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 theIdSegment
distributor and improve performance.IdSegmentDistributor
:RedisIdSegmentDistributor
:IdSegment
distributor based on Redis.JdbcIdSegmentDistributor
: The Jdbc-basedIdSegment
distributor supports various relational databases.ZookeeperIdSegmentDistributor
:IdSegment
distributor based on Zookeeper.MongoIdSegmentDistributor
:IdSegment
distributor based on MongoDB.
SegmentChainId
(recommend):SegmentChainId
(lock-free) is an enhancement ofSegmentId
, the design diagram is as follows.PrefetchWorker
maintains asafe distance
, so thatSegmentChainId
achieves approximatelyAtomicLong
TPS performance: 127,439,148+ ops/s JMH Benchmark .PrefetchWorker
maintains a safe distance (safeDistance
), and supports dynamicsafeDistance
expansion 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 customizeEPOCH
as 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. - 👍
SnowflakeId
does 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
machineId
needs to be set manually. If themachineId
is 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
MachineId
distributors.
cosid:
snowflake:
machine:
distributor:
type: manual
manual:
machine-id: 0
Manually distribute
MachineId
cosid:
snowflake:
machine:
distributor:
type: stateful_set
Use the stable identification ID provided by the
StatefulSet
ofKubernetes
as the machine number.
cosid:
snowflake:
machine:
distributor:
type: redis
Use Redis as the distribution store for the machine number.
cosid:
snowflake:
clock-backwards:
spin-threshold: 10
broken-threshold: 2000
The default DefaultClockBackwardsSynchronizer
clock 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
than spinThreshold
, use thread sleep to wait for clock synchronization, if it exceeds BrokenThreshold
(default value
2 seconds) will directly throw a ClockTooManyBackwardsException
exception.
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 LocalMachineStateStorage
local machine state storage uses a local file to store the machine number and the
most recent timestamp, which is used as a MachineState
cache.
cosid:
snowflake:
share:
clock-sync: true
The default SnowflakeId
will directly throw a ClockBackwardsException
when a clock moved backwards occurs, while
using the ClockSyncSnowflakeId
will use the ClockBackwardsSynchronizer
to actively wait for clock synchronization to
regenerate the ID, providing a more user-friendly experience.
SnowflakeId snowflakeId=SafeJavaScriptSnowflakeId.ofMillisecond(1);
The Number.MAX_SAFE_INTEGER
of JavaScript
has only 53-bit. If the 63-bit SnowflakeId
is directly returned to the
front end, the value will overflow. Usually we can convert SnowflakeId
to String type or customize SnowflakeId
Bit
allocation is used to shorten the number of bits of SnowflakeId
so that ID
does not overflow when it is provided to
the front end.
cosid:
snowflake:
share:
friendly: true
public 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: redis
Initialize the
cosid
table
create 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: true
After enabling enable-auto-init-id-segment:true
, the application will try to create the idSegment
record 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, because name
is 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: 100
cosid:
snowflake:
provider:
bizA:
# timestamp-bit:
sequence-bit: 12
bizB:
# timestamp-bit:
sequence-bit: 12
IdGenerator idGenerator=idGeneratorProvider.get("bizA");
In actual use, we generally do not use the same IdGenerator
for all business services, but different businesses use
different IdGenerator
, then IdGeneratorProvider
exists to solve this problem, and it is the container
of IdGenerator
, You can get the corresponding IdGenerator
by 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 |
---|---|
spring:
shardingsphere:
rules:
sharding:
sharding-algorithms:
alg-name:
type: COSID_MOD
props:
mod: 4
logic-name-prefix: t_table_
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 1
Benchmark 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
).