K
- the type of keys in hash containers, created by this builderC
- the container type, created by this builder, i. e. ChronicleMap
or ChronicleSet
B
- the concrete builder type, i. e. ChronicleMapBuilder
or ChronicleSetBuilder
public interface ChronicleHashBuilder<K,C extends ChronicleHash,B extends ChronicleHashBuilder<K,C,B>> extends Cloneable
ChronicleMapBuilder
and
ChronicleSetBuilder
, i. e. Chronicle hash container configurations.
ChronicleHashBuilder
is mutable. Configuration methods mutate the builder and return
the builder itself back to support chaining pattern, rather than the builder copies with
the corresponding configuration changed. To make an independent configuration, clone() the builder.
Modifier and Type | Method and Description |
---|---|
B |
actualEntriesPerSegment(long actualEntriesPerSegment) |
B |
actualSegments(int actualSegments) |
B |
bytesMarshallerFactory(BytesMarshallerFactory bytesMarshallerFactory)
Configures a
BytesMarshallerFactory to be used with BytesMarshallableSerializer , which is a default ObjectSerializer ,
to serialize/deserialize data to/from off-heap memory in hash containers, created by this
builder. |
B |
clone()
Clones this builder.
|
B |
constantKeySizeBySample(K sampleKey)
Configures the constant number of bytes, taken by serialized form of keys, put into hash
containers, created by this builder.
|
C |
create()
Creates a new hash container, storing it's data in off-heap memory, not mapped to any file.
|
C |
createPersistedTo(File file)
Opens a hash container residing the specified file, or creates a new one if the file not yet
exists and maps its off-heap memory to the file.
|
B |
entries(long entries)
Configures the maximum number of "entry size chunks", which
could be taken by the maximum number of entries, inserted into the hash containers, created
by this builder.
|
B |
entrySize(int entrySize)
Configures the size in bytes of allocation unit of hash container instances, created by this
builder.
|
B |
errorListener(ChronicleHashErrorListener errorListener) |
B |
immutableKeys()
Specifies that key objects, queried with the hash containers, created by this builder, are
inherently immutable.
|
ChronicleHashInstanceConfig<C> |
instance() |
B |
keyDeserializationFactory(ObjectFactory<K> keyDeserializationFactory)
Configures factory which is used to create a new key instance, if key class is either
Byteable , BytesMarshallable or Externalizable subclass in maps,
created by this builder. |
B |
keyMarshaller(BytesMarshaller<K> keyMarshaller)
Configures the
BytesMarshaller used to serialize/deserialize keys to/from off-heap
memory in hash containers, created by this builder. |
B |
keyMarshallers(BytesWriter<K> keyWriter,
BytesReader<K> keyReader)
Configures the marshallers, used to serialize/deserialize keys to/from off-heap
memory in hash containers, created by this builder.
|
B |
keySize(int keySize)
Configures the optimal number of bytes, taken by serialized form of keys, put into hash
containers, created by this builder.
|
B |
keySizeMarshaller(SizeMarshaller keySizeMarshaller)
Configures the marshaller used to serialize actual key sizes to off-heap memory
in hash containers, created by this builder.
|
B |
lockTimeOut(long lockTimeOut,
TimeUnit unit)
Configures timeout of locking on segments of hash
containers, created by this builder, when performing any queries, as well as bulk operations
like iteration.
|
B |
metaDataBytes(int metaDataBytes) |
B |
minSegments(int minSegments)
Set minimum number of segments in hash containers, constructed by this builder.
|
B |
objectSerializer(ObjectSerializer objectSerializer)
Configures the serializer used to serialize/deserialize data to/from off-heap memory, when
specified class doesn't implement a specific serialization interface like
Externalizable or BytesMarshallable (for example, if data is loosely typed and just
Object is specified as the data class), or nullable data, and if custom marshaller is
not configured, in hash containers, created by
this builder. |
B |
replication(byte identifier) |
B |
replication(byte identifier,
TcpTransportAndNetworkConfig tcpTransportAndNetwork)
Shortcut for
replication(SimpleReplication.builder()
.tcpTransportAndNetwork(tcpTransportAndNetwork).createWithId(identifier)) . |
B |
replication(SingleChronicleHashReplication replication)
Configures replication of the hash containers, created by this builder.
|
StatelessClientConfig<C> |
statelessClient(InetSocketAddress remoteAddress) |
B |
timeProvider(TimeProvider timeProvider)
Configures a time provider, used by hash containers, created by this builder, for needs of
replication consensus protocol (conflicting data updates resolution).
|
B clone()
ChronicleHashBuilder
s are mutable and changed on each configuration method call. Original
and cloned builders are independent.B minSegments(int minSegments)
ConcurrentHashMap
.minSegments
- the minimum number of segments in containers, constructed by this builderB keySize(int keySize)
constantKeySizeBySample(Object)
method instead of this one.
If key size varies moderately, specify the size higher than average, but lower than the maximum possible, to minimize average memory overuse. If key size varies in a wide range, it's better to use entry size in "chunk" mode and configure it directly.
keySize
- number of bytes, taken by serialized form of keysconstantKeySizeBySample(Object)
,
entrySize(int)
B constantKeySizeBySample(K sampleKey)
sampleKey
, all
keys should take the same number of bytes in serialized form, as this sample object.
If keys are of boxed primitive type or Byteable
subclass, i. e. if key size is
known statically, it is automatically accounted and this method shouldn't be called.
If key size varies, method keySize(int)
or entrySize(int)
should be
called instead of this one.
sampleKey
- the sample keykeySize(int)
B entrySize(int entrySize)
ChronicleMap
and ChronicleSet
store their data off-heap, so it is required
to serialize key (and values, in ChronicleMap
case) (unless they are direct Byteable
instances). Serialized key bytes (+ serialized value bytes, in ChronicleMap
case) + some metadata bytes comprise "entry space", which ChronicleMap
or ChronicleSet
should allocate. So entry size is a minimum allocation portion in the
hash containers, created by this builder. E. g. if entry size is 100, the created container
could only allocate 100, 200, 300... bytes for an entry. If say 150 bytes of entry space are
required by the entry, 200 bytes will be allocated, 150 used and 50 wasted. To minimize
memory overuse and improve speed, you should pay decent attention to this configuration.
There are three major patterns of this configuration usage:
ChronicleMap
case) sizes are constant. Configure them via constantKeySizeBySample(Object)
and ChronicleMapBuilder.constantValueSizeBySample(Object)
methods, and you will experience no memory waste at all.ChronicleMap
case) varies moderately. Specify them using corresponding methods, or
specify entry size directly by calling this method, by sizes somewhere between average and
maximum possible. The idea is to have most (90% or more) entries to fit a single "entry size"
with moderate memory waste (10-20% on average), rest 10% or less of entries should take 2
"entry sizes", thus with ~50% memory overuse.ChronicleMap
case) varies in a wide range. Then it's best to use entry size configuration in
chunk mode. Specify entry size so that most entries should take from 5 to several
dozens of "chunks". With this approach, average memory waste should be very low.
However, remember that
IllegalArgumentException
is thrown
on attempt to insert too large entry, compared to the configured or computed entry size.Example: if values in your ChronicleMap
are adjacency lists of some social graph,
where nodes are represented as long
ids, and adjacency lists are serialized in
efficient manner, for example as long[]
arrays. Typical number of connections is
100-300, maximum is 3000. In this case entry size of
50 * (8 bytes for each id) = 400 bytes would be a good choice:
Map<Long, long[]> socialGraph = ChronicleMapBuilder
.of(Long.class, long[].class)
// given that graph should have of 1 billion nodes, and 150 average adjacency list size
// => values takes 3 chuncks on average
.entries(1_000_000_000L * (150 / 50))
.entrySize(50 * 8)
.create();
It is minimum possible (because 3000 friends / 50 friends = 60 is close to 64 "max chunks by
single entry" limit, and ensures moderate average memory overuse (not more than 20%). entrySize
- the "chunk size" in bytesentries(long)
B entries(long entries)
IllegalStateException
might be
thrown, because currently ChronicleMap
and ChronicleSet
doesn't support
resizing.
ChronicleMap
case) is constant, this number
is equal to the maximum number of entries (because each entry takes exactly one "entry size"
memory unit).ChronicleMap
case) size varies
moderately, you should pass to this method the maximum number of entries + 5-25%, depending
on your data properties and configured key/value/entry sizes.entrySize(int)
method.You shouldn't put additional margin over the number, computed according the rules above.
This bad practice was popularized by HashMap.HashMap(int)
and HashSet.HashSet(int)
constructors, which accept "capacity", that should be multiplied by
"load factor" to obtain actual maximum expected number of entries. ChronicleMap
and
ChronicleSet
don't have a notion of load factor.
Default value is 2^20 (~ 1 million).
entries
- maximum size of the created maps, in memory allocation units, so-called "entry
size"entrySize(int)
B actualEntriesPerSegment(long actualEntriesPerSegment)
B actualSegments(int actualSegments)
B lockTimeOut(long lockTimeOut, TimeUnit unit)
ChronicleHashErrorListener.onLockTimeout(long)
is
called, and then thread tries to obtain the segment lock one more time, and so in a loop,
until thread is interrupted. However, you can configure error listener to throw an exception on the first (or n-th) lock
acquisition fail.
Default lock time out is 2 seconds.
lockTimeOut
- new lock timeout for segments of containers created by this builder, in
the given time unitsunit
- time unit of the given lock timeoutB errorListener(ChronicleHashErrorListener errorListener)
B metaDataBytes(int metaDataBytes)
B timeProvider(TimeProvider timeProvider)
Default time provider is TimeProvider.SYSTEM
.
timeProvider
- a new time provider for replication needsreplication(SingleChronicleHashReplication)
B bytesMarshallerFactory(BytesMarshallerFactory bytesMarshallerFactory)
BytesMarshallerFactory
to be used with BytesMarshallableSerializer
, which is a default ObjectSerializer
,
to serialize/deserialize data to/from off-heap memory in hash containers, created by this
builder.
Default BytesMarshallerFactory
is an instance of VanillaBytesMarshallerFactory
. This is a convenience configuration method, it has no effect
on the resulting hash containers, if custom data
marshallers are configured, data types extends one of specific serialization interfaces,
recognized by this builder (e. g. Externalizable
or BytesMarshallable
), or
ObjectSerializer
is configured.
bytesMarshallerFactory
- the marshaller factory to be used with the default ObjectSerializer
, i. e. BytesMarshallableSerializer
objectSerializer(ObjectSerializer)
B objectSerializer(ObjectSerializer objectSerializer)
Externalizable
or BytesMarshallable
(for example, if data is loosely typed and just
Object
is specified as the data class), or nullable data, and if custom marshaller is
not configured, in hash containers, created by
this builder. Please read ObjectSerializer
docs for more info and available options.
Default serializer is BytesMarshallableSerializer
, configured with the specified
or default BytesMarshallerFactory
.
objectSerializer
- the serializer used to serialize loosely typed or nullable data if
custom marshaller is not configuredbytesMarshallerFactory(BytesMarshallerFactory)
,
keyMarshaller(BytesMarshaller)
B keyMarshaller(@NotNull BytesMarshaller<K> keyMarshaller)
BytesMarshaller
used to serialize/deserialize keys to/from off-heap
memory in hash containers, created by this builder. See the
section about serialization in ChronicleMap manual for more information.keyMarshaller
- the marshaller used to serialize keyskeyMarshallers(BytesWriter, BytesReader)
,
objectSerializer(ObjectSerializer)
B keyMarshallers(@NotNull BytesWriter<K> keyWriter, @NotNull BytesReader<K> keyReader)
Configuring marshalling this way results to a little bit more compact in-memory layout
of the map, comparing to a single interface configuration:
keyMarshaller(BytesMarshaller)
.
Passing BytesInterop
(which is a subinterface of BytesWriter
) as the
first argument is supported, and even more advantageous from performance perspective.
keyWriter
- the new key object → Bytes
writer (interop) strategykeyReader
- the new Bytes
→ key object reader strategykeyMarshaller(BytesMarshaller)
B keySizeMarshaller(@NotNull SizeMarshaller keySizeMarshaller)
Default key size marshaller is so-called stop bit encoding marshalling. If constant key
size is configured, or defaulted if the key type is always constant and ChronicleHashBuilder
implementation knows about it, this configuration takes no effect,
because a special SizeMarshaller
implementation, which doesn't actually do any
marshalling, and just returns the known constant size on SizeMarshaller.readSize(
Bytes)
calls, is used instead of any SizeMarshaller
configured using this method.
keySizeMarshaller
- the new marshaller, used to serialize actual key sizes to off-heap
memoryB keyDeserializationFactory(@NotNull ObjectFactory<K> keyDeserializationFactory)
Byteable
, BytesMarshallable
or Externalizable
subclass in maps,
created by this builder. If custom key
marshaller is configured, this configuration is unused, because it is incapsulated in
BytesMarshaller.read(Bytes)
method (without provided instance to read the data into),
i. e. it's is the user-side responsibility.
Default key deserialization factory is NewInstanceObjectFactory
, which creates
a new key instance using Class.newInstance()
default constructor. You could provide
an AllocateInstanceObjectFactory
, which uses Unsafe.allocateInstance(Class)
(you might want to do this for better performance or if you don't want to initialize fields),
or a factory which calls a key class constructor with some arguments, or a factory which
internally delegates to instance pool or ThreadLocal
, to reduce allocations.
keyDeserializationFactory
- the key factory used to produce instances to deserialize
data inIllegalStateException
- if custom key marshaller is specified or key class is not
either Byteable
, BytesMarshallable
or Externalizable
B immutableKeys()
ChronicleMap
or ChronicleSet
are not required
to be immutable, as in ordinary Map
or Set
implementations, because they are
serialized off-heap. However, ChronicleMap
and ChronicleSet
implementations
can benefit from the knowledge that keys are not mutated between queries.
By default, ChronicleHashBuilder
s detects immutability automatically only for very
few standard JDK types (for example, for String
), it is not recommended to rely on
ChronicleHashBuilder
to be smart enough about this.
B replication(SingleChronicleHashReplication replication)
By default, hash containers, created by this builder doesn't replicate their data.
This method call overrides all previous replication configurations of this builder, made
either by this method or replication(byte, TcpTransportAndNetworkConfig)
shortcut method.
replication
- the replication configChronicleHashInstanceConfig.replicated(SingleChronicleHashReplication)
,
replication(byte, TcpTransportAndNetworkConfig)
B replication(byte identifier, TcpTransportAndNetworkConfig tcpTransportAndNetwork)
replication(SimpleReplication.builder()
.tcpTransportAndNetwork(tcpTransportAndNetwork).createWithId(identifier))
.identifier
- the network-wide identifier of the containers, created by this buildertcpTransportAndNetwork
- configuration of tcp connection and networkreplication(SingleChronicleHashReplication)
,
ChronicleHashInstanceConfig.replicated(byte, TcpTransportAndNetworkConfig)
B replication(byte identifier)
StatelessClientConfig<C> statelessClient(InetSocketAddress remoteAddress)
ChronicleHashInstanceConfig<C> instance()
C create() throws IOException
ChronicleHash.close()
called on the returned container, or after the container
object is collected during GC, or on JVM shutdown the off-heap memory used by the returned
container is freed.
This method is a shortcut for instance().create()
.
IOException
- if any IO error relates to off-heap memory allocation,
or establishing replication connections, occurscreatePersistedTo(File)
,
instance()
C createPersistedTo(File file) throws IOException
Multiple containers could give access to the same data simultaneously, either inside a single JVM or across processes. Access is synchronized correctly across all instances, i. e. hash container mapping the data from the first JVM isn't able to modify the data, concurrently accessed from the second JVM by another hash container instance, mapping the same data.
On container's close()
the data isn't removed, it remains on
disk and available to be opened again (given the same file name) or during different JVM
run.
This method is shortcut for instance().persistedTo(file).create()
.
file
- the file with existing hash container or a desired location of a new off-heap
persisted hash containerIOException
- if any IO error, related to off-heap memory allocation or file mapping,
or establishing replication connections, occursChronicleHash.file()
,
ChronicleHash.close()
,
create()
,
ChronicleHashInstanceConfig.persistedTo(File)
Copyright © 2014. All rights reserved.