diff --git a/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Append.java b/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Append.java index 9e3be3e..d5a4552 100644 --- a/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Append.java +++ b/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Append.java @@ -171,4 +171,9 @@ public class Append extends Mutation { public Append setACL(Map perms) { return (Append) super.setACL(perms); } + + @Override + public Append setTTL(long ttl) { + return (Append) super.setTTL(ttl); + } } diff --git a/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Delete.java b/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Delete.java index ea76d0d..d947ef8 100644 --- a/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Delete.java +++ b/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Delete.java @@ -461,4 +461,9 @@ public class Delete extends Mutation implements Comparable { public Delete setACL(Map perms) { return (Delete) super.setACL(perms); } + + @Override + public Delete setTTL(long ttl) { + throw new UnsupportedOperationException("Setting TTLs on Deletes is not supported"); + } } diff --git a/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Increment.java b/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Increment.java index b12ee02..b6e6a52 100644 --- a/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Increment.java +++ b/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Increment.java @@ -317,4 +317,9 @@ public class Increment extends Mutation implements Comparable { public Increment setACL(Map perms) { return (Increment) super.setACL(perms); } + + @Override + public Increment setTTL(long ttl) { + return (Increment) super.setTTL(ttl); + } } diff --git a/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Mutation.java b/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Mutation.java index b7f7d1b..28284e5 100644 --- a/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Mutation.java +++ b/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Mutation.java @@ -76,6 +76,11 @@ public abstract class Mutation extends OperationWithAttributes implements Row, C */ private static final String CONSUMED_CLUSTER_IDS = "_cs.id"; + /** + * The attribute for storing TTL for the result of the mutation. + */ + private static final String OP_ATTRIBUTE_TTL = "_ttl"; + protected byte [] row = null; protected long ts = HConstants.LATEST_TIMESTAMP; protected Durability durability = Durability.USE_DEFAULT; @@ -199,6 +204,12 @@ public abstract class Mutation extends OperationWithAttributes implements Row, C if (getId() != null) { map.put("id", getId()); } + // Add the TTL if set + // Long.MAX_VALUE is the default, and is interpreted to mean this attribute + // has not been set. + if (getTTL() != Long.MAX_VALUE) { + map.put("ttl", getTTL()); + } return map; } @@ -417,6 +428,29 @@ public abstract class Mutation extends OperationWithAttributes implements Row, C } /** + * Return the TTL requested for the result of the mutation, in milliseconds. + * @return the TTL requested for the result of the mutation, in milliseconds, + * or Long.MAX_VALUE if unset + */ + public long getTTL() { + byte[] ttlBytes = getAttribute(OP_ATTRIBUTE_TTL); + if (ttlBytes != null) { + return Bytes.toLong(ttlBytes); + } + return Long.MAX_VALUE; + } + + /** + * Set the TTL desired for the result of the mutation, in milliseconds. + * @param ttl the TTL desired for the result of the mutation, in milliseconds + * @return this + */ + public Mutation setTTL(long ttl) { + setAttribute(OP_ATTRIBUTE_TTL, Bytes.toBytes(ttl)); + return this; + } + + /** * Subclasses should override this method to add the heap size of their own fields. * @return the heap size to add (will be aligned). */ diff --git a/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Put.java b/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Put.java index f0d3a72..b9d652d 100644 --- a/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Put.java +++ b/hbase-client/src/main/java/org/apache/hadoop/hbase/client/Put.java @@ -460,4 +460,9 @@ public class Put extends Mutation implements HeapSize, Comparable { public Put setACL(Map perms) { return (Put) super.setACL(perms); } + + @Override + public Put setTTL(long ttl) { + return (Put) super.setTTL(ttl); + } } diff --git a/hbase-common/src/main/java/org/apache/hadoop/hbase/TagType.java b/hbase-common/src/main/java/org/apache/hadoop/hbase/TagType.java index 45c8476..9ed247e 100644 --- a/hbase-common/src/main/java/org/apache/hadoop/hbase/TagType.java +++ b/hbase-common/src/main/java/org/apache/hadoop/hbase/TagType.java @@ -28,4 +28,5 @@ public final class TagType { public static final byte VISIBILITY_TAG_TYPE = (byte) 2; public static final byte LOG_REPLAY_TAG_TYPE = (byte) 3; public static final byte VISIBILITY_EXP_SERIALIZATION_FORMAT_TAG_TYPE = (byte)4; + public static final byte TTL_TAG_TYPE = (byte)7; } diff --git a/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/GetClosestRowBeforeTracker.java b/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/GetClosestRowBeforeTracker.java index fce3b29..4d22c0e 100644 --- a/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/GetClosestRowBeforeTracker.java +++ b/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/GetClosestRowBeforeTracker.java @@ -34,7 +34,7 @@ import org.apache.hadoop.hbase.util.Bytes; /** * State and utility processing {@link HRegion#getClosestRowBefore(byte[], byte[])}. - * Like {@link ScanDeleteTracker} and {@link ScanDeleteTracker} but does not + * Like {@link ScanQueryMatcher} and {@link ScanDeleteTracker} but does not * implement the {@link DeleteTracker} interface since state spans rows (There * is no update nor reset method). */ @@ -42,7 +42,8 @@ import org.apache.hadoop.hbase.util.Bytes; class GetClosestRowBeforeTracker { private final KeyValue targetkey; // Any cell w/ a ts older than this is expired. - private final long oldestts; + private final long now; + private final long oldestUnexpiredTs; private Cell candidate = null; private final KVComparator kvcomparator; // Flag for whether we're doing getclosest on a metaregion. @@ -75,19 +76,12 @@ class GetClosestRowBeforeTracker { HConstants.DELIMITER) - this.rowoffset; } this.tablenamePlusDelimiterLength = metaregion? l + 1: -1; - this.oldestts = System.currentTimeMillis() - ttl; + this.now = System.currentTimeMillis(); + this.oldestUnexpiredTs = now - ttl; this.kvcomparator = c; this.deletes = new TreeMap>(new CellComparator.RowComparator()); } - /** - * @param kv - * @return True if this kv is expired. - */ - boolean isExpired(final Cell kv) { - return HStore.isExpired(kv, this.oldestts); - } - /* * Add the specified KeyValue to the list of deletes. * @param kv @@ -172,6 +166,15 @@ class GetClosestRowBeforeTracker { return false; } + /** + * @param cell + * @return true if the cell is expired + */ + public boolean isExpired(final Cell cell) { + return cell.getTimestamp() < this.oldestUnexpiredTs || + HStore.isCellTTLExpired(cell, this.oldestUnexpiredTs, this.now); + } + /* * Handle keys whose values hold deletes. * Add to the set of deletes and then if the candidate keys contain any that diff --git a/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/HRegion.java b/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/HRegion.java index fbf151a..8d1cb4f 100644 --- a/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/HRegion.java +++ b/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/HRegion.java @@ -84,6 +84,8 @@ import org.apache.hadoop.hbase.KeyValueUtil; import org.apache.hadoop.hbase.NotServingRegionException; import org.apache.hadoop.hbase.RegionTooBusyException; import org.apache.hadoop.hbase.TableName; +import org.apache.hadoop.hbase.Tag; +import org.apache.hadoop.hbase.TagType; import org.apache.hadoop.hbase.UnknownScannerException; import org.apache.hadoop.hbase.backup.HFileArchiver; import org.apache.hadoop.hbase.client.Append; @@ -2634,6 +2636,7 @@ public class HRegion implements HeapSize, PropagatingConfigurationObserver { // prepareDeleteTimestamps(mutation, familyMaps[i], byteNow); noOfDeletes++; } + rewriteCellTags(familyMaps[i], mutation); } lock(this.updatesLock.readLock(), numReadyToWrite); @@ -3106,6 +3109,59 @@ public class HRegion implements HeapSize, PropagatingConfigurationObserver { // } } + /** + * Possibly rewrite incoming cell tags. + */ + void rewriteCellTags(Map> familyMap, final Mutation m) { + // Check if we have any work to do and early out otherwise + // Update these checks as more logic is added here + + if (m.getTTL() == Long.MAX_VALUE) { + return; + } + + // From this point we know we have some work to do + + for (Map.Entry> e: familyMap.entrySet()) { + List cells = e.getValue(); + assert cells instanceof RandomAccess; + int listSize = cells.size(); + for (int i = 0; i < listSize; i++) { + Cell cell = cells.get(i); + List newTags = new ArrayList(); + Iterator tagIterator = CellUtil.tagsIterator(cell.getTagsArray(), + cell.getTagsOffset(), cell.getTagsLength()); + + // Carry forward existing tags + + while (tagIterator.hasNext()) { + + // Add any filters or tag specific rewrites here + + newTags.add(tagIterator.next()); + } + + // Cell TTL handling + + // Check again if we need to add a cell TTL because early out logic + // above may change when there are more tag based features in core. + if (m.getTTL() != Long.MAX_VALUE) { + // Add a cell TTL tag + newTags.add(new Tag(TagType.TTL_TAG_TYPE, Bytes.toBytes(m.getTTL()))); + } + + // Rewrite the cell with the updated set of tags + + cells.set(i, new KeyValue(cell.getRowArray(), cell.getRowOffset(), cell.getRowLength(), + cell.getFamilyArray(), cell.getFamilyOffset(), cell.getFamilyLength(), + cell.getQualifierArray(), cell.getQualifierOffset(), cell.getQualifierLength(), + cell.getTimestamp(), KeyValue.Type.codeToType(cell.getTypeByte()), + cell.getValueArray(), cell.getValueOffset(), cell.getValueLength(), + newTags)); + } + } + } + /* * Check if resources to support an update. * @@ -5172,6 +5228,9 @@ public class HRegion implements HeapSize, PropagatingConfigurationObserver { // processor.preBatchMutate(this, walEdit); // 7. Apply to memstore for (Mutation m : mutations) { + // Handle any tag based cell features + rewriteCellTags(m.getFamilyCellMap(), m); + for (CellScanner cellScanner = m.cellScanner(); cellScanner.advance();) { Cell cell = cellScanner.current(); CellUtil.setSequenceId(cell, mvccNum); @@ -5380,8 +5439,8 @@ public class HRegion implements HeapSize, PropagatingConfigurationObserver { // // Iterate the input columns and update existing values if they were // found, otherwise add new column initialized to the append value - // Avoid as much copying as possible. Every byte is copied at most - // once. + // Avoid as much copying as possible. We may need to rewrite and + // consolidate tags. Bytes are only copied once. // Would be nice if KeyValue had scatter/gather logic int idx = 0; for (Cell cell : family.getValue()) { @@ -5390,40 +5449,87 @@ public class HRegion implements HeapSize, PropagatingConfigurationObserver { // if (idx < results.size() && CellUtil.matchingQualifier(results.get(idx), cell)) { oldCell = results.get(idx); - // allocate an empty kv once + + // Process cell tags + List newTags = new ArrayList(); + + // Make a union of the set of tags in the old and new KVs + + if (oldCell.getTagsLength() > 0) { + Iterator i = CellUtil.tagsIterator(oldCell.getTagsArray(), + oldCell.getTagsOffset(), oldCell.getTagsLength()); + while (i.hasNext()) { + newTags.add(i.next()); + } + } + if (cell.getTagsLength() > 0) { + Iterator i = CellUtil.tagsIterator(cell.getTagsArray(), cell.getTagsOffset(), + cell.getTagsLength()); + while (i.hasNext()) { + newTags.add(i.next()); + } + } + + // Cell TTL handling + + if (append.getTTL() != Long.MAX_VALUE) { + // Add the new TTL tag + newTags.add(new Tag(TagType.TTL_TAG_TYPE, Bytes.toBytes(append.getTTL()))); + } + + // Rebuild tags + byte[] tagBytes = Tag.fromList(newTags); + + // allocate an empty cell once newCell = new KeyValue(row.length, cell.getFamilyLength(), cell.getQualifierLength(), now, KeyValue.Type.Put, oldCell.getValueLength() + cell.getValueLength(), - oldCell.getTagsLength() + cell.getTagsLength()); - // copy in the value - System.arraycopy(oldCell.getValueArray(), oldCell.getValueOffset(), - newCell.getValueArray(), newCell.getValueOffset(), - oldCell.getValueLength()); - System.arraycopy(cell.getValueArray(), cell.getValueOffset(), - newCell.getValueArray(), - newCell.getValueOffset() + oldCell.getValueLength(), - cell.getValueLength()); - // copy in the tags - System.arraycopy(oldCell.getTagsArray(), oldCell.getTagsOffset(), - newCell.getTagsArray(), newCell.getTagsOffset(), oldCell.getTagsLength()); - System.arraycopy(cell.getTagsArray(), cell.getTagsOffset(), newCell.getTagsArray(), - newCell.getTagsOffset() + oldCell.getTagsLength(), cell.getTagsLength()); + tagBytes.length); // copy in row, family, and qualifier System.arraycopy(cell.getRowArray(), cell.getRowOffset(), - newCell.getRowArray(), newCell.getRowOffset(), cell.getRowLength()); + newCell.getRowArray(), newCell.getRowOffset(), cell.getRowLength()); System.arraycopy(cell.getFamilyArray(), cell.getFamilyOffset(), - newCell.getFamilyArray(), newCell.getFamilyOffset(), - cell.getFamilyLength()); + newCell.getFamilyArray(), newCell.getFamilyOffset(), + cell.getFamilyLength()); System.arraycopy(cell.getQualifierArray(), cell.getQualifierOffset(), - newCell.getQualifierArray(), newCell.getQualifierOffset(), - cell.getQualifierLength()); + newCell.getQualifierArray(), newCell.getQualifierOffset(), + cell.getQualifierLength()); + // copy in the value + System.arraycopy(oldCell.getValueArray(), oldCell.getValueOffset(), + newCell.getValueArray(), newCell.getValueOffset(), + oldCell.getValueLength()); + System.arraycopy(cell.getValueArray(), cell.getValueOffset(), + newCell.getValueArray(), + newCell.getValueOffset() + oldCell.getValueLength(), + cell.getValueLength()); + // Copy in tag data + System.arraycopy(tagBytes, 0, newCell.getTagsArray(), newCell.getTagsOffset(), + tagBytes.length); idx++; } else { - // Append's KeyValue.Type==Put and ts==HConstants.LATEST_TIMESTAMP, - // so only need to update the timestamp to 'now' + // Append's KeyValue.Type==Put and ts==HConstants.LATEST_TIMESTAMP CellUtil.updateLatestStamp(cell, now); - newCell = cell; - } + + // Cell TTL handling + + if (append.getTTL() != Long.MAX_VALUE) { + List newTags = new ArrayList(1); + newTags.add(new Tag(TagType.TTL_TAG_TYPE, Bytes.toBytes(append.getTTL()))); + // Add the new TTL tag + newCell = new KeyValue(cell.getRowArray(), cell.getRowOffset(), + cell.getRowLength(), + cell.getFamilyArray(), cell.getFamilyOffset(), + cell.getFamilyLength(), + cell.getQualifierArray(), cell.getQualifierOffset(), + cell.getQualifierLength(), + cell.getTimestamp(), KeyValue.Type.codeToType(cell.getTypeByte()), + cell.getValueArray(), cell.getValueOffset(), cell.getValueLength(), + newTags); + } else { + newCell = cell; + } + } + CellUtil.setSequenceId(newCell, mvccNum); // Give coprocessors a chance to update the new cell if (coprocessorHost != null) { @@ -5598,12 +5704,22 @@ public class HRegion implements HeapSize, PropagatingConfigurationObserver { // // Iterate the input columns and update existing values if they were // found, otherwise add new column initialized to the increment amount int idx = 0; - for (Cell kv: family.getValue()) { - long amount = Bytes.toLong(CellUtil.cloneValue(kv)); + for (Cell cell: family.getValue()) { + long amount = Bytes.toLong(CellUtil.cloneValue(cell)); boolean noWriteBack = (amount == 0); + List newTags = new ArrayList(); + + // Carry forward any tags that might have been added by a coprocessor + if (cell.getTagsLength() > 0) { + Iterator i = CellUtil.tagsIterator(cell.getTagsArray(), + cell.getTagsOffset(), cell.getTagsLength()); + while (i.hasNext()) { + newTags.add(i.next()); + } + } Cell c = null; - if (idx < results.size() && CellUtil.matchingQualifier(results.get(idx), kv)) { + if (idx < results.size() && CellUtil.matchingQualifier(results.get(idx), cell)) { c = results.get(idx); if(c.getValueLength() == Bytes.SIZEOF_LONG) { amount += Bytes.toLong(c.getValueArray(), c.getValueOffset(), Bytes.SIZEOF_LONG); @@ -5612,32 +5728,36 @@ public class HRegion implements HeapSize, PropagatingConfigurationObserver { // throw new org.apache.hadoop.hbase.DoNotRetryIOException( "Attempted to increment field that isn't 64 bits wide"); } + // Carry tags forward from previous version + if (c.getTagsLength() > 0) { + Iterator i = CellUtil.tagsIterator(c.getTagsArray(), + c.getTagsOffset(), c.getTagsLength()); + while (i.hasNext()) { + newTags.add(i.next()); + } + } idx++; } // Append new incremented KeyValue to list - byte[] q = CellUtil.cloneQualifier(kv); + byte[] q = CellUtil.cloneQualifier(cell); byte[] val = Bytes.toBytes(amount); - int oldCellTagsLen = (c == null) ? 0 : c.getTagsLength(); - int incCellTagsLen = kv.getTagsLength(); - Cell newKV = new KeyValue(row.length, family.getKey().length, q.length, now, - KeyValue.Type.Put, val.length, oldCellTagsLen + incCellTagsLen); - System.arraycopy(row, 0, newKV.getRowArray(), newKV.getRowOffset(), row.length); - System.arraycopy(family.getKey(), 0, newKV.getFamilyArray(), newKV.getFamilyOffset(), - family.getKey().length); - System.arraycopy(q, 0, newKV.getQualifierArray(), newKV.getQualifierOffset(), q.length); - // copy in the value - System.arraycopy(val, 0, newKV.getValueArray(), newKV.getValueOffset(), val.length); - // copy tags - if (oldCellTagsLen > 0) { - System.arraycopy(c.getTagsArray(), c.getTagsOffset(), newKV.getTagsArray(), - newKV.getTagsOffset(), oldCellTagsLen); - } - if (incCellTagsLen > 0) { - System.arraycopy(kv.getTagsArray(), kv.getTagsOffset(), newKV.getTagsArray(), - newKV.getTagsOffset() + oldCellTagsLen, incCellTagsLen); + + // Add the TTL tag if the mutation carried one + if (increment.getTTL() != Long.MAX_VALUE) { + newTags.add(new Tag(TagType.TTL_TAG_TYPE, Bytes.toBytes(increment.getTTL()))); } + + Cell newKV = new KeyValue(row, 0, row.length, + family.getKey(), 0, family.getKey().length, + q, 0, q.length, + now, + KeyValue.Type.Put, + val, 0, val.length, + newTags); + CellUtil.setSequenceId(newKV, mvccNum); + // Give coprocessors a chance to update the new cell if (coprocessorHost != null) { newKV = coprocessorHost.postMutationBeforeWAL( diff --git a/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/HStore.java b/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/HStore.java index faf2eb1..ad5db36 100644 --- a/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/HStore.java +++ b/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/HStore.java @@ -56,6 +56,8 @@ import org.apache.hadoop.hbase.HConstants; import org.apache.hadoop.hbase.HRegionInfo; import org.apache.hadoop.hbase.KeyValue; import org.apache.hadoop.hbase.TableName; +import org.apache.hadoop.hbase.Tag; +import org.apache.hadoop.hbase.TagType; import org.apache.hadoop.hbase.classification.InterfaceAudience; import org.apache.hadoop.hbase.client.Scan; import org.apache.hadoop.hbase.conf.ConfigurationManager; @@ -1669,8 +1671,38 @@ public class HStore implements Store { return wantedVersions > maxVersions ? maxVersions: wantedVersions; } - static boolean isExpired(final Cell key, final long oldestTimestamp) { - return key.getTimestamp() < oldestTimestamp; + /** + * @param cell + * @param oldestTimestamp + * @return true if the cell is expired + */ + static boolean isCellTTLExpired(final Cell cell, final long oldestTimestamp, final long now) { + // Do not create an Iterator or Tag objects unless the cell actually has + // tags + if (cell.getTagsLength() > 0) { + // Look for a TTL tag first. Use it instead of the family setting if + // found. If a cell has multiple TTLs, resolve the conflict by using the + // first tag encountered. + Iterator i = CellUtil.tagsIterator(cell.getTagsArray(), cell.getTagsOffset(), + cell.getTagsLength()); + while (i.hasNext()) { + Tag t = i.next(); + if (TagType.TTL_TAG_TYPE == t.getType()) { + // Unlike in schema cell TTLs are stored in milliseconds, no need + // to convert + long ts = cell.getTimestamp(); + assert t.getTagLength() == Bytes.SIZEOF_LONG; + long ttl = Bytes.toLong(t.getBuffer(), t.getTagOffset(), t.getTagLength()); + if (ts + ttl < now) { + return true; + } + // Per cell TTLs cannot extend lifetime beyond family settings, so + // fall through to check that + break; + } + } + } + return false; } @Override diff --git a/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/ScanQueryMatcher.java b/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/ScanQueryMatcher.java index 964fad8..1160c7e 100644 --- a/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/ScanQueryMatcher.java +++ b/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/ScanQueryMatcher.java @@ -100,6 +100,10 @@ public class ScanQueryMatcher { */ private final long earliestPutTs; + /** The oldest timestamp we are interested in, based on TTL */ + private final long oldestUnexpiredTS; + private final long now; + /** readPoint over which the KVs are unconditionally included */ protected long maxReadPointToTrackVersions; @@ -152,7 +156,7 @@ public class ScanQueryMatcher { */ public ScanQueryMatcher(Scan scan, ScanInfo scanInfo, NavigableSet columns, ScanType scanType, long readPointToUse, long earliestPutTs, long oldestUnexpiredTS, - RegionCoprocessorHost regionCoprocessorHost) throws IOException { + long now, RegionCoprocessorHost regionCoprocessorHost) throws IOException { this.tr = scan.getTimeRange(); this.rowComparator = scanInfo.getComparator(); this.regionCoprocessorHost = regionCoprocessorHost; @@ -162,6 +166,9 @@ public class ScanQueryMatcher { scanInfo.getFamily()); this.filter = scan.getFilter(); this.earliestPutTs = earliestPutTs; + this.oldestUnexpiredTS = oldestUnexpiredTS; + this.now = now; + this.maxReadPointToTrackVersions = readPointToUse; this.timeToPurgeDeletes = scanInfo.getTimeToPurgeDeletes(); @@ -213,18 +220,18 @@ public class ScanQueryMatcher { * @param scanInfo The store's immutable scan info * @param columns * @param earliestPutTs Earliest put seen in any of the store files. - * @param oldestUnexpiredTS the oldest timestamp we are interested in, - * based on TTL + * @param oldestUnexpiredTS the oldest timestamp we are interested in, based on TTL + * @param now the current server time * @param dropDeletesFromRow The inclusive left bound of the range; can be EMPTY_START_ROW. * @param dropDeletesToRow The exclusive right bound of the range; can be EMPTY_END_ROW. * @param regionCoprocessorHost * @throws IOException */ public ScanQueryMatcher(Scan scan, ScanInfo scanInfo, NavigableSet columns, - long readPointToUse, long earliestPutTs, long oldestUnexpiredTS, byte[] dropDeletesFromRow, + long readPointToUse, long earliestPutTs, long oldestUnexpiredTS, long now, byte[] dropDeletesFromRow, byte[] dropDeletesToRow, RegionCoprocessorHost regionCoprocessorHost) throws IOException { this(scan, scanInfo, columns, ScanType.COMPACT_RETAIN_DELETES, readPointToUse, earliestPutTs, - oldestUnexpiredTS, regionCoprocessorHost); + oldestUnexpiredTS, now, regionCoprocessorHost); Preconditions.checkArgument((dropDeletesFromRow != null) && (dropDeletesToRow != null)); this.dropDeletesFromRow = dropDeletesFromRow; this.dropDeletesToRow = dropDeletesToRow; @@ -234,10 +241,10 @@ public class ScanQueryMatcher { * Constructor for tests */ ScanQueryMatcher(Scan scan, ScanInfo scanInfo, - NavigableSet columns, long oldestUnexpiredTS) throws IOException { + NavigableSet columns, long oldestUnexpiredTS, long now) throws IOException { this(scan, scanInfo, columns, ScanType.USER_SCAN, Long.MAX_VALUE, /* max Readpoint to track versions */ - HConstants.LATEST_TIMESTAMP, oldestUnexpiredTS, null); + HConstants.LATEST_TIMESTAMP, oldestUnexpiredTS, now, null); } /** @@ -295,12 +302,17 @@ public class ScanQueryMatcher { int qualifierOffset = cell.getQualifierOffset(); int qualifierLength = cell.getQualifierLength(); + long timestamp = cell.getTimestamp(); // check for early out based on timestamp alone if (columns.isDone(timestamp)) { return columns.getNextRowOrNextColumn(cell.getQualifierArray(), qualifierOffset, qualifierLength); } + // check if the cell is expired by cell TTL + if (HStore.isCellTTLExpired(cell, this.oldestUnexpiredTS, this.now)) { + return MatchCode.SKIP; + } /* * The delete logic is pretty complicated now. diff --git a/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/StoreScanner.java b/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/StoreScanner.java index 6d0098b..a998d8f 100644 --- a/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/StoreScanner.java +++ b/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/StoreScanner.java @@ -76,6 +76,7 @@ public class StoreScanner extends NonReversedNonLazyKeyValueScanner protected final Scan scan; protected final NavigableSet columns; protected final long oldestUnexpiredTS; + protected final long now; protected final int minVersions; protected final long maxRowSize; @@ -122,7 +123,8 @@ public class StoreScanner extends NonReversedNonLazyKeyValueScanner explicitColumnQuery = numCol > 0; this.scan = scan; this.columns = columns; - oldestUnexpiredTS = EnvironmentEdgeManager.currentTime() - ttl; + this.now = EnvironmentEdgeManager.currentTime(); + this.oldestUnexpiredTS = now - ttl; this.minVersions = minVersions; if (store != null && ((HStore)store).getHRegion() != null @@ -172,7 +174,7 @@ public class StoreScanner extends NonReversedNonLazyKeyValueScanner } matcher = new ScanQueryMatcher(scan, scanInfo, columns, ScanType.USER_SCAN, Long.MAX_VALUE, HConstants.LATEST_TIMESTAMP, - oldestUnexpiredTS, store.getCoprocessorHost()); + oldestUnexpiredTS, now, store.getCoprocessorHost()); this.store.addChangedReaderObserver(this); @@ -237,10 +239,10 @@ public class StoreScanner extends NonReversedNonLazyKeyValueScanner ((HStore)store).getHRegion().getReadpoint(IsolationLevel.READ_COMMITTED)); if (dropDeletesFromRow == null) { matcher = new ScanQueryMatcher(scan, scanInfo, null, scanType, smallestReadPoint, - earliestPutTs, oldestUnexpiredTS, store.getCoprocessorHost()); + earliestPutTs, oldestUnexpiredTS, now, store.getCoprocessorHost()); } else { matcher = new ScanQueryMatcher(scan, scanInfo, null, smallestReadPoint, earliestPutTs, - oldestUnexpiredTS, dropDeletesFromRow, dropDeletesToRow, store.getCoprocessorHost()); + oldestUnexpiredTS, now, dropDeletesFromRow, dropDeletesToRow, store.getCoprocessorHost()); } // Filter the list of scanners using Bloom filters, time range, TTL, etc. @@ -280,7 +282,7 @@ public class StoreScanner extends NonReversedNonLazyKeyValueScanner this(null, scan.getCacheBlocks(), scan, columns, scanInfo.getTtl(), scanInfo.getMinVersions(), readPt); this.matcher = new ScanQueryMatcher(scan, scanInfo, columns, scanType, - Long.MAX_VALUE, earliestPutTs, oldestUnexpiredTS, null); + Long.MAX_VALUE, earliestPutTs, oldestUnexpiredTS, now, null); // In unit tests, the store could be null if (this.store != null) { diff --git a/hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestHRegion.java b/hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestHRegion.java index b4e94bf..a46e4d0 100644 --- a/hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestHRegion.java +++ b/hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestHRegion.java @@ -78,6 +78,7 @@ import org.apache.hadoop.hbase.HBaseTestingUtility; import org.apache.hadoop.hbase.HColumnDescriptor; import org.apache.hadoop.hbase.HConstants; import org.apache.hadoop.hbase.RegionTooBusyException; +import org.apache.hadoop.hbase.Tag; import org.apache.hadoop.hbase.HConstants.OperationStatusCode; import org.apache.hadoop.hbase.HDFSBlocksDistribution; import org.apache.hadoop.hbase.HRegionInfo; @@ -90,6 +91,7 @@ import org.apache.hadoop.hbase.MultithreadedTestUtil.TestThread; import org.apache.hadoop.hbase.NotServingRegionException; import org.apache.hadoop.hbase.ServerName; import org.apache.hadoop.hbase.TableName; +import org.apache.hadoop.hbase.TagType; import org.apache.hadoop.hbase.Waiter; import org.apache.hadoop.hbase.client.Append; import org.apache.hadoop.hbase.client.Delete; @@ -111,6 +113,7 @@ import org.apache.hadoop.hbase.filter.NullComparator; import org.apache.hadoop.hbase.filter.PrefixFilter; import org.apache.hadoop.hbase.filter.SingleColumnValueExcludeFilter; import org.apache.hadoop.hbase.filter.SingleColumnValueFilter; +import org.apache.hadoop.hbase.io.hfile.HFile; import org.apache.hadoop.hbase.monitoring.MonitoredRPCHandler; import org.apache.hadoop.hbase.monitoring.MonitoredTask; import org.apache.hadoop.hbase.monitoring.TaskMonitor; @@ -136,6 +139,7 @@ import org.apache.hadoop.hbase.test.MetricsAssertHelper; import org.apache.hadoop.hbase.testclassification.MediumTests; import org.apache.hadoop.hbase.testclassification.VerySlowRegionServerTests; import org.apache.hadoop.hbase.util.Bytes; +import org.apache.hadoop.hbase.util.EnvironmentEdgeManager; import org.apache.hadoop.hbase.util.EnvironmentEdgeManagerTestHelper; import org.apache.hadoop.hbase.util.FSUtils; import org.apache.hadoop.hbase.util.IncrementingEnvironmentEdge; @@ -5672,6 +5676,133 @@ public class TestHRegion { } } + @Test + public void testCellTTLs() throws IOException { + IncrementingEnvironmentEdge edge = new IncrementingEnvironmentEdge(); + EnvironmentEdgeManager.injectEdge(edge); + + final byte[] row = Bytes.toBytes("testRow"); + final byte[] q1 = Bytes.toBytes("q1"); + final byte[] q2 = Bytes.toBytes("q2"); + final byte[] q3 = Bytes.toBytes("q3"); + final byte[] q4 = Bytes.toBytes("q4"); + + HTableDescriptor htd = new HTableDescriptor(TableName.valueOf("testCellTTLs")); + HColumnDescriptor hcd = new HColumnDescriptor(fam1); + hcd.setTimeToLive(10); // 10 seconds + htd.addFamily(hcd); + + Configuration conf = new Configuration(TEST_UTIL.getConfiguration()); + conf.setInt(HFile.FORMAT_VERSION_KEY, HFile.MIN_FORMAT_VERSION_WITH_TAGS); + + HRegion region = HRegion.createHRegion(new HRegionInfo(htd.getTableName(), + HConstants.EMPTY_BYTE_ARRAY, HConstants.EMPTY_BYTE_ARRAY), + TEST_UTIL.getDataTestDir(), conf, htd); + assertNotNull(region); + try { + long now = EnvironmentEdgeManager.currentTime(); + // Add a cell that will expire in 5 seconds via cell TTL + region.put(new Put(row).add(new KeyValue(row, fam1, q1, now, + HConstants.EMPTY_BYTE_ARRAY, new Tag[] { + // TTL tags specify ts in milliseconds + new Tag(TagType.TTL_TAG_TYPE, Bytes.toBytes(5000L)) } ))); + // Add a cell that will expire after 10 seconds via family setting + region.put(new Put(row).add(fam1, q2, now, HConstants.EMPTY_BYTE_ARRAY)); + // Add a cell that will expire in 15 seconds via cell TTL + region.put(new Put(row).add(new KeyValue(row, fam1, q3, now + 10000 - 1, + HConstants.EMPTY_BYTE_ARRAY, new Tag[] { + // TTL tags specify ts in milliseconds + new Tag(TagType.TTL_TAG_TYPE, Bytes.toBytes(5000L)) } ))); + // Add a cell that will expire in 20 seconds via family setting + region.put(new Put(row).add(fam1, q4, now + 10000 - 1, HConstants.EMPTY_BYTE_ARRAY)); + + // Flush so we are sure store scanning gets this right + region.flushcache(); + + // A query at time T+0 should return all cells + Result r = region.get(new Get(row)); + assertNotNull(r.getValue(fam1, q1)); + assertNotNull(r.getValue(fam1, q2)); + assertNotNull(r.getValue(fam1, q3)); + assertNotNull(r.getValue(fam1, q4)); + + // Increment time to T+5 seconds + edge.incrementTime(5000); + + r = region.get(new Get(row)); + assertNull(r.getValue(fam1, q1)); + assertNotNull(r.getValue(fam1, q2)); + assertNotNull(r.getValue(fam1, q3)); + assertNotNull(r.getValue(fam1, q4)); + + // Increment time to T+10 seconds + edge.incrementTime(5000); + + r = region.get(new Get(row)); + assertNull(r.getValue(fam1, q1)); + assertNull(r.getValue(fam1, q2)); + assertNotNull(r.getValue(fam1, q3)); + assertNotNull(r.getValue(fam1, q4)); + + // Increment time to T+15 seconds + edge.incrementTime(5000); + + r = region.get(new Get(row)); + assertNull(r.getValue(fam1, q1)); + assertNull(r.getValue(fam1, q2)); + assertNull(r.getValue(fam1, q3)); + assertNotNull(r.getValue(fam1, q4)); + + // Increment time to T+20 seconds + edge.incrementTime(10000); + + r = region.get(new Get(row)); + assertNull(r.getValue(fam1, q1)); + assertNull(r.getValue(fam1, q2)); + assertNull(r.getValue(fam1, q3)); + assertNull(r.getValue(fam1, q4)); + + // Fun with disappearing increments + + // Start at 1 + region.put(new Put(row).add(fam1, q1, Bytes.toBytes(1L))); + r = region.get(new Get(row)); + byte[] val = r.getValue(fam1, q1); + assertNotNull(val); + assertEquals(Bytes.toLong(val), 1L); + + // Increment with a TTL of 5 seconds + Increment incr = new Increment(row).addColumn(fam1, q1, 1L); + incr.setTTL(5000); + region.increment(incr); // 2 + + // New value should be 2 + r = region.get(new Get(row)); + val = r.getValue(fam1, q1); + assertNotNull(val); + assertEquals(Bytes.toLong(val), 2L); + + // Increment time to T+25 seconds + edge.incrementTime(5000); + + // Value should be back to 1 + r = region.get(new Get(row)); + val = r.getValue(fam1, q1); + assertNotNull(val); + assertEquals(Bytes.toLong(val), 1L); + + // Increment time to T+30 seconds + edge.incrementTime(5000); + + // Original value written at T+20 should be gone now via family TTL + r = region.get(new Get(row)); + assertNull(r.getValue(fam1, q1)); + + } finally { + HRegion.closeHRegion(region); + } + } + private static HRegion initHRegion(byte[] tableName, String callingMethod, byte[]... families) throws IOException { return initHRegion(tableName, callingMethod, HBaseConfiguration.create(), diff --git a/hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestQueryMatcher.java b/hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestQueryMatcher.java index 2b1f11f..a2b7383 100644 --- a/hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestQueryMatcher.java +++ b/hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestQueryMatcher.java @@ -91,11 +91,12 @@ public class TestQueryMatcher extends HBaseTestCase { } - private void _testMatch_ExplicitColumns(Scan scan, List expected) throws IOException { - // 2,4,5 + private void _testMatch_ExplicitColumns(Scan scan, List expected) throws IOException { + long now = EnvironmentEdgeManager.currentTime(); + // 2,4,5 ScanQueryMatcher qm = new ScanQueryMatcher(scan, new ScanInfo(fam2, 0, 1, ttl, false, 0, rowComparator), get.getFamilyMap().get(fam2), - EnvironmentEdgeManager.currentTime() - ttl); + now - ttl, now); List memstore = new ArrayList(); memstore.add(new KeyValue(row1, fam2, col1, 1, data)); @@ -175,9 +176,10 @@ public class TestQueryMatcher extends HBaseTestCase { expected.add(ScanQueryMatcher.MatchCode.INCLUDE); expected.add(ScanQueryMatcher.MatchCode.DONE); + long now = EnvironmentEdgeManager.currentTime(); ScanQueryMatcher qm = new ScanQueryMatcher(scan, new ScanInfo(fam2, 0, 1, ttl, false, 0, rowComparator), null, - EnvironmentEdgeManager.currentTime() - ttl); + now - ttl, now); List memstore = new ArrayList(); memstore.add(new KeyValue(row1, fam2, col1, 1, data)); @@ -231,7 +233,7 @@ public class TestQueryMatcher extends HBaseTestCase { long now = EnvironmentEdgeManager.currentTime(); ScanQueryMatcher qm = new ScanQueryMatcher(scan, new ScanInfo(fam2, 0, 1, testTTL, false, 0, rowComparator), get.getFamilyMap().get(fam2), - now - testTTL); + now - testTTL, now); KeyValue [] kvs = new KeyValue[] { new KeyValue(row1, fam2, col1, now-100, data), @@ -285,7 +287,7 @@ public class TestQueryMatcher extends HBaseTestCase { long now = EnvironmentEdgeManager.currentTime(); ScanQueryMatcher qm = new ScanQueryMatcher(scan, new ScanInfo(fam2, 0, 1, testTTL, false, 0, rowComparator), null, - now - testTTL); + now - testTTL, now); KeyValue [] kvs = new KeyValue[] { new KeyValue(row1, fam2, col1, now-100, data), @@ -343,7 +345,7 @@ public class TestQueryMatcher extends HBaseTestCase { NavigableSet cols = get.getFamilyMap().get(fam2); ScanQueryMatcher qm = new ScanQueryMatcher(scan, scanInfo, cols, Long.MAX_VALUE, - HConstants.OLDEST_TIMESTAMP, HConstants.OLDEST_TIMESTAMP, from, to, null); + HConstants.OLDEST_TIMESTAMP, HConstants.OLDEST_TIMESTAMP, now, from, to, null); List actual = new ArrayList(rows.length); byte[] prevRow = null; diff --git a/hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestTags.java b/hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestTags.java index a613319..17fe754 100644 --- a/hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestTags.java +++ b/hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestTags.java @@ -455,8 +455,13 @@ public class TestTags { tags = TestCoprocessorForTags.tags; assertEquals(5L, Bytes.toLong(kv.getValueArray(), kv.getValueOffset(), kv.getValueLength())); assertEquals(2, tags.size()); - assertEquals("tag1", Bytes.toString(tags.get(0).getValue())); - assertEquals("tag2", Bytes.toString(tags.get(1).getValue())); + // We cannot assume the ordering of tags + List tagValues = new ArrayList(); + for (Tag tag: tags) { + tagValues.add(Bytes.toString(tag.getValue())); + } + assertTrue(tagValues.contains("tag1")); + assertTrue(tagValues.contains("tag2")); TestCoprocessorForTags.checkTagPresence = false; TestCoprocessorForTags.tags = null; @@ -512,8 +517,13 @@ public class TestTags { kv = KeyValueUtil.ensureKeyValue(result.getColumnLatestCell(f, q)); tags = TestCoprocessorForTags.tags; assertEquals(2, tags.size()); - assertEquals("tag1", Bytes.toString(tags.get(0).getValue())); - assertEquals("tag2", Bytes.toString(tags.get(1).getValue())); + // We cannot assume the ordering of tags + tagValues.clear(); + for (Tag tag: tags) { + tagValues.add(Bytes.toString(tag.getValue())); + } + assertTrue(tagValues.contains("tag1")); + assertTrue(tagValues.contains("tag2")); TestCoprocessorForTags.checkTagPresence = false; TestCoprocessorForTags.tags = null; diff --git a/hbase-shell/src/main/ruby/hbase/table.rb b/hbase-shell/src/main/ruby/hbase/table.rb index f02ba81..07a850d 100644 --- a/hbase-shell/src/main/ruby/hbase/table.rb +++ b/hbase-shell/src/main/ruby/hbase/table.rb @@ -136,17 +136,19 @@ EOF set_attributes(p, attributes) if attributes visibility = args[VISIBILITY] set_cell_visibility(p, visibility) if visibility + ttl = args[TTL] + set_op_ttl(p, ttl) if ttl end #Case where attributes are specified without timestamp if timestamp.kind_of?(Hash) timestamp.each do |k, v| - if v.kind_of?(Hash) - set_attributes(p, v) if v - end - if v.kind_of?(String) - set_cell_visibility(p, v) if v - end - + if k == 'ATTRIBUTES' + set_attributes(p, v) + elsif k == 'VISIBILITY' + set_cell_visibility(p, v) + elsif k == "TTL" + set_op_ttl(p, v) + end end timestamp = nil end @@ -214,6 +216,8 @@ EOF visibility = args[VISIBILITY] set_attributes(incr, attributes) if attributes set_cell_visibility(incr, visibility) if visibility + ttl = args[TTL] + set_op_ttl(incr, ttl) if ttl end incr.addColumn(family, qualifier, value) @table.increment(incr) @@ -232,6 +236,8 @@ EOF visibility = args[VISIBILITY] set_attributes(append, attributes) if attributes set_cell_visibility(append, visibility) if visibility + ttl = args[TTL] + set_op_ttl(append, ttl) if ttl end append.add(family, qualifier, value.to_s.to_java_bytes) @table.append(append) @@ -540,6 +546,10 @@ EOF auths.to_java(:string))) end + def set_op_ttl(op, ttl) + op.setTTL(ttl.to_java(:long)) + end + #---------------------------- # Add general administration utilities to the shell # each of the names below adds this method name to the table diff --git a/hbase-shell/src/test/ruby/hbase/table_test.rb b/hbase-shell/src/test/ruby/hbase/table_test.rb index 7272229..fa2990d 100644 --- a/hbase-shell/src/test/ruby/hbase/table_test.rb +++ b/hbase-shell/src/test/ruby/hbase/table_test.rb @@ -530,5 +530,18 @@ module Hbase end end + define_test "mutation with TTL should expire" do + @test_table.put('ttlTest', 'x:a', 'foo', { TTL => 1000 } ) + begin + res = @test_table._get_internal('ttlTest', 'x:a') + assert_not_nil(res) + sleep 2 + res = @test_table._get_internal('ttlTest', 'x:a') + assert_nil(res) + ensure + @test_table.delete('ttlTest', 'x:a') + end + end + end end