.util.concurrent.TimeUnit; 051import org.apache.commons.lang3.ObjectUtils; 052import org.apache.hadoop.hbase.HBaseIOException; 053import org.apache.hadoop.hbase.HConstants; 054import org.apache.hadoop.hbase.HRegionLocation; 055import org.apache.hadoop.hbase.MetaTableAccessor; 056import org.apache.hadoop.hbase.RegionLocations; 057import org.apache.hadoop.hbase.ServerName; 058import org.apache.hadoop.hbase.TableName; 059import org.apache.hadoop.hbase.TableNotFoundException; 060import org.apache.hadoop.hbase.client.Scan.ReadType; 061import org.apache.hadoop.hbase.util.Bytes; 062import org.apache.hbase.thirdparty.com.google.common.base.Objects; 063import org.apache.yetus.audience.InterfaceAudience; 064import org.slf4j.Logger; 065import org.slf4j.LoggerFactory; 066 067/** 068 * The asynchronous locator for regions other than meta. 069 */ 070@InterfaceAudience.Private 071class AsyncNonMetaRegionLocator { 072 073 private static final Logger LOG = LoggerFactory.getLogger(AsyncNonMetaRegionLocator.class); 074 075 static final String MAX_CONCURRENT_LOCATE_REQUEST_PER_TABLE = 076 "hbase.client.meta.max.concurrent.locate.per.table"; 077 078 private static final int DEFAULT_MAX_CONCURRENT_LOCATE_REQUEST_PER_TABLE = 8; 079 080 static String LOCATE_PREFETCH_LIMIT = "hbase.client.locate.prefetch.limit"; 081 082 private static final int DEFAULT_LOCATE_PREFETCH_LIMIT = 10; 083 084 private final AsyncConnectionImpl conn; 085 086 private final int maxConcurrentLocateRequestPerTable; 087 088 private final int locatePrefetchLimit; 089 090 // The mode tells if HedgedRead, LoadBalance mode is supported. 091 // The default mode is CatalogReplicaMode.None. 092 private CatalogReplicaMode metaReplicaMode; 093 private CatalogReplicaLoadBalanceSelector metaReplicaSelector; 094 095 private final ConcurrentMap<TableName, TableCache> cache = new ConcurrentHashMap<>(); 096 097 private static final class LocateRequest { 098 099 private final byte[] row; 100 101 private final RegionLocateType locateType; 102 103 public LocateRequest(byte[] row, RegionLocateType locateType) { 104 this.row = row; 105 this.locateType = locateType; 106 } 107 108 @Override 109 public int hashCode() { 110 return Bytes.hashCode(row) ^ locateType.hashCode(); 111 } 112 113 @Override 114 public boolean equals(Object obj) { 115 if (obj == null || obj.getClass() != LocateRequest.class) { 116 return false; 117 } 118 LocateRequest that = (LocateRequest) obj; 119 return locateType.equals(that.locateType) && Bytes.equals(row, that.row); 120 } 121 } 122 123 private static final class TableCache { 124 125 private final ConcurrentNavigableMap<byte[], RegionLocations> cache = 126 new ConcurrentSkipListMap<>(BYTES_COMPARATOR); 127 128 private final Set<LocateRequest> pendingRequests = new HashSet<>(); 129 130 private final Map<LocateRequest, CompletableFuture<RegionLocations>> allRequests = 131 new LinkedHashMap<>(); 132 133 public boolean hasQuota(int max) { 134 return pendingRequests.size() < max; 135 } 136 137 public boolean isPending(LocateRequest req) { 138 return pendingRequests.contains(req); 139 } 140 141 public void send(LocateRequest req) { 142 pendingRequests.add(req); 143 } 144 145 public Optional<LocateRequest> getCandidate() { 146 return allRequests.keySet().stream().filter(r -> !isPending(r)).findFirst(); 147 } 148 149 public void clearCompletedRequests(RegionLocations locations) { 150 for (Iterator<Map.Entry<LocateRequest, CompletableFuture<RegionLocations>>> iter = 151 allRequests.entrySet().iterator(); iter.hasNext();) { 152 Map.Entry<LocateRequest, CompletableFuture<RegionLocations>> entry = iter.next(); 153 if (tryComplete(entry.getKey(), entry.getValue(), locations)) { 154 iter.remove(); 155 } 156 } 157 } 158 159 private boolean tryComplete(LocateRequest req, CompletableFuture<RegionLocations> future, 160 RegionLocations locations) { 161 if (future.isDone()) { 162 return true; 163 } 164 if (locations == null) { 165 return false; 166 } 167 HRegionLocation loc = ObjectUtils.firstNonNull(locations.getRegionLocations()); 168 // we should at least have one location available, otherwise the request should fail and 169 // should not arrive here 170 assert loc != null; 171 boolean completed; 172 if (req.locateType.equals(RegionLocateType.BEFORE)) { 173 // for locating the row before current row, the common case is to find the previous region 174 // in reverse scan, so we check the endKey first. In general, the condition should be 175 // startKey < req.row and endKey >= req.row. Here we split it to endKey == req.row || 176 // (endKey > req.row && startKey < req.row). The two conditions are equal since startKey < 177 // endKey. 178 byte[] endKey = loc.getRegion().getEndKey(); 179 int c = Bytes.compareTo(endKey, req.row); 180 completed = c == 0 || ((c > 0 || Bytes.equals(EMPTY_END_ROW, endKey)) && 181 Bytes.compareTo(loc.getRegion().getStartKey(), req.row) < 0); 182 } else { 183 completed = loc.getRegion().containsRow(req.row); 184 } 185 if (completed) { 186 future.complete(locations); 187 return true; 188 } else { 189 return false; 190 } 191 } 192 } 193 194 AsyncNonMetaRegionLocator(AsyncConnectionImpl conn) { 195 this.conn = conn; 196 this.maxConcurrentLocateRequestPerTable = conn.getConfiguration().getInt( 197 MAX_CONCURRENT_LOCATE_REQUEST_PER_TABLE, DEFAULT_MAX_CONCURRENT_LOCATE_REQUEST_PER_TABLE); 198 this.locatePrefetchLimit = 199 conn.getConfiguration().getInt(LOCATE_PREFETCH_LIMIT, DEFAULT_LOCATE_PREFETCH_LIMIT); 200 201 // Get the region locator's meta replica mode. 202 this.metaReplicaMode = CatalogReplicaMode.fromString(conn.getConfiguration() 203 .get(LOCATOR_META_REPLICAS_MODE, CatalogReplicaMode.NONE.toString())); 204 205 switch (this.metaReplicaMode) { 206 case LOAD_BALANCE: 207 String replicaSelectorClass = conn.getConfiguration(). 208 get(RegionLocator.LOCATOR_META_REPLICAS_MODE_LOADBALANCE_SELECTOR, 209 CatalogReplicaLoadBalanceSimpleSelector.class.getName()); 210 211 this.metaReplicaSelector = CatalogReplicaLoadBalanceSelectorFactory.createSelector( 212 replicaSelectorClass, META_TABLE_NAME, conn.getChoreService(), () -> { 213 int numOfReplicas = 1; 214 try { 215 RegionLocations metaLocations = conn.registry.getMetaRegionLocations().get( 216 conn.connConf.getReadRpcTimeoutNs(), TimeUnit.NANOSECONDS); 217 numOfReplicas = metaLocations.size(); 218 } catch (Exception e) { 219 LOG.error("Failed to get table {}'s region replication, ", META_TABLE_NAME, e); 220 } 221 return numOfReplicas; 222 }); 223 break; 224 case NONE: 225 // If user does not configure LOCATOR_META_REPLICAS_MODE, let's check the legacy config. 226 227 boolean useMetaReplicas = conn.getConfiguration().getBoolean(USE_META_REPLICAS, 228 DEFAULT_USE_META_REPLICAS); 229 if (useMetaReplicas) { 230 this.metaReplicaMode = CatalogReplicaMode.HEDGED_READ; 231 } 232 break; 233 default: 234 // Doing nothing 235 } 236 } 237 238 private TableCache getTableCache(TableName tableName) { 239 return computeIfAbsent(cache, tableName, TableCache::new); 240 } 241 242 private boolean isEqual(RegionLocations locs1, RegionLocations locs2) { 243 HRegionLocation[] locArr1 = locs1.getRegionLocations(); 244 HRegionLocation[] locArr2 = locs2.getRegionLocations(); 245 if (locArr1.length != locArr2.length) { 246 return false; 247 } 248 for (int i = 0; i < locArr1.length; i++) { 249 // do not need to compare region info 250 HRegionLocation loc1 = locArr1[i]; 251 HRegionLocation loc2 = locArr2[i]; 252 if (loc1 == null) { 253 if (loc2 != null) { 254 return false; 255 } 256 } else { 257 if (loc2 == null) { 258 return false; 259 } 260 if (loc1.getSeqNum() != loc2.getSeqNum()) { 261 return false; 262 } 263 if (!Objects.equal(loc1.getServerName(), loc2.getServerName())) { 264 return false; 265 } 266 } 267 } 268 return true; 269 } 270 271 // if we successfully add the locations to cache, return the locations, otherwise return the one 272 // which prevents us being added. The upper layer can use this value to complete pending requests. 273 private RegionLocations addToCache(TableCache tableCache, RegionLocations locs) { 274 LOG.trace("Try adding {} to cache", locs); 275 byte[] startKey = locs.getRegionLocation().getRegion().getStartKey(); 276 for (;;) { 277 RegionLocations oldLocs = tableCache.cache.putIfAbsent(startKey, locs); 278 if (oldLocs == null) { 279 return locs; 280 } 281 // check whether the regions are the same, this usually happens when table is split/merged, or 282 // deleted and recreated again. 283 RegionInfo region = locs.getRegionLocation().getRegion(); 284 RegionInfo oldRegion = oldLocs.getRegionLocation().getRegion(); 285 if (region.getEncodedName().equals(oldRegion.getEncodedName())) { 286 RegionLocations mergedLocs = oldLocs.mergeLocations(locs); 287 if (isEqual(mergedLocs, oldLocs)) { 288 // the merged one is the same with the old one, give up 289 LOG.trace("Will not add {} to cache because the old value {} " + 290 " is newer than us or has the same server name." + 291 " Maybe it is updated before we replace it", locs, oldLocs); 292 return oldLocs; 293 } 294 if (tableCache.cache.replace(startKey, oldLocs, mergedLocs)) { 295 return mergedLocs; 296 } 297 } else { 298 // the region is different, here we trust the one we fetched. This maybe wrong but finally 299 // the upper layer can detect this and trigger removal of the wrong locations 300 if (LOG.isDebugEnabled()) { 301 LOG.debug("The newnly fetch region {} is different from the old one {} for row '{}'," + 302 " try replaing the old one...", region, oldRegion, Bytes.toStringBinary(startKey)); 303 } 304 if (tableCache.cache.replace(startKey, oldLocs, locs)) { 305 return locs; 306 } 307 } 308 } 309 } 310 311 private void complete(TableName tableName, LocateRequest req, RegionLocations locs, 312 Throwable error) { 313 if (error != null) { 314 LOG.warn("Failed to locate region in '" + tableName + "', row='" + 315 Bytes.toStringBinary(req.row) + "', locateType=" + req.locateType, error); 316 } 317 Optional<LocateRequest> toSend = Optional.empty(); 318 TableCache tableCache = getTableCache(tableName); 319 if (locs != null) { 320 RegionLocations addedLocs = addToCache(tableCache, locs); 321 synchronized (tableCache) { 322 tableCache.pendingRequests.remove(req); 323 tableCache.clearCompletedRequests(addedLocs); 324 // Remove a complete locate request in a synchronized block, so the table cache must have 325 // quota to send a candidate request. 326 toSend = tableCache.getCandidate(); 327 toSend.ifPresent(r -> tableCache.send(r)); 328 } 329 toSend.ifPresent(r -> locateInMeta(tableName, r)); 330 } else { 331 // we meet an error 332 assert error != null; 333 synchronized (tableCache) { 334 tableCache.pendingRequests.remove(req); 335 // fail the request itself, no matter whether it is a DoNotRetryIOException, as we have 336 // already retried several times 337 CompletableFuture<?> future = tableCache.allRequests.remove(req); 338 if (future != null) { 339 future.completeExceptionally(error); 340 } 341 tableCache.clearCompletedRequests(null); 342 // Remove a complete locate request in a synchronized block, so the table cache must have 343 // quota to send a candidate request. 344 toSend = tableCache.getCandidate(); 345 toSend.ifPresent(r -> tableCache.send(r)); 346 } 347 toSend.ifPresent(r -> locateInMeta(tableName, r)); 348 } 349 } 350 351 // return whether we should stop the scan 352 private boolean onScanNext(TableName tableName, LocateRequest req, Result result) { 353 RegionLocations locs = MetaTableAccessor.getRegionLocations(result); 354 if (LOG.isDebugEnabled()) { 355 LOG.debug("The fetched location of '{}', row='{}', locateType={} is {}", tableName, 356 Bytes.toStringBinary(req.row), req.locateType, locs); 357 } 358 // remove HRegionLocation with null location, i.e, getServerName returns null. 359 if (locs != null) { 360 locs = locs.removeElementsWithNullLocation(); 361 } 362 363 // the default region location should always be presented when fetching from meta, otherwise 364 // let's fail the request. 365 if (locs == null || locs.getDefaultRegionLocation() == null) { 366 complete(tableName, req, null, 367 new HBaseIOException(String.format("No location found for '%s', row='%s', locateType=%s", 368 tableName, Bytes.toStringBinary(req.row), req.locateType))); 369 return true; 370 } 371 HRegionLocation loc = locs.getDefaultRegionLocation(); 372 RegionInfo info = loc.getRegion(); 373 if (info == null) { 374 complete(tableName, req, null, 375 new HBaseIOException(String.format("HRegionInfo is null for '%s', row='%s', locateType=%s", 376 tableName, Bytes.toStringBinary(req.row), req.locateType))); 377 return true; 378 } 379 if (info.isSplitParent()) { 380 return false; 381 } 382 complete(tableName, req, locs, null); 383 return true; 384 } 385 386 private void recordCacheHit() { 387 conn.getConnectionMetrics().ifPresent(MetricsConnection::incrMetaCacheHit); 388 } 389 390 private void recordCacheMiss() { 391 conn.getConnectionMetrics().ifPresent(MetricsConnection::incrMetaCacheMiss); 392 } 393 394 private RegionLocations locateRowInCache(TableCache tableCache, TableName tableName, byte[] row, 395 int replicaId) { 396 Map.Entry<byte[], RegionLocations> entry = tableCache.cache.floorEntry(row); 397 if (entry == null) { 398 recordCacheMiss(); 399 return null; 400 } 401 RegionLocations locs = entry.getValue(); 402 HRegionLocation loc = locs.getRegionLocation(replicaId); 403 if (loc == null) { 404 recordCacheMiss(); 405 return null; 406 } 407 byte[] endKey = loc.getRegion().getEndKey(); 408 if (isEmptyStopRow(endKey) || Bytes.compareTo(row, endKey) < 0) { 409 if (LOG.isTraceEnabled()) { 410 LOG.trace("Found {} in cache for {}, row='{}', locateType={}, replicaId={}", loc, tableName, 411 Bytes.toStringBinary(row), RegionLocateType.CURRENT, replicaId); 412 } 413 recordCacheHit(); 414 return locs; 415 } else { 416 recordCacheMiss(); 417 return null; 418 } 419 } 420 421 private RegionLocations locateRowBeforeInCache(TableCache tableCache, TableName tableName, 422 byte[] row, int replicaId) { 423 boolean isEmptyStopRow = isEmptyStopRow(row); 424 Map.Entry<byte[], RegionLocations> entry = 425 isEmptyStopRow ? tableCache.cache.lastEntry() : tableCache.cache.lowerEntry(row); 426 if (entry == null) { 427 recordCacheMiss(); 428 return null; 429 } 430 RegionLocations locs = entry.getValue(); 431 HRegionLocation loc = locs.getRegionLocation(replicaId); 432 if (loc == null) { 433 recordCacheMiss(); 434 return null; 435 } 436 if (isEmptyStopRow(loc.getRegion().getEndKey()) || 437 (!isEmptyStopRow && Bytes.compareTo(loc.getRegion().getEndKey(), row) >= 0)) { 438 if (LOG.isTraceEnabled()) { 439 LOG.trace("Found {} in cache for {}, row='{}', locateType={}, replicaId={}", loc, tableName, 440 Bytes.toStringBinary(row), RegionLocateType.BEFORE, replicaId); 441 } 442 recordCacheHit(); 443 return locs; 444 } else { 445 recordCacheMiss(); 446 return null; 447 } 448 } 449 450 private void locateInMeta(TableName tableName, LocateRequest req) { 451 if (LOG.isTraceEnabled()) { 452 LOG.trace("Try locate '" + tableName + "', row='" + Bytes.toStringBinary(req.row) + 453 "', locateType=" + req.locateType + " in meta"); 454 } 455 byte[] metaStartKey; 456 if (req.locateType.equals(RegionLocateType.BEFORE)) { 457 if (isEmptyStopRow(req.row)) { 458 byte[] binaryTableName = tableName.getName(); 459 metaStartKey = Arrays.copyOf(binaryTableName, binaryTableName.length + 1); 460 } else { 461 metaStartKey = createRegionName(tableName, req.row, ZEROES, false); 462 } 463 } else { 464 metaStartKey = createRegionName(tableName, req.row, NINES, false); 465 } 466 byte[] metaStopKey = 467 RegionInfo.createRegionName(tableName, HConstants.EMPTY_START_ROW, "", false); 468 Scan scan = new Scan().withStartRow(metaStartKey).withStopRow(metaStopKey, true) 469 .addFamily(HConstants.CATALOG_FAMILY).setReversed(true).setCaching(locatePrefetchLimit) 470 .setReadType(ReadType.PREAD); 471 472 switch (this.metaReplicaMode) { 473 case LOAD_BALANCE: 474 int metaReplicaId = this.metaReplicaSelector.select(tableName, req.row, req.locateType); 475 if (metaReplicaId != RegionInfo.DEFAULT_REPLICA_ID) { 476 // If the selector gives a non-primary meta replica region, then go with it. 477 // Otherwise, just go to primary in non-hedgedRead mode. 478 scan.setConsistency(Consistency.TIMELINE); 479 scan.setReplicaId(metaReplicaId); 480 } 481 break; 482 case HEDGED_READ: 483 scan.setConsistency(Consistency.TIMELINE); 484 break; 485 default: 486 // do nothing 487 } 488 489 conn.getTable(META_TABLE_NAME).scan(scan, new AdvancedScanResultConsumer() { 490 491 private boolean completeNormally = false; 492 493 private boolean tableNotFound = true; 494 495 @Override 496 public void onError(Throwable error) { 497 complete(tableName, req, null, error); 498 } 499 500 @Override 501 public void onComplete() { 502 if (tableNotFound) { 503 complete(tableName, req, null, new TableNotFoundException(tableName)); 504 } else if (!completeNormally) { 505 complete(tableName, req, null, new IOException( 506 "Unable to find region for '" + Bytes.toStringBinary(req.row) + "' in " + tableName)); 507 } 508 } 509 510 @Override 511 public void onNext(Result[] results, ScanController controller) { 512 if (results.length == 0) { 513 return; 514 } 515 tableNotFound = false; 516 int i = 0; 517 for (; i < results.length; i++) { 518 if (onScanNext(tableName, req, results[i])) { 519 completeNormally = true; 520 controller.terminate(); 521 i++; 522 break; 523 } 524 } 525 // Add the remaining results into cache 526 if (i < results.length) { 527 TableCache tableCache = getTableCache(tableName); 528 for (; i < results.length; i++) { 529 RegionLocations locs = MetaTableAccessor.getRegionLocations(results[i]); 530 if (locs == null) { 531 continue; 532 } 533 HRegionLocation loc = locs.getDefaultRegionLocation(); 534 if (loc == null) { 535 continue; 536 } 537 RegionInfo info = loc.getRegion(); 538 if (info == null || info.isOffline() || info.isSplitParent()) { 539 continue; 540 } 541 RegionLocations addedLocs = addToCache(tableCache, locs); 542 synchronized (tableCache) { 543 tableCache.clearCompletedRequests(addedLocs); 544 } 545 } 546 } 547 } 548 }); 549 } 550 551 private RegionLocations locateInCache(TableCache tableCache, TableName tableName, byte[] row, 552 int replicaId, RegionLocateType locateType) { 553 return locateType.equals(RegionLocateType.BEFORE) 554 ? locateRowBeforeInCache(tableCache, tableName, row, replicaId) 555 : locateRowInCache(tableCache, tableName, row, replicaId); 556 } 557 558 // locateToPrevious is true means we will use the start key of a region to locate the region 559 // placed before it. Used for reverse scan. See the comment of 560 // AsyncRegionLocator.getPreviousRegionLocation. 561 private CompletableFuture<RegionLocations> getRegionLocationsInternal(TableName tableName, 562 byte[] row, int replicaId, RegionLocateType locateType, boolean reload) { 563 // AFTER should be convert to CURRENT before calling this method 564 assert !locateType.equals(RegionLocateType.AFTER); 565 TableCache tableCache = getTableCache(tableName); 566 if (!reload) { 567 RegionLocations locs = locateInCache(tableCache, tableName, row, replicaId, locateType); 568 if (isGood(locs, replicaId)) { 569 return CompletableFuture.completedFuture(locs); 570 } 571 } 572 CompletableFuture<RegionLocations> future; 573 LocateRequest req; 574 boolean sendRequest = false; 575 synchronized (tableCache) { 576 // check again 577 if (!reload) { 578 RegionLocations locs = locateInCache(tableCache, tableName, row, replicaId, locateType); 579 if (isGood(locs, replicaId)) { 580 return CompletableFuture.completedFuture(locs); 581 } 582 } 583 req = new LocateRequest(row, locateType); 584 future = tableCache.allRequests.get(req); 585 if (future == null) { 586 future = new CompletableFuture<>(); 587 tableCache.allRequests.put(req, future); 588 if (tableCache.hasQuota(maxConcurrentLocateRequestPerTable) && !tableCache.isPending(req)) { 589 tableCache.send(req); 590 sendRequest = true; 591 } 592 } 593 } 594 if (sendRequest) { 595 locateInMeta(tableName, req); 596 } 597 return future; 598 } 599 600 CompletableFuture<RegionLocations> getRegionLocations(TableName tableName, byte[] row, 601 int replicaId, RegionLocateType locateType, boolean reload) { 602 // as we know the exact row after us, so we can just create the new row, and use the same 603 // algorithm to locate it. 604 if (locateType.equals(RegionLocateType.AFTER)) { 605 row = createClosestRowAfter(row); 606 locateType = RegionLocateType.CURRENT; 607 } 608 return getRegionLocationsInternal(tableName, row, replicaId, locateType, reload); 609 } 610 611 private void recordClearRegionCache() { 612 conn.getConnectionMetrics().ifPresent(MetricsConnection::incrMetaCacheNumClearRegion); 613 } 614 615 private void removeLocationFromCache(HRegionLocation loc) { 616 TableCache tableCache = cache.get(loc.getRegion().getTable()); 617 if (tableCache == null) { 618 return; 619 } 620 byte[] startKey = loc.getRegion().getStartKey(); 621 for (;;) { 622 RegionLocations oldLocs = tableCache.cache.get(startKey); 623 if (oldLocs == null) { 624 return; 625 } 626 HRegionLocation oldLoc = oldLocs.getRegionLocation(loc.getRegion().getReplicaId()); 627 if (!canUpdateOnError(loc, oldLoc)) { 628 return; 629 } 630 // Tell metaReplicaSelector that the location is stale. It will create a stale entry 631 // with timestamp internally. Next time the client looks up the same location, 632 // it will pick a different meta replica region. 633 if (this.metaReplicaMode == CatalogReplicaMode.LOAD_BALANCE) { 634 metaReplicaSelector.onError(loc); 635 } 636 637 RegionLocations newLocs = removeRegionLocation(oldLocs, loc.getRegion().getReplicaId()); 638 if (newLocs == null) { 639 if (tableCache.cache.remove(startKey, oldLocs)) { 640 recordClearRegionCache(); 641 return; 642 } 643 } else { 644 if (tableCache.cache.replace(startKey, oldLocs, newLocs)) { 645 recordClearRegionCache(); 646 return; 647 } 648 } 649 } 650 } 651 652 private void addLocationToCache(HRegionLocation loc) { 653 addToCache(getTableCache(loc.getRegion().getTable()), createRegionLocations(loc)); 654 } 655 656 private HRegionLocation getCachedLocation(HRegionLocation loc) { 657 TableCache tableCache = cache.get(loc.getRegion().getTable()); 658 if (tableCache == null) { 659 return null; 660 } 661 RegionLocations locs = tableCache.cache.get(loc.getRegion().getStartKey()); 662 return locs != null ? locs.getRegionLocation(loc.getRegion().getReplicaId()) : null; 663 } 664 665 void updateCachedLocationOnError(HRegionLocation loc, Throwable exception) { 666 Optional<MetricsConnection> connectionMetrics = conn.getConnectionMetrics(); 667 AsyncRegionLocatorHelper.updateCachedLocationOnError(loc, exception, this::getCachedLocation, 668 this::addLocationToCache, this::removeLocationFromCache, connectionMetrics.orElse(null)); 669 } 670 671 void clearCache(TableName tableName) { 672 TableCache tableCache = cache.remove(tableName); 673 if (tableCache == null) { 674 return; 675 } 676 synchronized (tableCache) { 677 if (!tableCache.allRequests.isEmpty()) { 678 IOException error = new IOException("Cache cleared"); 679 tableCache.allRequests.values().forEach(f -> f.completeExceptionally(error)); 680 } 681 } 682 conn.getConnectionMetrics() 683 .ifPresent(metrics -> metrics.incrMetaCacheNumClearRegion(tableCache.cache.size())); 684 } 685 686 void clearCache() { 687 cache.clear(); 688 } 689 690 void clearCache(ServerName serverName) { 691 for (TableCache tableCache : cache.values()) { 692 for (Map.Entry<byte[], RegionLocations> entry : tableCache.cache.entrySet()) { 693 byte[] regionName = entry.getKey(); 694 RegionLocations locs = entry.getValue(); 695 RegionLocations newLocs = locs.removeByServer(serverName); 696 if (locs == newLocs) { 697 continue; 698 } 699 if (newLocs.isEmpty()) { 700 tableCache.cache.remove(regionName, locs); 701 } else { 702 tableCache.cache.replace(regionName, locs, newLocs); 703 } 704 } 705 } 706 } 707 708 // only used for testing whether we have cached the location for a region. 709 RegionLocations getRegionLocationInCache(TableName tableName, byte[] row) { 710 TableCache tableCache = cache.get(tableName); 711 if (tableCache == null) { 712 return null; 713 } 714 return locateRowInCache(tableCache, tableName, row, RegionReplicaUtil.DEFAULT_REPLICA_ID); 715 } 716}