在 Akka Persistence 中,数据都缓存在服务内存(状态),后端存储的都是一些持久化的事件日志,没法使用类似 SQL 一样的 DSL 来进行分页查询。利用 Akka Streams 和 Actor 我们可以通过编码的方式来实现分页查询的效果,而且这个分页查询还是分步式并行的……
EventSourcedBehavior
Akka Persistence的EventSourcedBehavior
里实现了CQRS模型,通过commandHandler
与eventHandler
解耦了命令处理与事件处理。commandHandler
处理传入的命令并返回一个事件,并可选择将这个事件持久化;若事件需要持久化,则事件将被传给eventHandler
处理,eventHandler
处理完事件后将返回一个“新的”状态(也可以不更新,直接返回原状态)。
1 2 3 4 5
| def apply[Command, Event, State]( persistenceId: PersistenceId, emptyState: State, commandHandler: (State, Command) => Effect[Event, State], eventHandler: (State, Event) => State): EventSourcedBehavior[Command, Event, State]
|
建模
以我们习惯的数据库表建模来说,我们会有以下一张表:
1 2 3 4 5 6 7 8 9
| create table t_config ( data_id varchar(64), namespace varchar(64) not null, config_type varchar(32) not null, content text not null, constraint t_config_pk primary key (namespace, data_id) ); create index t_config_idx_data_id on t_config (data_id);
|
ConfigManager
actor 可以看作 t_config
表,它的 entityId
就是 namespace
, State 里保存了所有记录的主键值(ConfigManagerState
),这就相当于 t_config
表的 t_config_idx_data_id
索引。
而 ConfigEntity
actor 可看作 t_config
表里存储的记录,每个 actor 实例就是一行记录。它的 entityId
由 namespace
+ data_id
组成,这就相当于 t_config
表的 t_config_pk
复合主键。
这里我们定义两个 EventSourcedBehavior
:
ConfigManager
:拥有所有配置ID列表,并作为 State 保存在 EventSourcedBehavior
ConfigEntity
: 拥有每个配置数据,并作为 State 保存在 EventSourcedBehavior
实现
这里先贴出 ConfigManager
和 ConfigEntity
的部分代码,接下来再详解怎样实现分页查询。
ConfigManager
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61
| object ConfigManager { sealed trait Command extends CborSerializable sealed trait Event extends CborSerializable sealed trait Response extends CborSerializable
final case class Query(dataId: Option[String], configType: Option[String], page: Int, size: Int) extends Command final case class ReplyCommand(in: AnyRef, replyTo: ActorRef[Response]) extends Command private final case class InternalResponse(replyTo: ActorRef[Response], response: Response) extends Command
case class ConfigResponse(status: Int, message: String = "", data: Option[AnyRef] = None) extends Response
final case class ConfigManagerState(dataIds: Vector[String] = Vector()) extends CborSerializable
val TypeKey: EntityTypeKey[Command] = EntityTypeKey("ConfigManager") }
import ConfigManager._ class ConfigManager private (namespace: String, context: ActorContext[Command]) { private implicit val system = context.system private implicit val timeout: Timeout = 5.seconds import context.executionContext private val configEntity = ConfigEntity.init(context.system)
def eventSourcedBehavior(): EventSourcedBehavior[Command, Event, ConfigManagerState] = EventSourcedBehavior( PersistenceId.of(TypeKey.name, namespace), ConfigManagerState(), { case (state, ReplyCommand(in, replyTo)) => replyCommandHandler(state, replyTo, in) case (_, InternalResponse(replyTo, response)) => Effect.reply(replyTo)(response) }, eventHandler) private def processPageQuery( state: ConfigManagerState, replyTo: ActorRef[Response], in: Query): Effect[Event, ConfigManagerState] = { val offset = if (in.page > 0) (in.page - 1) * in.size else 0 val responseF = if (offset < state.dataIds.size) { Source(state.dataIds) .filter(dataId => in.dataId.forall(v => v.contains(dataId))) .mapAsync(20) { dataId => configEntity.ask[Option[ConfigState]](replyTo => ShardingEnvelope(dataId, ConfigEntity.Query(in.configType, replyTo))) } .collect { case Some(value) => value } .drop(offset) .take(in.size) .runWith(Sink.seq) .map(items => ConfigResponse(IntStatus.OK, data = Some(items))) } else { Future.successful(ConfigResponse(IntStatus.NOT_FOUND, data = Some(Nil))) } context.pipeToSelf(responseF) { case Success(value) => InternalResponse(replyTo, value) case Failure(e) => InternalResponse(replyTo, ConfigResponse(IntStatus.INTERNAL_ERROR, e.getLocalizedMessage)) } Effect.none } }
|
ConfigEntity
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
| object ConfigEntity { case class ConfigState(namespace: String, dataId: String, configType: String, content: String)
sealed trait Command extends CborSerializable sealed trait Event extends CborSerializable
final case class Query(configType: Option[String], replyTo: ActorRef[Option[ConfigState]]) extends Command
final case class ConfigEntityState(config: Option[ConfigState] = None) extends CborSerializable
val TypeKey: EntityTypeKey[Command] = EntityTypeKey("ConfigEntity") }
import ConfigEntity._ class ConfigEntity private (namespace: String, dataId: String, context: ActorContext[Command]) { def eventSourcedBehavior(): EventSourcedBehavior[Command, Event, ConfigEntityState] = EventSourcedBehavior(PersistenceId.of(TypeKey.name, dataId), ConfigEntityState(), commandHandler, eventHandler)
def commandHandler(state: ConfigEntityState, command: Command): Effect[Event, ConfigEntityState] = command match { case Query(configType, replyTo) => state.config match { case None => Effect.reply(replyTo)(None) case Some(config) => val resp = if (configType.forall(v => config.configType.contains(v))) Some(config) else None Effect.reply(replyTo)(resp) } } }
|
ConfigManager#processPageQuery
函数实现了大部分的分页查询逻辑(有部分逻辑需要由 ConfigEntity
处理)。
1 2 3 4 5 6
| val offset = if (in.page > 0) (in.page - 1) * in.size else 0 val responseF = if (offset < state.dataIds.size) { } else { Future.successful(ConfigResponse(IntStatus.OK, data = Some(Nil))) }
|
这里首先获取实际的分页数据偏移量 offset
,再于 ConfigManager
状态里保存的 dataIds
的大小进行判断,若 offset
< state.dataIds.size
则我们进行分页逻辑,否则直接返回一个空列表给前端。
1 2 3 4 5 6 7 8 9 10 11
| Source(state.dataIds) .filter(dataId => in.dataId.forall(v => v.contains(dataId))) .mapAsync(20) { dataId => configEntity.ask[Option[ConfigState]](replyTo => ShardingEnvelope(s"$namespace@$dataId", ConfigEntity.Query(in.configType, replyTo))) } .collect { case Some(value) => value } .drop(offset) .take(in.size) .runWith(Sink.seq) .map(items => ConfigResponse(IntStatus.OK, data = Some(items)))
|
这个 Akka Streams 流即是分页处理的主要实现,若是SQL的话,它类似:
1
| select * from t_config where data_id like '%"in.dataId"%' offset "offset" limit "in.size"
|
.mapAsync
在流执行流程中起了20个并发的异步操作,将委托每个匹配的 ConfigEntity
(由s"$namespace@$dataId"
生成entityId
)执行 config_type
字段的查询。这样,完整的SQL语句类似:
1
| select * from t_config where data_id like '%"in.dataId"%' and config_type = "in.configType" offset "offset" limit "in.size"
|
ConfigEntity
对 config_type
部分的查询逻辑实现如下:
1 2 3 4 5 6 7 8
| case Query(configType, replyTo) => state.config match { case None => Effect.reply(replyTo)(None) case Some(config) => val resp = if (configType.forall(v => config.configType.contains(v))) Some(config) else None Effect.reply(replyTo)(resp) }
|
若in.configType
为空,既不需要判断 config_type
这个字段,直接返回 Some(config)
即可,而这时的SQL语句类似:
1
| select * from t_config where data_id like '%"in.dataId"%' and true offset "offset" limit "in.size"
|
Tip这里有个小技巧,对于 Option[T]
字段的判断,直接使用了 .forall
方法,它等价于:
1 2 3 4
| option match { case Some(x) => p(x) case None => true }
|
小结
完整代码可在此 https://github.com/yangbajing/yangbajing-blog/tree/master/src/main/scala/blog/persistence/config 找到。