forked from geekan/MetaGPT-docs
-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
Showing
2 changed files
with
173 additions
and
0 deletions.
There are no files selected for viewing
Empty file.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,173 @@ | ||
| # 断点恢复 | ||
|
|
||
| ## 定义 | ||
| 断点恢复指在程序运行过程中，记录程序不同模块的产出并落盘，当程序碰到外部如`Ctrl-C`或内部执行异常如LLM Api网络异常导致退出等情况时。再次执行程序，能够从中断前的结果中恢复继续执行，而无需从0到1开始执行，降低开发者的时间和费用成本。 | ||
|
|
||
| ## 序列化与反序列化 | ||
| 为了能支持断点恢复操作，需要对程序中的不同模块产出进行结构化存储即序列化的过程，保存后续用于恢复操作的现场。序列化的操作根据不同模块的功能有所区分，比如角色信息，初始化后即可以进行序列化，过程中不会发生改变。记忆信息，需要执行过程中，实时进行序列化保证完整性（序列化耗时在整个程序执行中的占比很低）。 | ||
|
|
||
| ## 实现逻辑 | ||
|
|
||
| ### 可能产生中断的情况 | ||
|
|
||
| - 网络等问题，LLM-Api调用重试多次后仍失败 | ||
| - Action执行过程中，输出内容解析失败导致退出 | ||
| - 人为的`Ctrl-C`对程序进行中断 | ||
|
|
||
| ### 序列化存储结构 | ||
| 为了清晰化**整体项目的结构信息**，采用层级化的方式进行内容的序列化存储。 | ||
|
|
||
| 当程序发生中断后，对应不同模块在存储目录下的文件结构如下： | ||
| 结构概要 | ||
| ```bash | ||
| ./workspace | ||
| storage | ||
| team | ||
| team_info.json # 团队需求、预算等信息 | ||
| environment # 环境 | ||
| memory.json # 环境记忆 | ||
| roles.json # 团队内的角色基础信息 | ||
| history.json # 历史信息 | ||
| roles # 团队内角色 | ||
| RoleA_Alice # 具体一个角色 | ||
| memory.json # 角色记忆 | ||
| role_info.json # 包括角色身份、执行动作、监听动作等信息 | ||
| ``` | ||
|
|
||
| 每个`xxx.json`下的为对应内容的数据概要示例。 | ||
| ```bash | ||
| ./workspace | ||
| storage | ||
| team | ||
| team_info.json # investment and so on | ||
| { | ||
| "investment": 10.0, | ||
| "idea": "write a snake game" | ||
| } | ||
| environment | ||
| roles.json # roles' meta info | ||
| [ | ||
| { | ||
| "role_class": "RoleA", | ||
| "module_name": "tests.metagpt.serialize_deserialize.test_serdeser_base", | ||
| "role_name": "RoleA" | ||
| }, | ||
| { | ||
| "role_class": "RoleB", | ||
| "module_name": "tests.metagpt.serialize_deserialize.test_serdeser_base", | ||
| "role_name": "RoleB" | ||
| } | ||
| ] | ||
| memory.json | ||
| { | ||
| "storage": [ | ||
| ], | ||
| "index": { | ||
| } | ||
| } | ||
| history.json | ||
| { | ||
| "content": "\nHuman: write a snake game\nRole A: ActionPass run passed\nHuman: write a snake game" | ||
| } | ||
| roles | ||
| ProductManager_Alice | ||
| memory.json | ||
| { | ||
| "storage": [ | ||
| ], | ||
| "index": { | ||
| } | ||
| } | ||
| role_info.json # RoleSetting _actions _states | ||
| { | ||
| "name": "RoleA", | ||
| "profile": "Role A", | ||
| "goal": "RoleA's goal", | ||
| "constraints": "RoleA's constraints", | ||
| "desc": "", | ||
| "is_human": false, | ||
| "recovered": true, | ||
| "builtin_class_name": "RoleA", | ||
| "_role_id": "RoleA(Role A)", | ||
| "_states": [ | ||
| "0. <class 'tests.metagpt.serialize_deserialize.test_serdeser_base.ActionPass'>" | ||
| ], | ||
| "_actions": [ | ||
| ], | ||
| "_rc": { | ||
| }, | ||
| "role_class": "RoleA", | ||
| "module_name": "tests.metagpt.serialize_deserialize.test_serdeser_base" | ||
| } | ||
| ``` | ||
|
|
||
| ### 恢复时的执行顺序 | ||
| 由于MetaGPT是异步执行框架，对于下述几种典型的中断截点和恢复顺序。 | ||
|
|
||
| 1. 角色A（一个action）-> 角色B（2个action），角色A进行action选择时出现异常退出。 | ||
| 2. 角色A（一个action）-> 角色B（2个action），角色B第1个action执行正常，第2个action执行时出现异常退出。 | ||
| #### 情况1 | ||
| 执行入口重新执行后，各模块进行反序列化。角色A未观察到属于自己处理的Message，不处理。角色B恢复后，观察到一条之前未处理完毕的Message，则在`_observe`后重新执行对应的`react`操作，按react策略执行对应2个动作。 | ||
|
|
||
| #### 情况2 | ||
| 执行入口重新执行后，各模块进行反序列化。角色A未观察到属于自己处理的Message，不处理。角色B恢复后，`_observe`到一条之前未完整处理完毕的Message，在`react`中，知道自己在第2个action执行失败，则直接从第2个action开始执行。 | ||
|
|
||
|
|
||
| ### 从中断前的Message开始重新执行 | ||
| 一般来说，Message是不同角色间沟通协作的桥梁，当在Message的执行过程中发生中断后，由于该Message已经被该角色存入环境（Environment）记忆（Memory）中。在进行恢复中，如果直接加载环境内的全部Memory，该角色的`_observe`将不会观察到中断时引发当时执行`Message`，从而不能恢复该Message的继续执行。 | ||
| 因此，为了保证该Message在恢复时能够继续执行，需要在发生中断后，从角色记忆中删除对应的该条Message。 | ||
|
|
||
| ### 从中断前的Action开始重新执行 | ||
| 一般来说，Action是一个相对较小的执行模块粒度，当在Action的执行过程中发生中断后，需要知道多个Actions的执行顺序以及当前执行到哪个Action。当进行恢复时，定位到中断时的Action位置并重新执行该Action。 | ||
|
|
||
| ## 结果 | ||
|
|
||
| ### 断点恢复入口 | ||
| `python3 starup.py "xxx" --recover_path "./workspace/storage/team"` # 默认序列化到`./workspace/storage/team`中。 | ||
|
|
||
| ### 恢复后继续执行结果 | ||
| `python3 -s tests/metagpt/serialize_deserialize/test_team.py`的`test_team_recover_multi_roles_save`的执行case | ||
|
|
||
| `RoleB`的`ActionRaise`模拟Action异常，执行到该Action时发生异常，序列化项目后退出。 重新启动后，`RoleA`已经执行过，不继续执行。`RoleB`的`ActionOK`已经执行过，不继续执行。继续从`ActionRaise`执行，仍异常。 | ||
|
|
||
| ```bash | ||
| 2023-11-30 20:41:22.313 | DEBUG | metagpt.team:run:92 - n_round=3 | ||
| 2023-11-30 20:41:22.314 | DEBUG | metagpt.roles.role:_observe:389 - RoleA(Role A) observed: ['Human: write a snake game...'] | ||
| 2023-11-30 20:41:22.314 | DEBUG | metagpt.roles.role:_set_state:316 - [ActionPass] | ||
| 2023-11-30 20:41:22.314 | DEBUG | metagpt.roles.role:_react:412 - RoleA(Role A): self._rc.state=0, will do ActionPass | ||
| 2023-11-30 20:41:22.314 | INFO | metagpt.roles.role:_act:361 - RoleA(Role A): ready to ActionPass | ||
| 2023-11-30 20:41:22.315 | DEBUG | metagpt.roles.role:run:472 - RoleB(Role B): no news. waiting. | ||
| 2023-11-30 20:41:27.322 | DEBUG | metagpt.roles.role:_set_state:316 - [ActionPass] | ||
| 2023-11-30 20:41:27.322 | DEBUG | metagpt.team:run:92 - n_round=2 | ||
| 2023-11-30 20:41:27.323 | DEBUG | metagpt.roles.role:run:472 - RoleA(Role A): no news. waiting. | ||
| 2023-11-30 20:41:27.324 | DEBUG | metagpt.roles.role:_observe:389 - RoleB(Role B) observed: ['Role A: ActionPass run passe...'] | ||
| 2023-11-30 20:41:27.325 | DEBUG | metagpt.roles.role:_set_state:316 - [ActionOK, ActionRaise] | ||
| 2023-11-30 20:41:27.325 | INFO | metagpt.roles.role:_act:361 - RoleB(Role B): ready to ActionOK | ||
| 2023-11-30 20:41:32.327 | DEBUG | metagpt.roles.role:_set_state:316 - [ActionOK, ActionRaise] | ||
| 2023-11-30 20:41:32.328 | INFO | metagpt.roles.role:_act:361 - RoleB(Role B): ready to ActionRaise | ||
| 2023-11-30 20:41:32.329 | WARNING | metagpt.utils.utils:wrapper:82 - There is a exception in role's execution, in order to resume, we delete the newest role communication message in the role's memory. | ||
| 2023-11-30 20:41:32.331 | ERROR | metagpt.utils.utils:wrapper:61 - Exception occurs, start to serialize the project, exp: | ||
| Traceback (most recent call last): | ||
| ... | ||
| File "/Users/xxxx/work/code/MetaGPT/metagpt/roles/role.py", line 362, in _act | ||
| response = await self._rc.todo.run(self._rc.important_memory) | ||
| File "/Users/xxxx/work/code/MetaGPT/tests/metagpt/serialize_deserialize/test_serdeser_base.py", line 50, in run | ||
| raise RuntimeError("parse error in ActionRaise") | ||
| RuntimeError: parse error in ActionRaise | ||
|
|
||
| ############################# --------- 此处开始重新执行 ----------- ############################ | ||
| 2023-11-30 20:41:32.351 | DEBUG | metagpt.team:run:92 - n_round=3 | ||
| 2023-11-30 20:41:32.351 | DEBUG | metagpt.roles.role:run:472 - RoleA(Role A): no news. waiting. | ||
| 2023-11-30 20:41:32.352 | DEBUG | metagpt.roles.role:_observe:389 - RoleB(Role B) observed: ['Role A: ActionPass run passe...'] | ||
| 2023-11-30 20:41:32.352 | DEBUG | metagpt.roles.role:_set_state:316 - [ActionOK, ActionRaise] | ||
| 2023-11-30 20:41:32.352 | INFO | metagpt.roles.role:_act:361 - RoleB(Role B): ready to ActionRaise | ||
| 2023-11-30 20:41:32.353 | WARNING | metagpt.utils.utils:wrapper:82 - There is a exception in role's execution, in order to resume, we delete the newest role communication message in the role's memory. | ||
| 2023-11-30 20:41:32.353 | ERROR | metagpt.utils.utils:wrapper:61 - Exception occurs, start to serialize the project, exp: | ||
| Traceback (most recent call last): | ||
| ... | ||
| File "/Users/xxxx/work/code/MetaGPT/metagpt/roles/role.py", line 362, in _act | ||
| response = await self._rc.todo.run(self._rc.important_memory) | ||
| File "/Users/xxxx/work/code/MetaGPT/tests/metagpt/serialize_deserialize/test_serdeser_base.py", line 50, in run | ||
| raise RuntimeError("parse error in ActionRaise") | ||
| RuntimeError: parse error in ActionRaise | ||
| ``` |