esper ha模式对比
2017-12-25 本文已影响0人
wpb
esper ha模式对比
| **Selective State Residence ** | **Capture ** | |
|---|---|---|
| Engine State 引擎状态 | Engine state is not limited to the amount of heap memory and can be much larger than available heap memory. 内存状态不受堆内存限制，可以大于堆内存 | All engine state must fit into heap memory. Therefore engine state is limited to the amount of heap memory available. |
| Memory management 内存管理 | Comprehensive; Memory resources can be actively managed and engine state can be released from memory. 综合，状态可以从内存中释放 | None; all engine state must be in memory for operation. 状态必须都在内存中 |
| Incremental checkpoint 增量checkpoint | Yes, checkpoints are always implicitly incremental. 只能增量checkpoint | Yes, checkpoints can also full backup. 支持增量checkpoint & 全量checkpoint |
| Checkpoint pause-time checkpoint暂停时间 | Minimized 最小 | Pause-time spans the whole checkpoint. 整个checkpoint周期内全暂停 |
| Store Targets 存储目标 | Multiple 文件 数据库 redis | File System 文件 |
| Recovery State Reading 状态恢复 | Reads only engine metadata. The engine state is still available and the engine reads such state at the time of state use. 只加载引擎元数据，引擎状态等到使用时再加载 | Reads all engine state into memory. 加载所有状态 |
| Resilience service level configurable per statement 持久化级别配置粒度 | No; All statements participate. 全局配置 | Yes; Profiles can declare EPL statements to be transient, durable or resilient. statements 粒度 |
| State management API to extract and load state 通过API导出、加载状态 | Not available 不支持 | Available 支持 |
esperha 自动checkpoint设置
- 示例配置
<esperha-settings>
<checkpointing enabled="true" event-count="1000" interval-msec="1000"
thread-priority="6" />
</esperha-settings>
- tips：
- enabled设置成false，用于手动checkpoint
esperha ssr-持久化到fs
<store-env-ssr name="esperha-default-store" target-name="fs-one">
<state-management>
<state-management-passive/>
</state-management>
<targets>
<target-fs name="fs-one" >
<homedir-file>./checkpoint</homedir-file>
</target-fs>
</targets>
</store-env-ssr>
