CosId Universal, flexible, high-performance distributed ID generator
CosId aims to provide a universal, flexible and high-performance distributed ID generator.
CosIdGenerator: Stand-alone TPS performance：15,570,085 ops/s , three times that ofUUID.randomUUID(),global trend increasing based-time.SnowflakeId: Stand-alone TPS performance：4,096,000 ops/s JMH Benchmark , It mainly solves two major problems ofSnowflakeId: machine number allocation problem and clock backwards problem and provide a more friendly and flexible experience.SegmentId: Get a segment (Step) ID every time to reduce the network IO request frequency of theIdSegmentdistributor and improve performance.IdSegmentDistributor:RedisIdSegmentDistributor:IdSegmentdistributor based on Redis.JdbcIdSegmentDistributor: The Jdbc-basedIdSegmentdistributor supports various relational databases.ZookeeperIdSegmentDistributor:IdSegmentdistributor based on Zookeeper.MongoIdSegmentDistributor:IdSegmentdistributor based on MongoDB.
SegmentChainId(recommend):SegmentChainId(lock-free) is an enhancement ofSegmentId, the design diagram is as follows.PrefetchWorkermaintains asafe distance, so thatSegmentChainIdachieves approximatelyAtomicLongTPS performance: 127,439,148+ ops/s JMH Benchmark .PrefetchWorkermaintains a safe distance (safeDistance), and supports dynamicsafeDistanceexpansion and contraction based on hunger status.
SnowflakeId is a distributed ID algorithm that uses
Long(64-bit) bit partition to generate ID. The general bit allocation scheme is :timestamp(41-bit) +machineId(10-bit) +sequence(12-bit) = 63-bit。
- 41-bit
timestamp= (1L<<41)/(1000/3600/24/365) approximately 69 years of timestamp can be stored, that is, the usable absolute time isEPOCH+ 69 years. Generally, we need to customizeEPOCHas the product development time. In addition, we can increase the number of allocated bits by compressing other areas， The number of timestamp bits to extend the available time. - 10-bit
machineId= (1L<<10) = 1024 That is, 1024 copies of the same business can be deployed (there is no master-slave copy in the Kubernetes concept, and the definition of Kubernetes is directly used here) instances. Generally, there is no need to use so many, so it will be redefined according to the scale of deployment. - 12-bit
sequence= (1L<<12) * 1000 = 4096000 That is, a single machine can generate about 409W ID per second, and a global same-service cluster can generate4096000*1024=4194304000=4.19 billion (TPS).
It can be seen from the design of SnowflakeId:
- 👍 The first 41-bit are a
timestamp,So SnowflakeId is local monotonically increasing, and affected by global clock synchronization SnowflakeId is global trend increasing. - 👍
SnowflakeIddoes not have a strong dependency on any third-party middleware, and its performance is also very high. - 👍 The bit allocation scheme can be flexibly configured according to the needs of the business system to achieve the optimal use effect.
- 👎 Strong reliance on the local clock, potential clock moved backwards problems will cause ID duplication.
- 👎 The
machineIdneeds to be set manually. If themachineIdis manually assigned during actual deployment, it will be very inefficient.
It mainly solves two major problems of SnowflakeId: machine number allocation problem and clock backwards problem and
provide a more friendly and flexible experience.
Currently CosId provides the following three
MachineIddistributors.cosid: snowflake: machine: distributor: type: manual manual: machine-id: 0Manually distribute
MachineIdcosid: snowflake: machine: distributor: type: stateful_setUse the stable identification ID provided by the
StatefulSetofKubernetesas the machine number.
cosid: snowflake: machine: distributor: type: redisUse Redis as the distribution store for the machine number.
cosid: snowflake: clock-backwards: spin-threshold: 10 broken-threshold: 2000The default
DefaultClockBackwardsSynchronizerclock moved backwards synchronizer uses active wait synchronization strategy,spinThreshold(default value 10 milliseconds) is used to set the spin wait threshold, when it is greater thanspinThreshold, use thread sleep to wait for clock synchronization, if it exceedsBrokenThreshold(default value 2 seconds) will directly throw aClockTooManyBackwardsExceptionexception.public class MachineState { public static final MachineState NOT_FOUND = of(-1, -1); private final int machineId; private final long lastTimeStamp; public MachineState(int machineId, long lastTimeStamp) { this.machineId = machineId; this.lastTimeStamp = lastTimeStamp; } public int getMachineId() { return machineId; } public long getLastTimeStamp() { return lastTimeStamp; } public static MachineState of(int machineId, long lastStamp) { return new MachineState(machineId, lastStamp); } }cosid: snowflake: machine: state-storage: local: state-location: ./cosid-machine-state/The default
LocalMachineStateStoragelocal machine state storage uses a local file to store the machine number and the most recent timestamp, which is used as aMachineStatecache.cosid: snowflake: share: clock-sync: trueThe default
SnowflakeIdwill directly throw aClockBackwardsExceptionwhen a clock moved backwards occurs, while using theClockSyncSnowflakeIdwill use theClockBackwardsSynchronizerto actively wait for clock synchronization to regenerate the ID, providing a more user-friendly experience.SnowflakeId snowflakeId=SafeJavaScriptSnowflakeId.ofMillisecond(1);The
Number.MAX_SAFE_INTEGERofJavaScripthas only 53-bit. If the 63-bitSnowflakeIdis directly returned to the front end, the value will overflow. Usually we can convertSnowflakeIdto String type or customizeSnowflakeIdBit allocation is used to shorten the number of bits ofSnowflakeIdso thatIDdoes not overflow when it is provided to the front end.cosid: snowflake: share: friendly: truepublic class SnowflakeIdState { private final long id; private final int machineId; private final long sequence; private final LocalDateTime timestamp; /** * {@link #timestamp}-{@link #machineId}-{@link #sequence} */ private final String friendlyId; }public interface SnowflakeFriendlyId extends SnowflakeId { SnowflakeIdState friendlyId(long id); SnowflakeIdState ofFriendlyId(String friendlyId); default SnowflakeIdState friendlyId() { long id = generate(); return friendlyId(id); } }SnowflakeFriendlyId snowflakeFriendlyId=new DefaultSnowflakeFriendlyId(snowflakeId); SnowflakeIdState idState=snowflakeFriendlyId.friendlyId(); idState.getFriendlyId(); //20210623131730192-1-0
cosid: segment: enabled: true distributor: type: redisInitialize the
cosidtablecreate table if not exists cosid ( name varchar(100) not null comment '{namespace}.{name}', last_max_id bigint not null default 0, last_fetch_time bigint not null, constraint cosid_pk primary key (name) ) engine = InnoDB;spring: datasource: url: jdbc:mysql://localhost:3306/test_db username: root password: root cosid: segment: enabled: true distributor: type: jdbc jdbc: enable-auto-init-cosid-table: false enable-auto-init-id-segment: trueAfter enabling
enable-auto-init-id-segment:true, the application will try to create theidSegmentrecord when it starts to avoid manual creation. Similar to the execution of the following initialization sql script, there is no need to worry about misoperation, becausenameis the primary key.insert into cosid (name, last_max_id, last_fetch_time) value ('namespace.name', 0, unix_timestamp());
cosid: segment: enabled: true mode: chain chain: safe-distance: 5 prefetch-worker: core-pool-size: 2 prefetch-period: 1s distributor: type: redis share: offset: 0 step: 100 provider: bizC: offset: 10000 step: 100 bizD: offset: 10000 step: 100cosid: snowflake: provider: bizA: # timestamp-bit: sequence-bit: 12 bizB: # timestamp-bit: sequence-bit: 12IdGenerator idGenerator=idGeneratorProvider.get("bizA");In actual use, we generally do not use the same
IdGeneratorfor all business services, but different businesses use differentIdGenerator, thenIdGeneratorProviderexists to solve this problem, and it is the container ofIdGenerator, You can get the correspondingIdGeneratorby the business name.Kotlin DSL
implementation("me.ahoo.cosid:cosid-mybatis:${cosidVersion}")@Target({ElementType.FIELD}) @Documented @Retention(RetentionPolicy.RUNTIME) public @interface CosId { String value() default IdGeneratorProvider.SHARE; boolean friendlyId() default false; }public class LongIdEntity { @CosId(value = "safeJs") private Long id; public Long getId() { return id; } public void setId(Long id) { this.id = id; } } public class FriendlyIdEntity { @CosId(friendlyId = true) private String id; public String getId() { return id; } public void setId(String id) { this.id = id; } }@Mapper public interface OrderRepository { @Insert("insert into t_table (id) value (#{id});") void insert(LongIdEntity order); @Insert({ "<script>", "insert into t_friendly_table (id)", "VALUES" + "<foreach item='item' collection='list' open='' separator=',' close=''>" + "(#{item.id})" + "</foreach>", "</script>"}) void insertList(List<FriendlyIdEntity> list); }LongIdEntity entity=new LongIdEntity(); entityRepository.insert(entity); /** * { * "id": 208796080181248 * } */ return entity;spring: shardingsphere: rules: sharding: key-generators: cosid: type: COSID props: id-name: __share__
- Ease of use: supports multiple data types (
Long/LocalDateTime/DATE/String/SnowflakeId),The official implementation is to first convert to a string and then convert toLocalDateTime, the conversion success rate is affected by the time formatting characters.- Performance: Compared to
org.apache.shardingsphere.sharding.algorithm.sharding.datetime.IntervalShardingAlgorithm,The performance is 1200~4000 times higher.
PreciseShardingValue RangeShardingValue
- CosIdIntervalShardingAlgorithm
- type: COSID_INTERVAL
spring: shardingsphere: rules: sharding: sharding-algorithms: alg-name: type: COSID_INTERVAL props: logic-name-prefix: logic-name-prefix id-name: cosid-name datetime-lower: 2021-12-08 22:00:00 datetime-upper: 2022-12-01 00:00:00 sharding-suffix-pattern: yyyyMM datetime-interval-unit: MONTHS datetime-interval-amount: 1
- Performance: Compared to
org.apache.shardingsphere.sharding.algorithm.sharding.datetime.IntervalShardingAlgorithm,The performance is 1200~4000 times higher.And it has higher stability and no serious performance degradation.
PreciseShardingValue RangeShardingValue 项目中根据使用的场景（
jdbc/proxy/redis-cosid/redis/shardingsphere/zookeeper等）提供了对应的例子，实践过程中可以参照配置快速接入。Kotlin DSL
val cosidVersion = "1.14.5"; implementation("me.ahoo.cosid:cosid-spring-boot-starter:${cosidVersion}")<?xml version="1.0" encoding="UTF-8"?> <project xmlns="http://maven.apache.org/POM/4.0.0" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd"> <modelVersion>4.0.0</modelVersion> <artifactId>demo</artifactId> <properties> <cosid.version>1.14.5</cosid.version> </properties> <dependencies> <dependency> <groupId>me.ahoo.cosid</groupId> <artifactId>cosid-spring-boot-starter</artifactId> <version>${cosid.version}</version> </dependency> </dependencies> </project>spring: shardingsphere: datasource: names: ds0,ds1 ds0: type: com.zaxxer.hikari.HikariDataSource driver-class-name: com.mysql.cj.jdbc.Driver jdbcUrl: jdbc:mysql://localhost:3306/cosid_db_0 username: root password: root ds1: type: com.zaxxer.hikari.HikariDataSource driver-class-name: com.mysql.cj.jdbc.Driver jdbcUrl: jdbc:mysql://localhost:3306/cosid_db_1 username: root password: root props: sql-show: true rules: sharding: binding-tables: - t_order,t_order_item tables: cosid: actual-data-nodes: ds0.cosid t_table: actual-data-nodes: ds0.t_table_$->{0..1} table-strategy: standard: sharding-column: id sharding-algorithm-name: table-inline t_date_log: actual-data-nodes: ds0.t_date_log_202112 key-generate-strategy: column: id key-generator-name: snowflake table-strategy: standard: sharding-column: create_time sharding-algorithm-name: data-log-interval sharding-algorithms: table-inline: type: COSID_MOD props: mod: 2 logic-name-prefix: t_table_ data-log-interval: type: COSID_INTERVAL props: logic-name-prefix: t_date_log_ datetime-lower: 2021-12-08 22:00:00 datetime-upper: 2022-12-01 00:00:00 sharding-suffix-pattern: yyyyMM datetime-interval-unit: MONTHS datetime-interval-amount: 1 key-generators: snowflake: type: COSID props: id-name: snowflake cosid: namespace: ${spring.application.name} machine: enabled: true # stable: true # machine-bit: 10 # instance-id: ${HOSTNAME} distributor: type: redis # manual: # machine-id: 0 snowflake: enabled: true # epoch: 1577203200000 clock-backwards: spin-threshold: 10 broken-threshold: 2000 share: clock-sync: true friendly: true provider: order_item: # timestamp-bit: sequence-bit: 12 snowflake: sequence-bit: 12 safeJs: machine-bit: 3 sequence-bit: 9 segment: enabled: true mode: chain chain: safe-distance: 5 prefetch-worker: core-pool-size: 2 prefetch-period: 1s distributor: type: redis share: offset: 0 step: 100 provider: order: offset: 10000 step: 100 longId: offset: 10000 step: 100
- The development notebook : MacBook Pro (M1)
- All benchmark tests are carried out on the development notebook.
- Deploying Redis on the development notebook.
gradle cosid-core:jmh # or java -jar cosid-core/build/libs/cosid-core-1.14.5-jmh.jar -bm thrpt -wi 1 -rf json -f 1Benchmark Mode Cnt Score Error Units SnowflakeIdBenchmark.millisecondSnowflakeId_friendlyId thrpt 4020311.665 ops/s SnowflakeIdBenchmark.millisecondSnowflakeId_generate thrpt 4095403.859 ops/s SnowflakeIdBenchmark.safeJsMillisecondSnowflakeId_generate thrpt 511654.048 ops/s SnowflakeIdBenchmark.safeJsSecondSnowflakeId_generate thrpt 539818.563 ops/s SnowflakeIdBenchmark.secondSnowflakeId_generate thrpt 4206843.941 ops/s
In statistics, a percentile (or a centile) is a score below which a given percentage of scores in its frequency distribution falls (exclusive definition) or a score at or below which a given percentage falls (inclusive definition). For example, the 50th percentile (the median) is the score below which (exclusive) or at or below which (inclusive) 50% of the scores in the distribution may be found.
CosId (
SegmentChainId) is 5 times faster than Leaf(segment).
