Chronicle Map

Community support: Issues, Chronicle mailing list, Stackoverflow, Chronicle User's group

3 min to understand everything about Chronicle Map

Chronicle Map is an in-memory key-value store designed for low-latency and/or multi-process applications. Notably trading, financial market applications.

Features

Ultra low latency: Chronicle Map targets median latency of both read and write queries of less than 1 microsecond in certain tests.
High concurrency: write queries scale well up to the number of hardware execution threads in the server. Read queries never block each other.
(Optional) persistence to disk
Multi-key queries
(Optional, closed-source) eventually-consistent, fully-redundant, asynchronous replication across servers, "last write wins" strategy by default, allows to implement custom state-based CRDT strategy.

Unique features

Multiple processes could access a Chronicle Map concurrently. At the same time, the data store is in-process for each of the accessing processes. (Out-of-process approach to IPC is simply incompatible with Chronicle Map's median latency target of < 1 μs.)
Replication without logs, with constant footprint cost, guarantees progress even if the network doesn't sustain write rates.

Chronicle Map has two meanings: the language-agnostic data store and the implementation of this data store for the JVM. Currently, this is the only implementation.

From Java perspective, ChronicleMap is a ConcurrentMap implementation which stores the entries off-heap, serializing/deserializing key and value objects to/from off-heap memory transparently. Chronicle Map supports

Key and value objects caching/reusing for making zero allocations (garbage) on queries.
Flyweight values for eliminating serialization/deserialization cost and allowing direct read/write access to off-heap memory.

Primary Chronicle Map use cases

Replacing slower key-value stores, like Redis and Memcached, when used within a single server.
Replacing similar JVM-centric solutions, like Coherence and Hazelcast, for speed and/or certain Chronicle Map features those solutions lack.
Moving parts of the application state out of the Java heap for either
- Reducing the heap size, for reducing GC pressure, or fitting 32 GB for using Compressed Oops
- Inter-process communication
- Persistence
- Replication across servers
Drop-in ConcurrentHashMap replacement, Chronicle Map performs better in some cases.

What guarantees does Chronicle Map provide in ACID terms?

Atomicity - single-key queries are atomic if Chronicle Map is properly configured, multi-key queries are not atomic.
Consistency - doesn't make sense for key-value stores
Isolation - yes (for both single- and multi-key queries).
Durability - no, at most, Chronicle Map could be persisted to disk. Durability with Chronicle Map is provided by another level of architecture, for example all requests are sent to several nodes - master and hot standby. Clustering/distributed architecture is out of the scope of the Chronicle Map project, there are projects on top of Chronicle Map which address these questions, e. g. Chronicle Enterprise.

Project status: ready for production.

What is the Chronicle Map's data structure? In one sentence and simplified, a Chronicle Map data store is a big chunk of shared memory (optionally mapped to disk), split into independent segments, each segment has an independent memory allocation for storing the entries, a hash table for search, and a lock in shared memory (implemented via CAS loops) for managing concurrent access. Read the Chronicle Map data store design overview for more.

Chronicle Map is not

A document store. No secondary indexes.
A multimap. Using a ChronicleMap<K, Collection<V>> as multimap is technically possible, but often leads to problems (see a StackOverflow answer for details). Developing a proper multimap with Chronicle Map's design principles is possible, contact us if you would consider sponsoring such development.

Chronicle Map doesn't support

Range queries, iteration over the entries in alphabetical order. Keys in Chronicle Map are not sorted.
LRU entry eviction

Features

Feature	Availability
In-memory off-heap Map	Open-source
Persistence to disk	Open-source
Remote calls	Closed-source, on-demand
Eventually-consistent replication (100% redundancy)
Partially-redundant replication
Synchronous replication
Entry expiration timeouts

License

Chronicle Map is distributed under LGPLv3. If you want to obtain this software under more permissive license, please contact the copyright holders.

Peer projects

Chronicle Engine - reactive processing framework supporting Chronicle Map as a backend.
Chronicle Enterprise - extended version of Chronicle Engine.
Chronicle Journal - another key-value built by Chronicle Software, with different properties.

Difference between Chronicle Map 2 and 3

Functional changes in Chronicle Map 3:

Chronicle Map 3 has formal data store specification, that verbalizes the guarantees which the data store provides and gives a way to verify those guarantees.
Added support for multi-key queries.
"Listeners" mechanism fully reworked, see the Behaviour Customization section. This has a number of important consequences, most notable is:
- Possibility to define replication eventual-consistency strategy, different from "last write wins", e. g. any state-based CRDT.
"Stateless clients" functionality (i. e. remote calls) is moved to Chronicle Engine.
Replication is done via Chronicle Engine.
Chronicle Map 2 has hard creation-time limit on the number of entries storable in a Chronicle Map. If the size exceeds this limit, an exception is thrown. In Chronicle Map 3, this limitation is removed, though the number of entries still has to be configured on the Chronicle Map creation, exceeding this configured limit is possible, but discouraged. See the Number of entries configuration section.
Chronicle Map 3 supports entry checksums that allows to detect data corruption, hence brings additional safety.
Chronicle Map 3 allows to recover after failures and corruptions.
A number of smaller improvements and fixes.

Non-functional changes:

Chronicle Map 3 requires Java version 8 or newer, while Chronicle Map 2 supports Java 7.
Chronicle Map 3 has versioning policy and compatibility policy. Chronicle Map 2 doesn't have such documents.

If you use Chronicle Map 2, you might be looking for Chronicle Map 2 Tutorial or Chronicle Map 2 Javadoc.

Download the library

Maven Artifact Download

<dependency>
  <groupId>net.openhft</groupId>
  <artifactId>chronicle-map</artifactId>
  <version><!--replace with the latest version--></version>
</dependency>

Click here to get the Latest Version Number

Maven Snapshot Download

If you want to try out the latest pre-release code, you can download the snapshot artifact manually from https://oss.sonatype.org/content/repositories/snapshots/net/openhft/chronicle-map/. A better way is to add the following to your setting.xml, to allow maven to download snapshots:

<repository>
    <id>Snapshot Repository</id>
    <name>Snapshot Repository</name>
    <url>https://oss.sonatype.org/content/repositories/snapshots</url>
    <snapshots>
        <enabled>true</enabled>
    </snapshots>
</repository>

and define the snapshot version in your pom.xml, for example:

<dependency>
  <groupId>net.openhft</groupId>
  <artifactId>chronicle-map</artifactId>
  <version><!--replace with the latest snapshot version--></version>
</dependency>

Create a `ChronicleMap` Instance

Creating an instance of ChronicleMap is a little more complex than just calling a constructor. To create an instance you have to use the ChronicleMapBuilder.

In-memory Chronicle Map

import net.openhft.chronicle.map.*
.....

interface PostalCodeRange {
    int minCode();
    void minCode(int minCode);

    int maxCode();
    void maxCode(int maxCode);
}

ChronicleMapBuilder<CharSequence, PostalCodeRange> cityPostalCodesMapBuilder =
    ChronicleMapBuilder.of(CharSequence.class, PostalCodeRange.class)
        .averageKey("Amsterdam")
        .entries(50_000);
ChronicleMap<CharSequence, PostalCodeRange> cityPostalCodes =
    cityPostalCodesMapBuilder.create();

// Or shorter form, without builder variable extraction:

ChronicleMap<Integer, PostalCodeRange> cityPostalCodes = ChronicleMap
    .of(CharSequence.class, PostalCodeRange.class)
    .averageKey("Amsterdam")
    .entries(50_000)
    .create();

This snippet creates an in-memory Chronicle Map store, supposed to store about 50 000 city name -> postal code mappings. It is accessible within a single JVM process - the process it is created within. The data is accessible while the process is alive, when the process is terminated, the data is vanished.

Persisted Chronicle Map

Replace .create() calls with .createPersistedTo(cityPostalCodesFile), if you want the Chronicle Map to either

Outlive the process it was created within, e. g. to support hot application redeploy
Be accessible from multiple concurrent processes on the same server
Persist the data to disk

The cityPostalCodesFile has to represent the same location on your server among all Java processes, wishing to access this Chronicle Map instance, e. g. System.getProperty("java.io.tmpdir") + "/cityPostalCodes.dat".

The name and location of the file is entirely up to you.

Note than when you create a ChronicleMap instance with .createPersistedTo(file), and the given file already exists in the system, you open a view to the existing Chronicle Map data store from this JVM process rather than creating a new Chronicle Map data store. I. e. it could already contain some entries. No special action with the data is performed during such operation. If you want to clean up corrupted entries and ensure that the data store is in correct state, see Recovery section.

"ChronicleMap instance" vs "Chronicle Map data store"

In this tutorial, ChronicleMap instance (or simply ChronicleMap) term is used to refer to on-heap object, providing access to a Chronicle Map data store (or Chronicle Map key-value store, or Chronicle Map store, or simply Chronicle Map, with space between two words in contrast to ChronicleMap), which could be purely in-memory, or persisted to disk. Currently Java implementation doesn't allow to create multiple accessor ChronicleMap objects to a single in-memory Chronicle Map store, i. e. there is always a one-to-one correspondence, that may lead to confusion. Persisted Chronicle Map store, however, allows to create multiple accessor ChronicleMap instances either within a single JVM process (though it is not recommended to do that), or from concurrent JVM processes, that is perfectly OK.

When no processes access the file, it could be freely moved to another location in the system, and even to another server, even running different operating system, opened from another location and you will observe the same data.

If you don't need the Chronicle Map instance to survive the server restart, i. e. you don't need persistence to disk, only multi-process access, choose the file to be mounted on tmpfs, e. g. on Linux it is as easy as placing you file in /dev/shm directory.

You must configure .entries(entries) -- the supposed ChronicleMap size. Try to configure the entries so that the created Chronicle Map is going to serve about 99% requests being less or equal than this number of entries in size.

You shouldn't put additional margin over the actual target number of entries. This bad practice was popularized by new HashMap(capacity) and new HashSet(capacity) constructors, which accept capacity, that should be multiplied by load factor to obtain the actual maximum expected number of entries in the container. ChronicleMap and ChronicleSet don't have a notion of load factor.

See ChronicleMapBuilder#entries() Javadocs for more.

Once ChronicleMap instance is created, it's configurations are sealed and couldn't be changed though the ChronicleMapBuilder instance.

Single ChronicleMap instance per JVM. If you want to access a Chronicle Map data store concurrently within a Java process, you should not create a separate ChronicleMap instance per thread. Within the JVM environment, ChronicleMap instance is a ConcurrentMap, and could be accessed concurrently the same way as e. g. ConcurrentHashMap.

Recovery

If a process, accessing a persisted Chronicle Map, terminated abnormally: crashed, SIGKILLed, or terminated because the host operating system crashed, or the machine lost power, the Chronicle Map might remain in an inaccessible or corrupted state. When the Chronicle Map is opened next time from another process, it should be done via .recoverPersistedTo() method in ChronicleMapBuilder. Unlike createPersistedTo(), this method scans all memory of Chronicle Map store for inconsistencies, if some found, it cleans them up.

.recoverPersistedTo() needs to access the Chronicle Map exclusively. If a concurrent process is accessing the Chronicle Map while another process is attempting to perform recovery, result of operations on the accessing process side, and results of recovery are unspecified. The data could be corrupted further. You must ensure no other process is accessing the Chronicle Map store when calling for .recoverPersistedTo() on this store.

Example:

ChronicleMap<Integer, PostalCodeRange> cityPostalCodes = ChronicleMap
    .of(CharSequence.class, PostalCodeRange.class)
    .averageKey("Amsterdam")
    .entries(50_000)
    .recoverPersistedTo(cityPostalCodesFile, true);
// or
ChronicleMap<Integer, PostalCodeRange> cityPostalCodes = ChronicleMap
    .of(CharSequence.class, PostalCodeRange.class)
    // assuming ChronicleMapBuilder configurations at the moment of
    // cityPostalCodes Chronicle Map creation are not known
    .recoverPersistedTo(cityPostalCodesFile, false);

The second parameter in recoverPersistedTo() method is called sameBuilderConfig, it should be true if ChronicleMapBuilder is configured in exactly the same way, as when the Chronicle Map was created, and using the same version of Chronicle Map library, or false, if initial configurations are not known, of current version of Chronicle Map library differs from the version, used to create this Chronicle Map initially.

If sameBuilderConfig is true, recoverPersistedTo() checks that the recovered Chronicle Map's header memory (containing serialized configurations) is not corrupted, because it "knows" all the right configurations and what should be written to the header. If the header is corrupted, it recovers it.

If sameBuilderConfig is false, recoverPersistedTo() relies on the configurations written to the Chronicle Map's header, assuming it is not corrupted. If it is corrupted, at best it will lead to a runtime exception thrown from recoverPersistedTo() without touching the data, stored in the Chronicle Map (but this effectively means inability to recover the Chronicle Map), at worst it will proceed "successfully" but will corrupt the Chronicle Map further.

However, the subject header memory is never updated on ordinary operations with Chronicle Map, so it couldn't be corrupted if an accessing process crashed, or the operating system crashed, or even the machine lost power. Only hardware memory or disk corruption or a bug in the file system could lead to Chronicle Map header memory corruption.

.recoverPersistedTo() is harmless if the previous process accessing the Chronicle Map terminated normally, however this is a computationally expensive procedure that should generally be avoided.

To share configuration, Chronicle Map creation and recovery code you could use .createOrRecoverPersistedTo(persistenceFile) method in ChronicleMapBuilder, which is equivalent to createPersistedTo(persistenceFile) call, if the persistence file doesn't yet exist, and to recoverPersistedTo(persistenceFile, true), if the file already exists, e. g.:

ChronicleMap<Integer, PostalCodeRange> cityPostalCodes = ChronicleMap
    .of(CharSequence.class, PostalCodeRange.class)
    .averageKey("Amsterdam")
    .entries(50_000)
    .createOrRecoverPersistedTo(cityPostalCodesFile);

If the Chronicle Map is configured to store entry checksums along with entries, recovery procedure checks for each entry that the checksums is correct, otherwise it assumes the entry is corrupted and deletes it from the Chronicle Map. If checksums are to stored, recovery procedure cannot guarantee correctness of entry data. See Entry checksums section for more information.

Key and Value Types

Either key or value type of ChronicleMap<K, V> could be:

Types with best possible out of the box support:
- Any value interface
- Any class implementing Byteable interface from Chronicle Bytes
- Any class implementing BytesMarshallable interface from Chronicle Bytes. The implementation class should have a public no-arg constructor.
- byte[] and ByteBuffer
- CharSequence, String and StringBuilder. Note that these char sequence types are serialized using UTF-8 encoding by default. If you need a different encoding, refer to the example in the custom CharSequence encoding section.
- Integer, Long and Double
Types supported out of the box, but not particularly efficiently. You might want to implement more efficient custom serializers for them:
- Any class implementing java.io.Externalizable. The implementation class should have a public no-arg constructor.
- Any type implementing java.io.Serializable, including boxed primitive types (except listed above) and array types
Any other type, if custom serializers are provided.

Prefer value interfaces. They don't generate garbage and have close to zero serialization/deserialization costs. Prefer them even to boxed primitives, for example, try to use net.openhft.chronicle.core.values.IntValue instead of Integer.

Generally, you must hint the ChronicleMapBuilder with the average sizes of the keys and values, which are going to be inserted into the ChronicleMap. This is needed to allocate the proper volume of the shared memory. Do this via averageKey() (preferred) or averageKeySize() and averageValue() or averageValueSize() respectively.

See the example above: averageKey("Amsterdam") is called, because it is assumed that "Amsterdam" (9 bytes in UTF-8 encoding) is the average length for city names, some names are shorter (Tokyo, 5 bytes), some names are longer (San Francisco, 13 bytes).

Another example: if values in your ChronicleMap are adjacency lists of some social graph, where nodes are represented as long ids, and adjacency lists are long[] arrays. The average number of friends is 150. Configure the ChronicleMap as follows:

Map<Long, long[]> socialGraph = ChronicleMap
    .of(Long.class, long[].class)
    .entries(1_000_000_000L)
    .averageValue(new long[150])
    .create();

You could omit specifying key or value average sizes, if their types are boxed Java primitives or value interfaces. They are constantly-sized and Chronicle Map knows about that.

If the key or value type is constantly sized, or keys or values only of a certain size appear in your Chronicle Map domain, you should prefer to configure constantKeySizeBySample() or constantValueSizeBySample(), instead of averageKey() or averageValue(), for example:

ChronicleSet<UUID> uuids =
    ChronicleSet.of(UUID.class)
        // All UUIDs take 16 bytes.
        .constantKeySizeBySample(UUID.randomUUID())
        .entries(1_000_000)
        .create();

Custom serializers

Chronicle Map allows to configure custom marshallers for key or value types which are not supported out of the box, or serialize supported types like String in some custom way (i. e. in encoding, different from UTF-8), or serialize supported types more efficiently than it is done by default.

There are three pairs of serialization interfaces, only one of them should be implemented and provided to the ChronicleMapBuilder for the key or value type:

`BytesWriter` and `BytesReader`

This pair of interfaces is configured via ChronicleMapBuilder.keyMarshallers() or valueMarshallers() for the key or value type of the map, respectively.

This pair of interfaces is the most suitable, if the size of the serialized form is not known in advance, i. e. the easiest way to compute the size of the serialized form of some object of the type, is performing serialization itself and looking at the number of written bytes.

This pair of interfaces is the least efficient and the simplest to implement, so it should also be used when efficiency is not the top priority, or when gains of using other pairs of interfaces (which are more complicated to implement) are marginal.

Basically you should implement two serialization methods:

void write(Bytes out, @NotNull T toWrite) from BytesWriter interface, which writes the given toWrite instance of the serialized type to the given out bytes sink.
T read(Bytes in, @Nullable T using) from BytesReader interface, which reads the serialized object into the given using instance (if the serialized type is reusable, the using object is not null, and suitable for reusing for this particular serialized object), or a newly created instance. The returned object contains the serialized data; it may be identical or not identical to the passed using instance.

For example, here is the implementation of BytesWriter and BytesReader for CharSequence[] value type (array of CharSequences):

public final class CharSequenceArrayBytesMarshaller
        implements BytesWriter<CharSequence[]>, BytesReader<CharSequence[]>,
        ReadResolvable<CharSequenceArrayBytesMarshaller> {

    static final CharSequenceArrayBytesMarshaller INSTANCE = new CharSequenceArrayBytesMarshaller();

    private CharSequenceArrayBytesMarshaller() {}

    @Override
    public void write(Bytes out, @NotNull CharSequence[] toWrite) {
        out.writeInt(toWrite.length);
        for (CharSequence cs : toWrite) {
            // Assume elements non-null for simplicity
            Objects.requireNonNull(cs);
            out.writeUtf8(cs);
        }
    }

    @NotNull
    @Override
    public CharSequence[] read(Bytes in, @Nullable CharSequence[] using) {
        int len = in.readInt();
        if (using == null)
            using = new CharSequence[len];
        if (using.length != len)
            using = Arrays.copyOf(using, len);
        for (int i = 0; i < len; i++) {
            CharSequence cs = using[i];
            if (cs instanceof StringBuilder) {
                in.readUtf8((StringBuilder) cs);
            } else {
                StringBuilder sb = new StringBuilder(0);
                in.readUtf8(sb);
                using[i] = sb;
            }
        }
        return using;
    }

    @Override
    public void writeMarshallable(@NotNull WireOut wireOut) {
        // no fields to write
    }

    @Override
    public void readMarshallable(@NotNull WireIn wireIn) {
        // no fields to read
    }

    @Override
    public CharSequenceArrayBytesMarshaller readResolve() {
        return INSTANCE;
    }
}

Usage example:

try (ChronicleMap<String, CharSequence[]> map = ChronicleMap
        .of(String.class, CharSequence[].class)
        .averageKey("fruits")
        .valueMarshaller(CharSequenceArrayBytesMarshaller.INSTANCE)
        .averageValue(new CharSequence[]{"banana", "pineapple"})
        .entries(2)
        .create()) {
    map.put("fruits", new CharSequence[]{"banana", "pineapple"});
    map.put("vegetables", new CharSequence[] {"carrot", "potato"});
    Assert.assertEquals(2, map.get("fruits").length);
    Assert.assertEquals(2, map.get("vegetables").length);
}

The total size of serialization form for some CharSequence[] array is 4 bytes for storing the array length, plus the sum of sizes of all CharSequences, in UTF-8 encoding. Computing this size without actual encoding has comparable computational cost with performing actual encoding, that makes CharSequence[] type to meet the second criteria (see above) which makes BytesWriter and BytesReader the most suitable pair of serialization interfaces to implement for the type.

Note how read() implementation attempts to reuse not only the array object, but also the elements, minimizing the amount of produced garbage. This is a recommended practice.

Some additional notes:

If the reader or writer interface implementation is not configurable and doesn't have per-instance cache or state fields, i. e. it doesn't have instance fields at all, there is a convention to make such implementation classes final, give them a private constructor and expose a single INSTANCE constant - a sole instance of this implementation in the JVM.
- Don't make marshaller class enum, because there are some issues with enum serialization/ deserialization.
- For such no-state implementations, don't forget to implement ReadResolvable interface and return INSTANCE, otherwise you have no guarantee that INSTANCE constant is the only alive instance of this implementation in the JVM.
If both writer and reader interface implementations have no fields, it might be a good idea to merge them into a single type, in order to keep writing and reading logic together.

Custom `CharSequence` encoding

Another example shows how to serialize CharSequences using custom encoding (rather than UTF-8):

Writer:

public final class CharSequenceCustomEncodingBytesWriter
        implements BytesWriter<CharSequence>,
        StatefulCopyable<CharSequenceCustomEncodingBytesWriter> {

    // config fields, non-final because read in readMarshallable()
    private Charset charset;
    private int inputBufferSize;

    // cache fields
    private transient CharsetEncoder charsetEncoder;
    private transient CharBuffer inputBuffer;
    private transient ByteBuffer outputBuffer;

    public CharSequenceCustomEncodingBytesWriter(Charset charset, int inputBufferSize) {
        this.charset = charset;
        this.inputBufferSize = inputBufferSize;
        initTransients();
    }

    private void initTransients() {
        charsetEncoder = charset.newEncoder();
        inputBuffer = CharBuffer.allocate(inputBufferSize);
        int outputBufferSize = (int) (inputBufferSize * charsetEncoder.averageBytesPerChar());
        outputBuffer = ByteBuffer.allocate(outputBufferSize);
    }

    @Override
    public void write(Bytes out, @NotNull CharSequence cs) {
        // Write the actual cs length for accurate StringBuilder.ensureCapacity() while reading
        out.writeStopBit(cs.length());
        long encodedSizePos = out.writePosition();
        out.writeSkip(4);
        charsetEncoder.reset();
        inputBuffer.clear();
        outputBuffer.clear();
        int csPos = 0;
        boolean endOfInput = false;
        // this loop inspired by the CharsetEncoder.encode(CharBuffer) implementation
        while (true) {
            if (!endOfInput) {
                int nextCsPos = Math.min(csPos + inputBuffer.remaining(), cs.length());
                append(inputBuffer, cs, csPos, nextCsPos);
                inputBuffer.flip();
                endOfInput = nextCsPos == cs.length();
                csPos = nextCsPos;
            }

            CoderResult cr = inputBuffer.hasRemaining() ?
                    charsetEncoder.encode(inputBuffer, outputBuffer, endOfInput) :
                    CoderResult.UNDERFLOW;

            if (cr.isUnderflow() && endOfInput)
                cr = charsetEncoder.flush(outputBuffer);

            if (cr.isUnderflow()) {
                if (endOfInput) {
                    break;
                } else {
                    inputBuffer.compact();
                    continue;
                }
            }

            if (cr.isOverflow()) {
                outputBuffer.flip();
                writeOutputBuffer(out);
                outputBuffer.clear();
                continue;
            }

            try {
                cr.throwException();
            } catch (CharacterCodingException e) {
                throw new IORuntimeException(e);
            }
        }
        outputBuffer.flip();
        writeOutputBuffer(out);

        out.writeInt(encodedSizePos, (int) (out.writePosition() - encodedSizePos - 4));
    }

    private void writeOutputBuffer(Bytes out) {
        int remaining = outputBuffer.remaining();
        out.write(out.writePosition(), outputBuffer, 0, remaining);
        out.writeSkip(remaining);
    }

    /**
     * Need this method because {@link CharBuffer#append(CharSequence, int, int)} produces garbage
     */
    private static void append(CharBuffer charBuffer, CharSequence cs, int start, int end) {
        for (int i = start; i < end; i++) {
            charBuffer.put(cs.charAt(i));
        }
    }

    @Override
    public void readMarshallable(@NotNull WireIn wireIn) {
        charset = (Charset) wireIn.read(() -> "charset").object();
        inputBufferSize = wireIn.read(() -> "inputBufferSize").int32();
        initTransients();
    }

    @Override
    public void writeMarshallable(@NotNull WireOut wireOut) {
        wireOut.write(() -> "charset").object(charset);
        wireOut.write(() -> "inputBufferSize").int32(inputBufferSize);
    }

    @Override
    public CharSequenceCustomEncodingBytesWriter copy() {
        return new CharSequenceCustomEncodingBytesWriter(charset, inputBufferSize);
    }
}

Reader:

public final class CharSequenceCustomEncodingBytesReader
        implements BytesReader<CharSequence>,
        StatefulCopyable<CharSequenceCustomEncodingBytesReader> {

    // config fields, non-final because read in readMarshallable()
    private Charset charset;
    private int inputBufferSize;

    // cache fields
    private transient CharsetDecoder charsetDecoder;
    private transient ByteBuffer inputBuffer;
    private transient CharBuffer outputBuffer;

    public CharSequenceCustomEncodingBytesReader(Charset charset, int inputBufferSize) {
        this.charset = charset;
        this.inputBufferSize = inputBufferSize;
        initTransients();
    }

    private void initTransients() {
        charsetDecoder = charset.newDecoder();
        inputBuffer = ByteBuffer.allocate(inputBufferSize);
        int outputBufferSize = (int) (inputBufferSize * charsetDecoder.averageCharsPerByte());
        outputBuffer = CharBuffer.allocate(outputBufferSize);
    }

    @NotNull
    @Override
    public CharSequence read(Bytes in, @Nullable CharSequence using) {
        long csLengthAsLong = in.readStopBit();
        if (csLengthAsLong > Integer.MAX_VALUE) {
            throw new IORuntimeException("cs len shouldn't be more than " + Integer.MAX_VALUE +
                    ", " + csLengthAsLong + " read");
        }
        int csLength = (int) csLengthAsLong;
        StringBuilder sb;
        if (using instanceof StringBuilder) {
            sb = (StringBuilder) using;
            sb.setLength(0);
            sb.ensureCapacity(csLength);
        } else {
            sb = new StringBuilder(csLength);
        }

        int remainingBytes = in.readInt();
        charsetDecoder.reset();
        inputBuffer.clear();
        outputBuffer.clear();
        boolean endOfInput = false;
        // this loop inspired by the CharsetDecoder.decode(ByteBuffer) implementation
        while (true) {
            if (!endOfInput) {
                int inputChunkSize = Math.min(inputBuffer.remaining(), remainingBytes);
                inputBuffer.limit(inputBuffer.position() + inputChunkSize);
                in.read(inputBuffer);
                inputBuffer.flip();
                remainingBytes -= inputChunkSize;
                endOfInput = remainingBytes == 0;
            }

            CoderResult cr = inputBuffer.hasRemaining() ?
                    charsetDecoder.decode(inputBuffer, outputBuffer, endOfInput) :
                    CoderResult.UNDERFLOW;

            if (cr.isUnderflow() && endOfInput)
                cr = charsetDecoder.flush(outputBuffer);

            if (cr.isUnderflow()) {
                if (endOfInput) {
                    break;
                } else {
                    inputBuffer.compact();
                    continue;
                }
            }

            if (cr.isOverflow()) {
                outputBuffer.flip();
                sb.append(outputBuffer);
                outputBuffer.clear();
                continue;
            }

            try {
                cr.throwException();
            } catch (CharacterCodingException e) {
                throw new IORuntimeException(e);
            }
        }
        outputBuffer.flip();
        sb.append(outputBuffer);

        return sb;
    }

    @Override
    public void readMarshallable(@NotNull WireIn wireIn) throws IORuntimeException {
        charset = (Charset) wireIn.read(() -> "charset").object();
        inputBufferSize = wireIn.read(() -> "inputBufferSize").int32();
        initTransients();
    }

    @Override
    public void writeMarshallable(@NotNull WireOut wireOut) {
        wireOut.write(() -> "charset").object(charset);
        wireOut.write(() -> "inputBufferSize").int32(inputBufferSize);
    }

    @Override
    public CharSequenceCustomEncodingBytesReader copy() {
        return new CharSequenceCustomEncodingBytesReader(charset, inputBufferSize);
    }
}

Usage example:

Charset charset = Charset.forName("GBK");
int charBufferSize = 100;
int bytesBufferSize = 200;
CharSequenceCustomEncodingBytesWriter writer =
        new CharSequenceCustomEncodingBytesWriter(charset, charBufferSize);
CharSequenceCustomEncodingBytesReader reader =
        new CharSequenceCustomEncodingBytesReader(charset, bytesBufferSize);
try (ChronicleMap<String, CharSequence> englishToChinese = ChronicleMap
        .of(String.class, CharSequence.class)
        .valueMarshallers(reader, writer)
        .averageKey("hello")
        .averageValue("你好")
        .entries(10)
        .create()) {
    englishToChinese.put("hello", "你好");
    englishToChinese.put("bye", "再见");

    Assert.assertEquals("你好", englishToChinese.get("hello").toString());
    Assert.assertEquals("再见", englishToChinese.get("bye").toString());
}

Some notes on this case of custom serialization:

Both CharSequenceCustomEncodingBytesWriter and CharSequenceCustomEncodingBytesReader have configurations (charset and input buffer size), hence they are implemented as normal classes rather than classes with private constructors and a single INSTANCE.
Both writer and reader classes have some "cache" fields, their contents are mutated during writing and reading. That is why they have to implement StatefulCopyable interface. See Understanding StatefulCopyable section for more infromation on this.

`SizedWriter` and `SizedReader`

This pair of interfaces is configured via ChronicleMapBuilder.keyMarshallers() or valueMarshallers() for the key or value type of the map, respectively (overloaded methods, those for BytesWriter and BytesReader interfaces have the same names).

The main two methods to implement in SizedWriter interface:

long size(@NotNull T toWrite); returns the number of bytes, which is written by the subsequent write() method given the same toWrite instance of the serialized type.
void write(Bytes out, long size, @NotNull T toWrite); writes the given toWrite instance to the given out bytes sink. Additionally size is provided, which is the value computed for the same toWrite instance by calling size() method. Hence, when SizedWriter is used internally, size() method is always called before write(), caching logic in SizedWriter implementation may rely on this. This method should advance writePosition() of the given out Bytes exactly by the given size number of bytes.

SizedReader's T read(Bytes in, long size, @Nullable T using); method is similar to the corresponding read() method in BytesReader interface, except the size of the serialized object is provided.

This pair of interfaces is suitable, if

The serialized form of the type effectively includes the size of the rest of the serialized form in the beginning,
But the type is not a plain sequence of bytes like byte[] or ByteBuffer (for those types DataAccess and SizedReader pair of interfaces is a much better choice).

Examples of such types include lists and arrays of constant-sized elements.

Compared to BytesWriter and BytesReader, this pair of interfaces allows to share the information about the serialized form between serialization logic and the Chronicle Map, saving a few bytes by storing the serialization size only once rather than twice. This pair of interfaces is not much more difficult to implement, but the gains are also not big.

Example: serializing lists of simple Point structures:

public final class Point {

    public static Point of(double x, double y) {
        Point p = new Point();
        p.x = x;
        p.y = y;
        return p;
    }

    double x, y;
}

Serializer implementation:

public final class PointListSizedMarshaller
        implements SizedReader<List<Point>>, SizedWriter<List<Point>>,
        ReadResolvable<PointListSizedMarshaller> {

    static final PointListSizedMarshaller INSTANCE = new PointListSizedMarshaller();

    private PointListSizedMarshaller() {}

    /** A point takes 16 bytes in serialized form: 8 bytes for both x and y value */
    private static final long ELEMENT_SIZE = 16;

    @Override
    public long size(@NotNull List<Point> toWrite) {
        return toWrite.size() * ELEMENT_SIZE;
    }

    @Override
    public void write(Bytes out, long size, @NotNull List<Point> toWrite) {
        toWrite.forEach(point -> {
            out.writeDouble(point.x);
            out.writeDouble(point.y);
        });
    }

    @NotNull
    @Override
    public List<Point> read(@NotNull Bytes in, long size, List<Point> using) {
        if (size % ELEMENT_SIZE != 0) {
            throw new IORuntimeException("Bytes size should be a multiple of " + ELEMENT_SIZE +
                    ", " + size + " read");
        }
        long listSizeAsLong = size / ELEMENT_SIZE;
        if (listSizeAsLong > Integer.MAX_VALUE) {
            throw new IORuntimeException("List size couldn't be more than " + Integer.MAX_VALUE +
                    ", " + listSizeAsLong + " read");
        }
        int listSize = (int) listSizeAsLong;
        if (using == null) {
            using = new ArrayList<>(listSize);
            for (int i = 0; i < listSize; i++) {
                using.add(null);
            }
        } else if (using.size() < listSize) {
            while (using.size() < listSize) {
                using.add(null);
            }
        } else if (using.size() > listSize) {
            using.subList(listSize, using.size()).clear();
        }
        for (int i = 0; i < listSize; i++) {
            Point point = using.get(i);
            if (point == null)
                using.set(i, point = new Point());
            point.x = in.readDouble();
            point.y = in.readDouble();
        }
        return using;
    }

    @Override
    public void writeMarshallable(@NotNull WireOut wireOut) {
        // no fields to write
    }

    @Override
    public void readMarshallable(@NotNull WireIn wireIn) {
        // no fields to read
    }

    @Override
    public PointListSizedMarshaller readResolve() {
        return INSTANCE;
    }
}

Usage example:

try (ChronicleMap<String, List<Point>> objects = ChronicleMap
        .of(String.class, (Class<List<Point>>) (Class) List.class)
        .averageKey("range")
        .valueMarshaller(PointListSizedMarshaller.INSTANCE)
        .averageValue(asList(of(0, 0), of(1, 1)))
        .entries(10)
        .create()) {
    objects.put("range", asList(of(0, 0), of(1, 1)));
    objects.put("square", asList(of(0, 0), of(0, 100), of(100, 100), of(100, 0)));

    Assert.assertEquals(2, objects.get("range").size());
    Assert.assertEquals(4, objects.get("square").size());
}

`DataAccess` and `SizedReader`

This pair of interfaces is configured via ChronicleMapBuilder.keyReaderAndDataAccess() or valueReaderAndDataAccess() for the key or value type of the map, respectively.

The reader part, SizedReader, is the same as in SizedWriter and SizedReader pair, so DataAccess is an "advanced" interface to replace SizedWriter.

The main method in DataAccess is Data<T> getData(@NotNull T instance), it returns a Data accessor which is used to write "serialized" form of the instance to off-heap memory. Data.size() on the returned Data object is used for the same purpose as SizedWriter.size() method in SizedWriter and SizedReader pair interfaces. Data.writeTo() is used instead of SizedWriter.write().

DataAccess assumes that the Data object, returned from the getData() method is cached in some way, that is why it also has uninit() method to clear references to the serialized object after query operation to a Chronicle Map is over (to prevent memory leaks). This, in its turn, implies that DataAccess implementation is stateful, therefore DataAccess is made a subinterface of StatefulCopyable to force all DataAccess implementations to implement StatefulCopyable as well. See Understanding StatefulCopyable for more infromation on this. If your DataAccess implementation is not actually stateful, it is free to return this from StatefulCopyable.copy() method.

DataAccess interface is primarily intended for "serializing" objects that are already sequences of bytes and in fact doesn't require serialization, like byte[], ByteBuffer, arrays of Java primitives. For such types of objects, DataAccess allows to bypass intermediate buffering, copying data directly from objects to Chronicle Map's off-heap memory.

For example, look at the DataAccess implementation for byte[]:

public final class ByteArrayDataAccess extends AbstractData<byte[]> implements DataAccess<byte[]> {

    /** Cache field */
    private transient HeapBytesStore<byte[]> bs;

    /** State field */
    private transient byte[] array;

    public ByteArrayDataAccess() {
        initTransients();
    }

    private void initTransients() {
        bs = HeapBytesStore.uninitialized();
    }

    @Override
    public RandomDataInput bytes() {
        return bs;
    }

    @Override
    public long offset() {
        return bs.start();
    }

    @Override
    public long size() {
        return bs.capacity();
    }

    @Override
    public byte[] get() {
        return array;
    }

    @Override
    public byte[] getUsing(@Nullable byte[] using) {
        if (using == null || using.length != array.length)
            using = new byte[array.length];
        System.arraycopy(array, 0, using, 0, array.length);
        return using;
    }

    @Override
    public Data<byte[]> getData(@NotNull byte[] instance) {
        array = instance;
        bs.init(instance);
        return this;
    }

    @Override
    public void uninit() {
        array = null;
        bs.uninit();
    }

    @Override
    public DataAccess<byte[]> copy() {
        return new ByteArrayDataAccess();
    }

    @Override
    public void writeMarshallable(@NotNull WireOut wireOut) {
        // no fields to write
    }

    @Override
    public void readMarshallable(@NotNull WireIn wireIn) {
        // no fields to read
        initTransients();
    }
}

Note that getData() method returns this, and the DataAccess implementation implements Data interface as well. This is a recommended practice, because it reduces the number of objects involved (hence pointer chasing), and keeps DataAccess and Data logic together.

Data interface puts constrains on equals(), hashCode() and toString() implementations, this is why ByteArrayDataAccess subclasses AbstractData and inherits proper implementations from it. This is OK to serializer strategy implementation to have equals(), hashCode() and toString() from a very different domain, because those methods are never called on serializers inside Chronicle Map.

The easiest way to implement equals(), hashCode() and toString() is to extend AbstractData class, if it is not possible (the Data implementation already extends some other class), do this by delegating to dataEquals(), dataHashCode() and dataToString() default methods, provided right in Data interface.

Corresponding SizedReader for byte[]:

public final class ByteArraySizedReader
        implements SizedReader<byte[]>, Marshallable, ReadResolvable<ByteArraySizedReader> {

    public static final ByteArraySizedReader INSTANCE = new ByteArraySizedReader();

    private ByteArraySizedReader() {}

    @NotNull
    @Override
    public byte[] read(@NotNull Bytes in, long size, @Nullable byte[] using) {
        if (size < 0L || size > (long) Integer.MAX_VALUE) {
            throw new IORuntimeException("byte[] size should be non-negative int, " +
                    size + " given. Memory corruption?");
        }
        int arrayLength = (int) size;
        if (using == null || arrayLength != using.length)
            using = new byte[arrayLength];
        in.read(using);
        return using;
    }

    @Override
    public void writeMarshallable(@NotNull WireOut wireOut) {
        // no fields to write
    }

    @Override
    public void readMarshallable(@NotNull WireIn wireIn) {
        // no fields to read
    }

    @Override
    public ByteArraySizedReader readResolve() {
        return INSTANCE;
    }
}

(Nothing is required to use this pair of serializers with byte[] keys of values in Chronicle Map, it is used by default, if you configure byte[] key or value type.)

Understanding `StatefulCopyable`

Problems:

Sometimes on writing, reading or in other methods, defined in serialization interfaces, it is needed to operate with intermediate objects, holding some writing/reading state.

In the simplest and most common case, those objects are of some kind of buffers between serialized or deserialized instances and output or input Bytes: for example, see writer and reader implementations in the custom CharSequence encoding section.

To avoid producing a lot of garbage, those intermediate objects should be cached.

Data object, returned from getData() method of DataAccess serialization interface, should be cached in order to avoid producing garbage.
SizedWriter and DataAccess serialization interfaces, and the Data object, returned from DataAccess.getData() method, since it is cached (see the previous point) have several methods: size() and write() in SizedWriter, getData() and uninit() in DataAccess, many methods in Data. It is usually essential to save some computation results between calls to those methods from inside Chronicle Map implementation, while working some key or value during some query to Chronicle Map, because otherwise expensive computations are performed several times, that is very inefficient.

All these problems require to access some mutable, context-dependant fields from within serializer interface implementations. But, only a single instance of serializer implementation is configured for a Chronicle Map (either by keyMarshallers() or valueMarshallers() or keyReaderAndDataAccess() or valueReaderAndDataAccess() method in ChronicleMapBuilder). On the other hand, ChronicleMap is a ConcurrentMap, i. e. serializer implementations could be accessed from multiple threads concurrently. Moreover, a single ChronicleMapBuilder (with a single pair of serializer instances configured for keys and values) could be used to construct many independent ChronicleMaps, which could also be accessed concurrently.

Inefficient and fragile solution: a single instance of serialization interface, static ThreadLocal fields in serializer implementation class.

It is inefficient, because ThreadLocals are accessed each time a serializer interface method is called, that is much slower than accessing vanilla object fields.

It is fragile, because if you need ThreadLocal fields for preserving some state between calls to multiple methods in serializer interfaces over a single query to a Chronicle Map (the 3rd point in the problems list above), you should consider that a single serializer instance could be used for both keys and values of the same Chronicle Map (calls to some methods of serializer interface for serializing the key and the value over a Chronicle Map query are interspersed in unspecified order). Plus, you should consider that a single serializer instance could be used to access multiple independent Chronicle Map instances in the same thread, calls to serializers within maps could be interspersed via contexts access. So, ThreadLocal fields should be isolated not just per accessing thread, but also per serialized object domain (is the serialized object a key or a value in Chronicle Map), and per accessing Chronicle Map instance, that is hard to implement correctly.

Recommended solution: StatefulCopyable. A serializer implementation has ordinary instance fields for caching anything and preserving state between calls to different methods. It implements StatefulCopyable interface with a single method copy(), that is like Object.clone(), but copies only configuration fields of the serializer instance, not cache or state fields.

Call to copy() method at any point of a serializer instance lifetime should return an instance of the same serializer implementation in the state, exactly equal to the state of the current instance at the moment after construction and initial configuration. I. e. with "empty" cache and state. As a consequence, copy() method is transitive, i. e. serializer.copy().copy().copy() should return an object in exactly the same state, as after a single copy() call.

Chronicle Map implementation recognizes that configured BytesWriter, BytesReader, SizedWriter, SizedReader or DataAccess instance implements StatefulCopyable, and "populates" it internally via copy() for each thread, Chronicle Map instance and serialized object domain (keys or values).

See examples of implementing this interface in the custom CharSequence encoding section above.

It is allowed to return this from the copy() method, if with some configurations the serializer implementation doesn't have state. Typically this is the case when serializer is configured with sub-serializers which might be StatefulCopyable or not, for example ListMarshaller class.

Custom serialization checklist

Choose the most suitable pair of serialization interfaces: BytesWriter and BytesReader, SizedWriter and SizedReader or DataAccess and SizedReader. Recommendations on which pair to choose are given in the linked sections, describing each pair.
If implementation of the writer or reader part is configuration-less, give it a private constructor and define a single INSTANCE constant -- a sole instance of this marshaller class in the JVM. Implement ReadResolvable and return INSTANCE from readResolve() method. But, don't make implementation a Java enum.
If both the writer and reader are configuration-less, merge them into a single -Marshaller implementation class.
Make best effort in reusing using objects on the reader side (BytesReader or SizedReader), including nesting objects.
Make best effort in caching intermediate serialization results on writer side while working with some object, e. g. try not to make expensive computations in both size() and write() methods of SizedWriter implementation, but rather lazily compute them and cache in an serializer instance field.
Make best effort in reusing intermediate objects, used for reading or writing. Store them in instance fields of serializer implementation.
If a serializer implementation is stateful or have cache fields, implement StatefulCopyable. See Understanding StatefulCopyable section for more info.
Implement writeMarshallable() and readMarshallable() by writing and reading configuration fields (but not state or cache fields) of the serializer instance one-by-one, using the given WireOut/WireIn object. See Custom CharSequence encoding section for some non-trivial example of implementing these methods. See also Wire tutorial.
Don't forget to initialize transient/cache/state fileds of the instance in the end of readMarshallable() implementation. This is needed, because fefore calling readMarshallable(), Wire framework creates a serializer instance by means of Unsafe.allocateInstance() rather than calling any constructor.
If implementing DataAccess, consider implementation to be Data also, and return this from getData() method.
Don't forget to implement equals(), hashCode() and toString() in Data implementation, returned from DataAccess.getData() method, regardless if this is actually the same DataAccess object, or a separate object.
Except DataAccess which is also a Data, serializers shouldn't override Object's equals(), hashCode() and toString() (these methods are never called on serializers inside Chronicle Map library); they shouldn't implement Serializable or Externalizable (but have to implement net.openhft.chronicle.wire.Marshallable); shouldn't implement Cloneable (but have to implement StatefulCopyable, if they are stateful or have cache fields).
After implementing custom serializers, don't forget to actually apply them to ChronicleMapBuilder by keyMarshallers(), keyReaderAndDataAccess(), valueMarshallers() or valueReaderAndDataAccess() methods.

`ChronicleMap` usage patterns

Single-key queries

First of all, ChronicleMap supports all operations from Map: get(), put(), etc, including methods added in Java 8, like compute() and merge(), and ConcurrentMap interfaces: putIfAbsent(), replace(). All operations, including those which include "two steps", e. g. compute(), are correctly synchronized in terms of ConcurrentMap interface.

This means, you could use ChronicleMap instance just like a HashMap or ConcurrentHashMap:

PostalCodeRange amsterdamCodes = Values.newHeapInstance(PostalCodeRange.class);
amsterdamCodes.minCode(1011);
amsterdamCodes.maxCode(1183);
cityPostalCodes.put("Amsterdam", amsterdamCodes);

...

PostalCodeRange amsterdamCodes = cityPostalCodes.get("Amsterdam");

However, this approach often generates garbage, because the values should be deserialized from off-heap memory to on-heap, the new value object are allocated. There are several possibilities to reuse objects efficiently:

Value interfaces instead of boxed primitives

If you want to create a ChronicleMap where keys are long ids, use LongValue instead of Long key:

ChronicleMap<LongValue, Order> orders = ChronicleMap
    .of(LongValue.class, Order.class)
    .entries(1_000_000)
    .create();

LongValue key = Values.newHeapInstance(LongValue.class);
key.setValue(id);
orders.put(key, order);

...

long[] orderIds = ...
// Allocate a single heap instance for inserting all keys from the array.
// This could be a cached or ThreadLocal value as well, eliminating
// allocations altogether.
LongValue key = Values.newHeapInstance(LongValue.class);
for (long id : orderIds) {
    // Reuse the heap instance for each key
    key.setValue(id);
    Order order = orders.get(key);
    // process the order...
}

`chronicleMap.getUsing()`

Use ChronicleMap#getUsing(K key, V using) to reuse the value object. It works if:

The value type is CharSequence, pass StringBuilder as the using argument. For example:

ChronicleMap<LongValue, CharSequence> names = ...
StringBuilder name = new StringBuilder();
for (long id : ids) {
   key.setValue(id);
   names.getUsing(key, name);
   // process the name...
}

In this case, calling names.getUsing(key, name) is equivalent to

name.setLength(0);
name.append(names.get(key));

with the difference that it doesn't generate garbage.

The value type is value interface, pass heap instance to read the data into it without new object allocation:

ThreadLocal<PostalCodeRange> cachedPostalCodeRange =
   ThreadLocal.withInitial(() -> Values.newHeapInstance(PostalCodeRange.class));

...

PostalCodeRange range = cachedPostalCodeRange.get();
cityPostalCodes.getUsing(city, range);
// process the range...

If the value type implements BytesMarshallable, or Externalizable, ChronicleMap attempts to reuse the given using object by deserializing the value into the given object.
If custom marshaller is configured in the ChronicleMapBuilder via .valueMarshaller(), ChronicleMap attempts to reuse the given object by calling readUsing() method from the marshaller interface.

If ChronicleMap fails to reuse the object in getUsing(), it makes no harm, it falls back to object creation, like in get() method. In particular, even null is allowed to be passed as using object. It allows "lazy" using object initialization pattern:

// a field
PostalCodeRange cachedRange = null;

...

// in a method
cachedRange = cityPostalCodes.getUsing(city, cachedRange);
// process the range...

In this example, cachedRange is null initially, on the first getUsing() call the heap value is allocated, and saved in a cachedRange field for later reuse.

If the value type is a value interface, don't use flyweight implementation as getUsing() argument. This is dangerous, because on reusing flyweight points to the ChronicleMap memory directly, but the access is not synchronized. At least you could read inconsistent value state, at most - corrupt the ChronicleMap memory.

For accessing the ChronicleMap value memory directly use the following technique:

Working with an entry within a context section

try (ExternalMapQueryContext<CharSequence, PostalCodeRange, ?> c =
        cityPostalCodes.queryContext("Amsterdam")) {
    MapEntry<CharSequence, PostalCodeRange> entry = c.entry();
    if (entry != null) {
        PostalCodeRange range = entry.value().get();
        // Access the off-heap memory directly, by calling range
        // object getters.
        // This is very rewarding, when the value has a lot of fields
        // and expensive to copy to heap all of them, when you need to access
        // just a few fields.
    } else {
        // city not found..
    }
}

Multi-key queries

In this example, consistent graph edge addition and removals are implemented via multi-key queries:

public static boolean addEdge(
        ChronicleMap<Integer, Set<Integer>> graph, int source, int target) {
    if (source == target)
        throw new IllegalArgumentException("loops are forbidden");
    ExternalMapQueryContext<Integer, Set<Integer>, ?> sourceC = graph.queryContext(source);
    ExternalMapQueryContext<Integer, Set<Integer>, ?> targetC = graph.queryContext(target);
    // order for consistent lock acquisition => avoid dead lock
    if (sourceC.segmentIndex() <= targetC.segmentIndex()) {
        return innerAddEdge(source, sourceC, target, targetC);
    } else {
        return innerAddEdge(target, targetC, source, sourceC);
    }
}

private static boolean innerAddEdge(
        int source, ExternalMapQueryContext<Integer, Set<Integer>, ?> sourceContext,
        int target, ExternalMapQueryContext<Integer, Set<Integer>, ?> targetContext) {
    try (ExternalMapQueryContext<Integer, Set<Integer>, ?> sc = sourceContext) {
        try (ExternalMapQueryContext<Integer, Set<Integer>, ?> tc = targetContext) {
            sc.updateLock().lock();
            tc.updateLock().lock();
            MapEntry<Integer, Set<Integer>> sEntry = sc.entry();
            if (sEntry != null) {
                MapEntry<Integer, Set<Integer>> tEntry = tc.entry();
                if (tEntry != null) {
                    return addEdgeBothPresent(sc, sEntry, source, tc, tEntry, target);
                } else {
                    addEdgePresentAbsent(sc, sEntry, source, tc, target);
                    return true;
                }
            } else {
                MapEntry<Integer, Set<Integer>> tEntry = tc.entry();
                if (tEntry != null) {
                    addEdgePresentAbsent(tc, tEntry, target, sc, source);
                } else {
                    addEdgeBothAbsent(sc, source, tc, target);
                }
                return true;
            }
        }
    }
}

private static boolean addEdgeBothPresent(
        MapQueryContext<Integer, Set<Integer>, ?> sc,
        @NotNull MapEntry<Integer, Set<Integer>> sEntry, int source,
        MapQueryContext<Integer, Set<Integer>, ?> tc,
        @NotNull MapEntry<Integer, Set<Integer>> tEntry, int target) {
    Set<Integer> sNeighbours = sEntry.value().get();
    if (sNeighbours.add(target)) {
        Set<Integer> tNeighbours = tEntry.value().get();
        boolean added = tNeighbours.add(source);
        assert added;
        sEntry.doReplaceValue(sc.wrapValueAsData(sNeighbours));
        tEntry.doReplaceValue(tc.wrapValueAsData(tNeighbours));
        return true;
    } else {
        return false;
    }
}

private static void addEdgePresentAbsent(
        MapQueryContext<Integer, Set<Integer>, ?> sc,
        @NotNull MapEntry<Integer, Set<Integer>> sEntry, int source,
        MapQueryContext<Integer, Set<Integer>, ?> tc, int target) {
    Set<Integer> sNeighbours = sEntry.value().get();
    boolean added = sNeighbours.add(target);
    assert added;
    sEntry.doReplaceValue(sc.wrapValueAsData(sNeighbours));

    addEdgeOneSide(tc, source);
}

private static void addEdgeBothAbsent(MapQueryContext<Integer, Set<Integer>, ?> sc, int source,
        MapQueryContext<Integer, Set<Integer>, ?> tc, int target) {
    addEdgeOneSide(sc, target);
    addEdgeOneSide(tc, source);
}

private static void addEdgeOneSide(MapQueryContext<Integer, Set<Integer>, ?> tc, int source) {
    Set<Integer> tNeighbours = new HashSet<>();
    tNeighbours.add(source);
    MapAbsentEntry<Integer, Set<Integer>> tAbsentEntry = tc.absentEntry();
    assert tAbsentEntry != null;
    tAbsentEntry.doInsert(tc.wrapValueAsData(tNeighbours));
}

public static boolean removeEdge(
        ChronicleMap<Integer, Set<Integer>> graph, int source, int target) {
    ExternalMapQueryContext<Integer, Set<Integer>, ?> sourceC = graph.queryContext(source);
    ExternalMapQueryContext<Integer, Set<Integer>, ?> targetC = graph.queryContext(target);
    // order for consistent lock acquisition => avoid dead lock
    if (sourceC.segmentIndex() <= targetC.segmentIndex()) {
        return innerRemoveEdge(source, sourceC, target, targetC);
    } else {
        return innerRemoveEdge(target, targetC, source, sourceC);
    }
}

private static boolean innerRemoveEdge(
        int source, ExternalMapQueryContext<Integer, Set<Integer>, ?> sourceContext,
        int target, ExternalMapQueryContext<Integer, Set<Integer>, ?> targetContext) {
    try (ExternalMapQueryContext<Integer, Set<Integer>, ?> sc = sourceContext) {
        try (ExternalMapQueryContext<Integer, Set<Integer>, ?> tc = targetContext) {
            sc.updateLock().lock();
            MapEntry<Integer, Set<Integer>> sEntry = sc.entry();
            if (sEntry == null)
                return false;
            Set<Integer> sNeighbours = sEntry.value().get();
            if (!sNeighbours.remove(target))
                return false;

            tc.updateLock().lock();
            MapEntry<Integer, Set<Integer>> tEntry = tc.entry();
            if (tEntry == null)
                throw new IllegalStateException("target node should be present in the graph");
            Set<Integer> tNeighbours = tEntry.value().get();
            if (!tNeighbours.remove(source))
                throw new IllegalStateException("the target node have an edge to the source");
            sEntry.doReplaceValue(sc.wrapValueAsData(sNeighbours));
            tEntry.doReplaceValue(tc.wrapValueAsData(tNeighbours));
            return true;
        }
    }
}

Usage:

HashSet<Integer> averageValue = new HashSet<>();
for (int i = 0; i < AVERAGE_CONNECTIVITY; i++) {
    averageValue.add(i);
}
ChronicleMap<Integer, Set<Integer>> graph = ChronicleMapBuilder
        .of(Integer.class, (Class<Set<Integer>>) (Class) Set.class)
        .entries(100)
        .averageValue(averageValue)
        .create();

addEdge(graph, 1, 2);
removeEdge(graph, 1, 2);

Close `ChronicleMap`

Unlike ConcurrentHashMap, ChronicleMap stores its data off heap, often in a memory mapped file. Its recommended that you call close() once you have finished working with a ChronicleMap.

map.close()

This is especially important when working with ChronicleMap replication, as failure to call close may prevent you from restarting a replicated map on the same port. In the event that your application crashes it may not be possible to call close(). Your operating system will usually close dangling ports automatically, so although it is recommended that you close() when you have finished with the map, its not something that you must do, it's just something that we recommend you should do.

WARNING

If you call close() too early before you have finished working with the map, this can cause your JVM to crash. Close MUST BE the last thing that you do with the map.

Behaviour Customization

You could customize ChronicleMap behaviour on several levels:

ChronicleMapBuilder.entryOperations() define the "inner" listening level, all operations with entries, either during ordinary map method calls, remote calls, replication or modifications during iteration over the map, operate via this configured SPI.
ChronicleMapBuilder.mapMethods() is the higher-level of listening for local calls of Map methods. Methods in MapMethods interface correspond to Map interface methods with the same names, and define their implementations for ChronicleMap.
ChronicleMapBuilder.remoteOperations() is for listening and customizing behaviour of remote calls, and replication events.

All executions around ChronicleMap go through the three tiers (or the two bottom):

Query tier: MapQueryContext interface
Entry tier: MapEntry and MapAbsentEntry interfaces
Data tier: Data interface

MapMethods and MapRemoteOperations methods accept query context, i. e. these SPI is above the Query tier. MapEntryOperations methods accept MapEntry or MapAbsentEntry, i. e. this SPI is between Query and Entry tiers.

Combined, interception SPI interfaces and ChronicleMap.queryContext() API are powerful enough to

Log all operations of some kind on ChronicleMap (e. g. all remove, insert or update operations)
Log some specific operations on ChronicleMap (e. g. log only acquireUsing() calls, which has created a new entry)
Forbid performing operations of some kind on the ChronicleMap instance
Backup all changes to ChronicleMap to some durable storage, e. g. SQL database
Perform multi-Chronicle Map operations correctly in concurrent environment, by acquiring locks on all ChronicleMaps before updating them.
Perform multi-key operations on a single ChronicleMap correctly in concurrent environment, by acquiring locks on all keys before updating the entries
Define own replication/reconciliation logic for distributed Chronicle Maps
Dump statistics of the Chronicle Map instance -- each segment's load, size in bytes of each entry, etc.

Example - Simple logging

Just log all modification operations on ChronicleMap

class SimpleLoggingMapEntryOperations<K, V> implements MapEntryOperations<K, V, Void> {

    private static final SimpleLoggingMapEntryOperations INSTANCE =
            new SimpleLoggingMapEntryOperations();

    public static <K, V> MapEntryOperations<K, V, Void> simpleLoggingMapEntryOperations() {
        return SimpleLoggingMapEntryOperations.INSTANCE;
    }

    private SimpleLoggingMapEntryOperations() {}

    @Override
    public Void remove(@NotNull MapEntry<K, V> entry) {
        System.out.println("remove " + entry.key() + ": " + entry.value());
        entry.doRemove();
        return null;
    }

    @Override
    public Void replaceValue(@NotNull MapEntry<K, V> entry, Data<V, ?> newValue) {
        System.out.println("replace " + entry.key() + ": " + entry.value() + " -> " + newValue);
        entry.doReplaceValue(newValue);
        return null;
    }

    @Override
    public Void insert(@NotNull MapAbsentEntry<K, V> absentEntry, Data<V, ?> value) {
        System.out.println("insert " + absentEntry.absentKey() + " -> " + value);
        absentEntry.doInsert(value);
        return null;
    }
}

Usage:

ChronicleMap<IntValue, IntValue> map = ChronicleMap
        .of(Integer.class, IntValue.class)
        .entries(100)
        .entryOperations(simpleLoggingMapEntryOperations())
        .create();

// do anything with the map

Example - BiMap

Possible bidirectional map (i. e. a map that preserves the uniqueness of its values as well as that of its keys) implementation over Chronicle Maps.

enum DualLockSuccess {SUCCESS, FAIL}

class BiMapMethods<K, V> implements MapMethods<K, V, DualLockSuccess> {
    @Override
    public void remove(MapQueryContext<K, V, DualLockSuccess> q, ReturnValue<V> returnValue) {
        while (true) {
            q.updateLock().lock();
            try {
                MapEntry<K, V> entry = q.entry();
                if (entry != null) {
                    returnValue.returnValue(entry.value());
                    if (q.remove(entry) == SUCCESS)
                        return;
                }
            } finally {
                q.readLock().unlock();
            }
        }
    }

    @Override
    public void put(MapQueryContext<K, V, DualLockSuccess> q, Data<V, ?> value,
                    ReturnValue<V> returnValue) {
        while (true) {
            q.updateLock().lock();
            try {
                MapEntry<K, V> entry = q.entry();
                if (entry != null) {
                    throw new IllegalStateException();
                } else {
                    if (q.insert(q.absentEntry(), value) == SUCCESS)
                        return;
                }
            } finally {
                q.readLock().unlock();
            }
        }
    }

    @Override
    public void putIfAbsent(MapQueryContext<K, V, DualLockSuccess> q, Data<V, ?> value,
                            ReturnValue<V> returnValue) {
        while (true) {
            try {
                if (q.readLock().tryLock()) {
                    MapEntry<?, V> entry = q.entry();
                    if (entry != null) {
                        returnValue.returnValue(entry.value());
                        return;
                    }
                    // Key is absent
                    q.readLock().unlock();
                }
                q.updateLock().lock();
                MapEntry<?, V> entry = q.entry();
                if (entry != null) {
                    returnValue.returnValue(entry.value());
                    return;
                }
                // Key is absent
                if (q.insert(q.absentEntry(), value) == SUCCESS)
                    return;
            } finally {
                q.readLock().unlock();
            }
        }
    }

    @Override
    public boolean remove(MapQueryContext<K, V, DualLockSuccess> q, Data<V, ?> value) {
        while (true) {
            q.updateLock().lock();
            MapEntry<K, V> entry = q.entry();
            try {
                if (entry != null && bytesEquivalent(entry.value(), value)) {
                    if (q.remove(entry) == SUCCESS) {
                        return true;
                    } else {
                        //noinspection UnnecessaryContinue
                        continue;
                    }
                } else {
                    return false;
                }
            } finally {
                q.readLock().unlock();
            }
        }
    }

    @Override
    public void acquireUsing(MapQueryContext<K, V, DualLockSuccess> q,
                             ReturnValue<V> returnValue) {
        throw new UnsupportedOperationException();
    }

    @Override
    public void replace(MapQueryContext<K, V, DualLockSuccess> q, Data<V, ?> value,
                        ReturnValue<V> returnValue) {
        throw new UnsupportedOperationException();
    }

    @Override
    public boolean replace(MapQueryContext<K, V, DualLockSuccess> q, Data<V, ?> oldValue,
                           Data<V, ?> newValue) {
        throw new UnsupportedOperationException();
    }

    @Override
    public void compute(MapQueryContext<K, V, DualLockSuccess> q,
                        BiFunction<? super K, ? super V, ? extends V> remappingFunction,
                        ReturnValue<V> returnValue) {
        throw new UnsupportedOperationException();
    }

    @Override
    public void merge(MapQueryContext<K, V, DualLockSuccess> q, Data<V, ?> value,
                      BiFunction<? super V, ? super V, ? extends V> remappingFunction,
                      ReturnValue<V> returnValue) {
        throw new UnsupportedOperationException();
    }
}

class BiMapEntryOperations<K, V> implements MapEntryOperations<K, V, DualLockSuccess> {
    ChronicleMap<V, K> reverse;

    public void setReverse(ChronicleMap<V, K> reverse) {
        this.reverse = reverse;
    }

    @Override
    public DualLockSuccess remove(@NotNull MapEntry<K, V> entry) {
        try (ExternalMapQueryContext<V, K, ?> rq = reverse.queryContext(entry.value())) {
            if (!rq.updateLock().tryLock()) {
                if (entry.context() instanceof MapQueryContext)
                    return FAIL;
                throw new IllegalStateException("Concurrent modifications to reverse map " +
                        "during remove during iteration");
            }
            MapEntry<V, K> reverseEntry = rq.entry();
            if (reverseEntry != null) {
                entry.doRemove();
                reverseEntry.doRemove();
                return SUCCESS;
            } else {
                throw new IllegalStateException(entry.key() + " maps to " + entry.value() +
                        ", but in the reverse map this value is absent");
            }
        }
    }

    @Override
    public DualLockSuccess replaceValue(@NotNull MapEntry<K, V> entry, Data<V, ?> newValue) {
        throw new UnsupportedOperationException();
    }

    @Override
    public DualLockSuccess insert(@NotNull MapAbsentEntry<K, V> absentEntry,
                                  Data<V, ?> value) {
        try (ExternalMapQueryContext<V, K, ?> rq = reverse.queryContext(value)) {
            if (!rq.updateLock().tryLock())
                return FAIL;
            MapAbsentEntry<V, K> reverseAbsentEntry = rq.absentEntry();
            if (reverseAbsentEntry != null) {
                absentEntry.doInsert(value);
                reverseAbsentEntry.doInsert(absentEntry.absentKey());
                return SUCCESS;
            } else {
                Data<K, ?> reverseKey = rq.entry().value();
                if (reverseKey.equals(absentEntry.absentKey())) {
                    // recover
                    absentEntry.doInsert(value);
                    return SUCCESS;
                }
                throw new IllegalArgumentException("Try to associate " +
                        absentEntry.absentKey() + " with " + value + ", but in the reverse " +
                        "map this value already maps to " + reverseKey);
            }
        }
    }
}

Usage:

BiMapEntryOperations<Integer, CharSequence> biMapOps1 = new BiMapEntryOperations<>();
ChronicleMap<Integer, CharSequence> map1 = ChronicleMapBuilder
        .of(Integer.class, CharSequence.class)
        .entries(100)
        .actualSegments(1)
        .averageValueSize(10)
        .entryOperations(biMapOps1)
        .mapMethods(new BiMapMethods<>())
        .create();

BiMapEntryOperations<CharSequence, Integer> biMapOps2 = new BiMapEntryOperations<>();
ChronicleMap<CharSequence, Integer> map2 = ChronicleMapBuilder
        .of(CharSequence.class, Integer.class)
        .entries(100)
        .actualSegments(1)
        .averageKeySize(10)
        .entryOperations(biMapOps2)
        .mapMethods(new BiMapMethods<>())
        .create();

biMapOps1.setReverse(map2);
biMapOps2.setReverse(map1);

map1.put(1, "1");
System.out.println(map2.get("1"));

Example - Monitor Chronicle Map Statistics

    public static <K, V> void printMapStats(ChronicleMap<K, V> map) {
        for (int i = 0; i < map.segments(); i++) {
            try (MapSegmentContext<K, V, ?> c = map.segmentContext(i)) {
                System.out.printf("segment %d contains %d entries\n", i, c.size());
                c.forEachSegmentEntry(e -> System.out.printf("%s, %d bytes -> %s, %d bytes\n",
                        e.key(), e.key().size(), e.value(), e.value().size()));
            }
        }
    }

Entry checksums

Chronicle Map 3 is able to store entry checksums along with entries. With entry checksums it is possible to identify partially written entries (in case of operating system or power failure while) and corrupted entries (in case of hardware memory or disk corruption) and clean them up during recovery procedure.

Entry checksums are 32-bit numbers, computed by a hash function with good avalanche effect. Theoretically there is still about a one-billionth chance that after entry corruption it passes the sum check.

By default, entry checksums are ON if the Chronicle Map is persisted to disk (i. e. created via createPersistedTo() method), and OFF if the Chronicle Map is purely in-memory. Storing checksums for a purely in-memory Chronicle Map hardly makes any practical sense, but you might want to disable storing checksums for a persisted Chronicle Map by calling .checksumEntries(false) on the ChronicleMapBuilder used to create a map. It makes sense if you don't need extra safety checksums provide.

Entry checksums are computed automatically when an entry is inserted into a Chronicle Map, and re-computed automatically on operations which update the whole value, e. g. map.put(), map.replace(), map.compute(), mapEntry.doReplaceValue() (See MapEntry interface in Javadocs. But if you update values directly, bypassing Chronicle Map logic, keeping entry checksum up-to-date is also your responsibility.

It is strongly recommended to update off-heap memory of values directly only within a context and update or write lock held. Within a context, you are provided with an entry object of MapEntry type. To re-compute entry checksum manually, cast that object to ChecksumEntry type and call .updateChecksum() method on it:

try (ChronicleMap<Integer, LongValue> map = ChronicleMap
        .of(Integer.class, LongValue.class)
        .entries(1)
        // Entry checksums make sense only for persisted Chronicle Maps, and are ON by
        // default for such maps
        .createPersistedTo(file)) {

    LongValue value = Values.newHeapInstance(LongValue.class);
    value.setValue(42);
    map.put(1, value);

    try (ExternalMapQueryContext<Integer, LongValue, ?> c = map.queryContext(1)) {
        // Update lock required for calling ChecksumEntry.checkSum()
        c.updateLock().lock();
        MapEntry<Integer, LongValue> entry = c.entry();
        Assert.assertNotNull(entry);
        ChecksumEntry checksumEntry = (ChecksumEntry) entry;
        Assert.assertTrue(checksumEntry.checkSum());

        // to access off-heap bytes, should call value().getUsing() with Native value
        // provided. Simple get() return Heap value by default
        LongValue nativeValue =
                entry.value().getUsing(Values.newNativeReference(LongValue.class));
        // This value bytes update bypass Chronicle Map internals, so checksum is not
        // updated automatically
        nativeValue.setValue(43);
        Assert.assertFalse(checksumEntry.checkSum());

        // Restore correct checksum
        checksumEntry.updateChecksum();
        Assert.assertTrue(checksumEntry.checkSum());
    }
}

FAQ's

Question

I am investigating Chronicle Map as a potential replacement of Redis - with concurrency in mind. In our architecture we would be looking to replace a "large" Redis instance that currently has multiple clients connecting to it causing latency pileups due to Redis' blocking nature.

The issue is that we need to make requests in random batches of ~1000. With Redis we are able to make a single request via a Lua script (or multi-get / multi-set commands) and receive a single response. In the documentation on Chronicle Maps stateless client I see that the remote calls are blocking and can be made only one key at a time, so for us the solution is not obvious.

While I am considering passing off each individual key task to a threadpool running X blocking threads at a time, I wonder if there might be a better solution that could take advantage of doing RPC in batches and perhaps work asynchronously. As I do not see this available currently, my questions are whether this is an enhancement you might consider or if you could perhaps point me to if/how we could write our own solution for doing this - which we'd be open to contributing back...

Also, is there a reason these 1000 gets have be done serially in one thread? Why not submit 1000 get() tasks to a pool of say 20 threads, shouldn't this improve throughput / reduce latency?

Answer

stateless client is has been to be replaced with Engine. It is already not supported for ChronicleMap 3.x. ( for the rest of the answer on this question please refer to Chronile Engine FAQ's )

For get()s, sure parallelizing will reduce costs. For put()s, if you have concurrency requirements, i. e. multi-key lock before updating all of them, of cause it should be in one thread.

The stateless client which was avaible in the last version gives better performance if you use a number of threads - see https://github.com/OpenHFT/Chronicle-Map#how-to-speed-up-the-chronicle-map-stateless-client https://github.com/OpenHFT/Chronicle-Map#how-to-speed-up-the-chronicle-map-stateless-client

I don’t see that you would gain a performance benefit, in using batches, unless you are compressing the batch of data. All the data will have to be sent via tcp anyway, even if its in a batch.

Note : under high load the chronicle map stateless client consolidates many small TCP request into a single request ( when run with a number of threads. )

hhy5277 / chronicle-map Goto Github PK

chronicle-map's Introduction

Chronicle Map

3 min to understand everything about Chronicle Map

Chronicle Map is not

Features

License

Peer projects

Chronicle Map 3 Tutorial

Contents

Difference between Chronicle Map 2 and 3

Download the library

Maven Artifact Download

Maven Snapshot Download

Create a ChronicleMap Instance

In-memory Chronicle Map

Persisted Chronicle Map

"ChronicleMap instance" vs "Chronicle Map data store"

Recovery

Key and Value Types

Custom serializers

BytesWriter and BytesReader

Custom CharSequence encoding

SizedWriter and SizedReader

DataAccess and SizedReader

Understanding StatefulCopyable

Custom serialization checklist

ChronicleMap usage patterns

Single-key queries

Value interfaces instead of boxed primitives

chronicleMap.getUsing()

Working with an entry within a context section

Multi-key queries

Close ChronicleMap

WARNING

Behaviour Customization

Example - Simple logging

Example - BiMap

Example - Monitor Chronicle Map Statistics

Entry checksums

FAQ's

Question

Answer

chronicle-map's People

Contributors

Watchers

Recommend Projects

Recommend Topics

Recommend Org

Create a `ChronicleMap` Instance

"`ChronicleMap` instance" vs "Chronicle Map data store"

`BytesWriter` and `BytesReader`

Custom `CharSequence` encoding

`SizedWriter` and `SizedReader`

`DataAccess` and `SizedReader`

Understanding `StatefulCopyable`

`ChronicleMap` usage patterns

`chronicleMap.getUsing()`

Close `ChronicleMap`