/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.cassandra.db;
import java.io.File;
import java.nio.ByteBuffer;
import java.util.*;
import java.util.concurrent.*;
import java.util.concurrent.atomic.AtomicLong;
import com.google.common.base.Function;
import com.google.common.base.Throwables;
import org.cliffc.high_scale_lib.NonBlockingHashSet;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.apache.cassandra.concurrent.JMXEnabledThreadPoolExecutor;
import org.apache.cassandra.concurrent.NamedThreadFactory;
import org.apache.cassandra.concurrent.StageManager;
import org.apache.cassandra.config.DatabaseDescriptor;
import org.apache.cassandra.db.columniterator.OnDiskAtomIterator;
import org.apache.cassandra.db.columniterator.SimpleAbstractColumnIterator;
import org.apache.cassandra.db.commitlog.ReplayPosition;
import org.apache.cassandra.db.filter.AbstractColumnIterator;
import org.apache.cassandra.db.filter.NamesQueryFilter;
import org.apache.cassandra.db.filter.SliceQueryFilter;
import org.apache.cassandra.db.index.SecondaryIndexManager;
import org.apache.cassandra.db.marshal.AbstractType;
import org.apache.cassandra.io.sstable.SSTableMetadata;
import org.apache.cassandra.io.sstable.SSTableReader;
import org.apache.cassandra.io.sstable.SSTableWriter;
import org.apache.cassandra.io.util.DiskAwareRunnable;
import org.apache.cassandra.utils.Allocator;
import org.github.jamm.MemoryMeter;
public class Memtable
{
private static final Logger logger = LoggerFactory.getLogger(Memtable.class);
/*
* switchMemtable puts Memtable.getSortedContents on the writer executor. When the write is complete,
* we turn the writer into an SSTableReader and add it to ssTables where it is available for reads.
*
* There are two other things that switchMemtable does.
* First, it puts the Memtable into memtablesPendingFlush, where it stays until the flush is complete
* and it's been added as an SSTableReader to ssTables_. Second, it adds an entry to commitLogUpdater
* that waits for the flush to complete, then calls onMemtableFlush. This allows multiple flushes
* to happen simultaneously on multicore systems, while still calling onMF in the correct order,
* which is necessary for replay in case of a restart since CommitLog assumes that when onMF is
* called, all data up to the given context has been persisted to SSTables.
*/
private static final ExecutorService flushWriter
= new JMXEnabledThreadPoolExecutor(DatabaseDescriptor.getFlushWriters(),
StageManager.KEEPALIVE,
TimeUnit.SECONDS,
new LinkedBlockingQueue<Runnable>(DatabaseDescriptor.getFlushQueueSize()),
new NamedThreadFactory("FlushWriter"),
"internal");
// size in memory can never be less than serialized size
private static final double MIN_SANE_LIVE_RATIO = 1.0;
// max liveratio seen w/ 1-byte columns on a 64-bit jvm was 19. If it gets higher than 64 something is probably broken.
private static final double MAX_SANE_LIVE_RATIO = 64.0;
// we want to limit the amount of concurrently running and/or queued meterings, because counting is slow (can be
// minutes, for a large memtable and a busy server). so we could keep memtables
// alive after they're flushed and would otherwise be GC'd. the approach we take is to bound the number of
// outstanding/running meterings to a maximum of one per CFS using this set; the executor's queue is unbounded but
// will implicitly be bounded by the number of CFS:s.
private static final Set<ColumnFamilyStore> meteringInProgress = new NonBlockingHashSet<ColumnFamilyStore>();
private static final ExecutorService meterExecutor = new JMXEnabledThreadPoolExecutor(1,
Integer.MAX_VALUE,
TimeUnit.MILLISECONDS,
new LinkedBlockingQueue<Runnable>(),
new NamedThreadFactory("MemoryMeter"),
"internal");
private final MemoryMeter meter;
volatile static ColumnFamilyStore activelyMeasuring;
private final AtomicLong currentSize = new AtomicLong(0);
private final AtomicLong currentOperations = new AtomicLong(0);
// We index the memtable by RowPosition only for the purpose of being able
// to select key range using Token.KeyBound. However put() ensures that we
// actually only store DecoratedKey.
private final ConcurrentNavigableMap<RowPosition, ColumnFamily> columnFamilies = new ConcurrentSkipListMap<RowPosition, ColumnFamily>();
public final ColumnFamilyStore cfs;
private final long creationTime;
private final Allocator allocator = DatabaseDescriptor.getMemtableAllocator();
// We really only need one column by allocator but one by memtable is not a big waste and avoids needing allocators to know about CFS
private final Function<IColumn, IColumn> localCopyFunction = new Function<IColumn, IColumn>()
{
public IColumn apply(IColumn c)
{
return c.localCopy(cfs, allocator);
};
};
// Record the comparator of the CFS at the creation of the memtable. This
// is only used when a user update the CF comparator, to know if the
// memtable was created with the new or old comparator.
public final AbstractType initialComparator;
public Memtable(ColumnFamilyStore cfs)
{
this.cfs = cfs;
this.creationTime = System.currentTimeMillis();
this.initialComparator = cfs.metadata.comparator;
Callable<Set<Object>> provider = new Callable<Set<Object>>()
{
public Set<Object> call() throws Exception
{
// avoid counting this once for each row
Set<Object> set = Collections.newSetFromMap(new IdentityHashMap<Object, Boolean>());
set.add(Memtable.this.cfs.metadata);
return set;
}
};
meter = new MemoryMeter().omitSharedBufferOverhead().withTrackerProvider(provider);
}
public long getLiveSize()
{
long estimatedSize = (long) (currentSize.get() * cfs.liveRatio);
// cap the estimate at both ends by what the allocator can tell us
if (estimatedSize < allocator.getMinimumSize())
return allocator.getMinimumSize();
if (estimatedSize > allocator.getMaximumSize())
return allocator.getMaximumSize();
return estimatedSize;
}
public long getSerializedSize()
{
return currentSize.get();
}
public long getOperations()
{
return currentOperations.get();
}
/**
* Should only be called by ColumnFamilyStore.apply. NOT a public API.
* (CFS handles locking to avoid submitting an op
* to a flushing memtable. Any other way is unsafe.)
*/
void put(DecoratedKey key, ColumnFamily columnFamily, SecondaryIndexManager.Updater indexer)
{
resolve(key, columnFamily, indexer);
}
public void updateLiveRatio() throws RuntimeException
{
if (!MemoryMeter.isInitialized())
{
// hack for openjdk. we log a warning about this in the startup script too.
logger.warn("MemoryMeter uninitialized (jamm not specified as java agent); assuming liveRatio of {}. "
+ " Usually this means cassandra-env.sh disabled jamm because you are using a buggy JRE; "
+ " upgrade to the Sun JRE instead", cfs.liveRatio);
return;
}
if (!meteringInProgress.add(cfs))
{
logger.debug("Metering already pending or active for {}; skipping liveRatio update", cfs);
return;
}
meterExecutor.submit(new MeteringRunnable(cfs));
}
private void resolve(DecoratedKey key, ColumnFamily cf, SecondaryIndexManager.Updater indexer)
{
ColumnFamily previous = columnFamilies.get(key);
if (previous == null)
{
// AtomicSortedColumns doesn't work for super columns (see #3821)
ColumnFamily empty = cf.cloneMeShallow(cf.isSuper() ? ThreadSafeSortedColumns.factory() : AtomicSortedColumns.factory(), false);
// We'll add the columns later. This avoids wasting works if we get beaten in the putIfAbsent
previous = columnFamilies.putIfAbsent(new DecoratedKey(key.token, allocator.clone(key.key)), empty);
if (previous == null)
previous = empty;
}
long sizeDelta = previous.addAllWithSizeDelta(cf, allocator, localCopyFunction, indexer);
currentSize.addAndGet(sizeDelta);
currentOperations.addAndGet((cf.getColumnCount() == 0)
? cf.isMarkedForDelete() ? 1 : 0
: cf.getColumnCount());
}
// for debugging
public String contents()
{
StringBuilder builder = new StringBuilder();
builder.append("{");
for (Map.Entry<RowPosition, ColumnFamily> entry : columnFamilies.entrySet())
{
builder.append(entry.getKey()).append(": ").append(entry.getValue()).append(", ");
}
builder.append("}");
return builder.toString();
}
public void flushAndSignal(final CountDownLatch latch, final Future<ReplayPosition> context)
{
flushWriter.execute(new FlushRunnable(latch, context));
}
public String toString()
{
return String.format("Memtable-%s@%s(%s/%s serialized/live bytes, %s ops)",
cfs.getColumnFamilyName(), hashCode(), currentSize, getLiveSize(), currentOperations);
}
/**
* @param startWith Include data in the result from and including this key and to the end of the memtable
* @return An iterator of entries with the data from the start key
*/
public Iterator<Map.Entry<DecoratedKey, ColumnFamily>> getEntryIterator(final RowPosition startWith, final RowPosition stopAt)
{
return new Iterator<Map.Entry<DecoratedKey, ColumnFamily>>()
{
private Iterator<Map.Entry<RowPosition, ColumnFamily>> iter = stopAt.isMinimum()
? columnFamilies.tailMap(startWith).entrySet().iterator()
: columnFamilies.subMap(startWith, true, stopAt, true).entrySet().iterator();
public boolean hasNext()
{
return iter.hasNext();
}
public Map.Entry<DecoratedKey, ColumnFamily> next()
{
Map.Entry<RowPosition, ColumnFamily> entry = iter.next();
// Actual stored key should be true DecoratedKey
assert entry.getKey() instanceof DecoratedKey;
return (Map.Entry<DecoratedKey, ColumnFamily>)(Object)entry; // yes, it's ugly
}
public void remove()
{
iter.remove();
}
};
}
public boolean isClean()
{
return columnFamilies.isEmpty();
}
/**
* obtain an iterator of columns in this memtable in the specified order starting from a given column.
*/
public static OnDiskAtomIterator getSliceIterator(final DecoratedKey key, final ColumnFamily cf, SliceQueryFilter filter)
{
assert cf != null;
final Iterator<IColumn> filteredIter = filter.reversed ? cf.reverseIterator(filter.slices) : cf.iterator(filter.slices);
return new AbstractColumnIterator()
{
public ColumnFamily getColumnFamily()
{
return cf;
}
public DecoratedKey getKey()
{
return key;
}
public boolean hasNext()
{
return filteredIter.hasNext();
}
public OnDiskAtom next()
{
return filteredIter.next();
}
};
}
public static OnDiskAtomIterator getNamesIterator(final DecoratedKey key, final ColumnFamily cf, final NamesQueryFilter filter)
{
assert cf != null;
final boolean isStandard = !cf.isSuper();
return new SimpleAbstractColumnIterator()
{
private Iterator<ByteBuffer> iter = filter.columns.iterator();
public ColumnFamily getColumnFamily()
{
return cf;
}
public DecoratedKey getKey()
{
return key;
}
protected OnDiskAtom computeNext()
{
while (iter.hasNext())
{
ByteBuffer current = iter.next();
IColumn column = cf.getColumn(current);
if (column != null)
// clone supercolumns so caller can freely removeDeleted or otherwise mutate it
return isStandard ? column : ((SuperColumn)column).cloneMe();
}
return endOfData();
}
};
}
public ColumnFamily getColumnFamily(DecoratedKey key)
{
return columnFamilies.get(key);
}
void clearUnsafe()
{
columnFamilies.clear();
}
public long creationTime()
{
return creationTime;
}
class FlushRunnable extends DiskAwareRunnable
{
private final CountDownLatch latch;
private final Future<ReplayPosition> context;
private final long estimatedSize;
FlushRunnable(CountDownLatch latch, Future<ReplayPosition> context)
{
this.latch = latch;
this.context = context;
long keySize = 0;
for (RowPosition key : columnFamilies.keySet())
{
// make sure we don't write non-sensical keys
assert key instanceof DecoratedKey;
keySize += ((DecoratedKey)key).key.remaining();
}
estimatedSize = (long) ((keySize // index entries
+ keySize // keys in data file
+ currentSize.get()) // data
* 1.2); // bloom filter and row index overhead
}
public long getExpectedWriteSize()
{
return estimatedSize;
}
protected void runWith(File sstableDirectory) throws Exception
{
assert sstableDirectory != null : "Flush task is not bound to any disk";
SSTableReader sstable = writeSortedContents(context, sstableDirectory);
cfs.replaceFlushed(Memtable.this, sstable);
latch.countDown();
}
protected Directories getDirectories()
{
return cfs.directories;
}
private SSTableReader writeSortedContents(Future<ReplayPosition> context, File sstableDirectory)
throws ExecutionException, InterruptedException
{
logger.info("Writing " + Memtable.this.toString());
SSTableReader ssTable;
// errors when creating the writer that may leave empty temp files.
SSTableWriter writer = createFlushWriter(cfs.getTempSSTablePath(sstableDirectory));
try
{
// (we can't clear out the map as-we-go to free up memory,
// since the memtable is being used for queries in the "pending flush" category)
for (Map.Entry<RowPosition, ColumnFamily> entry : columnFamilies.entrySet())
{
ColumnFamily cf = entry.getValue();
if (cf.isMarkedForDelete())
{
// When every node is up, there's no reason to write batchlog data out to sstables
// (which in turn incurs cost like compaction) since the BL write + delete cancel each other out,
// and BL data is strictly local, so we don't need to preserve tombstones for repair.
// If we have a data row + row level tombstone, then writing it is effectively an expensive no-op so we skip it.
// See CASSANDRA-4667.
if (cfs.columnFamily.equals(SystemTable.BATCHLOG_CF) && cfs.table.name.equals(Table.SYSTEM_KS) && !cf.isEmpty())
continue;
// Pedantically, you could purge column level tombstones that are past GcGRace when writing to the SSTable.
// But it can result in unexpected behaviour where deletes never make it to disk,
// as they are lost and so cannot override existing column values. So we only remove deleted columns if there
// is a CF level tombstone to ensure the delete makes it into an SSTable.
ColumnFamilyStore.removeDeletedColumnsOnly(cf, Integer.MIN_VALUE);
}
writer.append((DecoratedKey)entry.getKey(), cf);
}
if (writer.getFilePointer() > 0)
{
ssTable = writer.closeAndOpenReader();
logger.info(String.format("Completed flushing %s (%d bytes) for commitlog position %s",
ssTable.getFilename(), new File(ssTable.getFilename()).length(), context.get()));
}
else
{
writer.abort();
ssTable = null;
logger.info("Completed flushing; nothing needed to be retained. Commitlog position was {}",
context.get());
}
return ssTable;
}
catch (Throwable e)
{
writer.abort();
throw Throwables.propagate(e);
}
}
public SSTableWriter createFlushWriter(String filename) throws ExecutionException, InterruptedException
{
SSTableMetadata.Collector sstableMetadataCollector = SSTableMetadata.createCollector().replayPosition(context.get());
return new SSTableWriter(filename,
columnFamilies.size(),
cfs.metadata,
cfs.partitioner,
sstableMetadataCollector);
}
}
private static class MeteringRunnable implements Runnable
{
// we might need to wait in the meter queue for a while. measure whichever memtable is active at that point,
// rather than keeping the original memtable referenced (and thus un-freeable) until this runs.
private final ColumnFamilyStore cfs;
public MeteringRunnable(ColumnFamilyStore cfs)
{
this.cfs = cfs;
}
public void run()
{
try
{
activelyMeasuring = cfs;
Memtable memtable = cfs.getMemtableThreadSafe();
long start = System.currentTimeMillis();
// ConcurrentSkipListMap has cycles, so measureDeep will have to track a reference to EACH object it visits.
// So to reduce the memory overhead of doing a measurement, we break it up to row-at-a-time.
long deepSize = memtable.meter.measure(memtable.columnFamilies);
int objects = 0;
for (Map.Entry<RowPosition, ColumnFamily> entry : memtable.columnFamilies.entrySet())
{
deepSize += memtable.meter.measureDeep(entry.getKey()) + memtable.meter.measureDeep(entry.getValue());
objects += entry.getValue().getColumnCount();
}
double newRatio = (double) deepSize / memtable.currentSize.get();
if (newRatio < MIN_SANE_LIVE_RATIO)
{
logger.warn("setting live ratio to minimum of {} instead of {}", MIN_SANE_LIVE_RATIO, newRatio);
newRatio = MIN_SANE_LIVE_RATIO;
}
if (newRatio > MAX_SANE_LIVE_RATIO)
{
logger.warn("setting live ratio to maximum of {} instead of {}", MAX_SANE_LIVE_RATIO, newRatio);
newRatio = MAX_SANE_LIVE_RATIO;
}
// we want to be very conservative about our estimate, since the penalty for guessing low is OOM
// death. thus, higher estimates are believed immediately; lower ones are averaged w/ the old
if (newRatio > cfs.liveRatio)
cfs.liveRatio = newRatio;
else
cfs.liveRatio = (cfs.liveRatio + newRatio) / 2.0;
logger.info("{} liveRatio is {} (just-counted was {}). calculation took {}ms for {} columns",
cfs, cfs.liveRatio, newRatio, System.currentTimeMillis() - start, objects);
}
finally
{
activelyMeasuring = null;
meteringInProgress.remove(cfs);
}
}
}
}