bugfix: fix the case that could not retry acquire global lock

[caohdgege]

Ⅰ. Describe what this PR did

fix: will not retry acquire lock when LOCK_RETRY_POLICY_BRANCH_ROLLBACK_ON_CONFLICT is true and autoCommit is false

Ⅱ. Does this pull request fix one issue?

fixes #3070

Ⅲ. Why don't you add test cases (unit test/integration test)?

Ⅳ. Describe how to verify it

Ⅴ. Special notes for reviews

[codecov-commenter]

Codecov Report

Merging #3133 into develop will increase coverage by 0.00%.
The diff coverage is 71.42%.

@@            Coverage Diff             @@
##             develop    #3133   +/-   ##
==========================================
  Coverage      50.46%   50.47%           
- Complexity      3111     3112    +1     
==========================================
  Files            593      593           
  Lines          19571    19572    +1     
  Branches        2427     2427           
==========================================
+ Hits            9876     9878    +2     
+ Misses          8702     8701    -1     
  Partials         993      993           

Impacted Files	Coverage Δ	Complexity Δ
...n/java/io/seata/rm/datasource/ConnectionProxy.java	29.13% <66.66%> (+1.13%)	13.00 <1.00> (ø)
...ta/rm/datasource/exec/AbstractDMLBaseExecutor.java	75.00% <100.00%> (-0.61%)	9.00 <0.00> (ø)
...o/seata/rm/datasource/AbstractConnectionProxy.java	12.37% <0.00%> (+1.03%)	6.00% <0.00%> (+1.00%)

[caohdgege]

补个图

[codecov-io]

Codecov Report

Merging #3133 (09c8d42) into develop (d8c09ab) will increase coverage by 0.01%.
The diff coverage is 71.42%.

@@              Coverage Diff              @@
##             develop    #3133      +/-   ##
=============================================
+ Coverage      52.15%   52.16%   +0.01%     
- Complexity      3509     3510       +1     
=============================================
  Files            638      638              
  Lines          21109    21110       +1     
  Branches        2614     2615       +1     
=============================================
+ Hits           11009    11012       +3     
+ Misses          9014     9012       -2     
  Partials        1086     1086              

Impacted Files	Coverage Δ	Complexity Δ
...n/java/io/seata/rm/datasource/ConnectionProxy.java	25.69% <66.66%> (+1.04%)	14.00 <2.00> (+1.00)
...ta/rm/datasource/exec/AbstractDMLBaseExecutor.java	58.82% <100.00%> (-0.80%)	11.00 <0.00> (ø)
...java/io/seata/rm/datasource/ConnectionContext.java	82.47% <0.00%> (+1.03%)	37.00% <0.00%> (+1.00%)
...o/seata/rm/datasource/AbstractConnectionProxy.java	13.18% <0.00%> (+1.09%)	6.00% <0.00%> (+1.00%)

[ggndnn]

@caohdgege I think LOCK_RETRY_POLICY_BRANCH_ROLLBACK_ON_CONFLICT==true means to release local record lock before retry when global lock conflict, that was not achieved by this pr.

[caohdgege]

@caohdgege I think LOCK_RETRY_POLICY_BRANCH_ROLLBACK_ON_CONFLICT==true means to release local record lock before retry when global lock conflict, that was not achieved by this pr.

这个之前在群里跟清铭哥讨论过，在autocommit=false 的时候，进行释放本地锁有点难，然后重试策略又比较重要，暂时先通过让它不释放本地锁直接重试，后续再优化

[ggndnn]

@caohdgege I think LOCK_RETRY_POLICY_BRANCH_ROLLBACK_ON_CONFLICT==true means to release local record lock before retry when global lock conflict, that was not achieved by this pr.

这个之前在群里跟清铭哥讨论过，在autocommit=false 的时候，进行释放本地锁有点难，然后重试策略又比较重要，暂时先通过让它不释放本地锁直接重试，后续再优化

If we have to do this, should we change the configuration meaning? Otherwise, it is easy to cause misunderstanding. By the way, how about to make use of savepoint?

[caohdgege]

@caohdgege I think LOCK_RETRY_POLICY_BRANCH_ROLLBACK_ON_CONFLICT==true means to release local record lock before retry when global lock conflict, that was not achieved by this pr.

这个之前在群里跟清铭哥讨论过，在autocommit=false 的时候，进行释放本地锁有点难，然后重试策略又比较重要，暂时先通过让它不释放本地锁直接重试，后续再优化

If we have to do this, should we change the configuration meaning? Otherwise, it is easy to cause misunderstanding. By the way, how about to make use of savepoint?

我个人的理解上，如果在 autocommit=false 的时候，如果要释放本地锁然后重试意味着要 重新执行对应的DML以及业务操作 ,用savepoint或者记录到context只能保证重新执行对应的DML，没有执行对应的业务操作。现在没能找到一个方案可以实现重新执行对应的DML以及业务操作，所以才采用了这么一个临时方案。或者云嵩哥有没有什么更好的建议？
对于修改配置的定义，云嵩哥觉得改成什么好？

[ggndnn]

@caohdgege I think LOCK_RETRY_POLICY_BRANCH_ROLLBACK_ON_CONFLICT==true means to release local record lock before retry when global lock conflict, that was not achieved by this pr.

这个之前在群里跟清铭哥讨论过，在autocommit=false 的时候，进行释放本地锁有点难，然后重试策略又比较重要，暂时先通过让它不释放本地锁直接重试，后续再优化

If we have to do this, should we change the configuration meaning? Otherwise, it is easy to cause misunderstanding. By the way, how about to make use of savepoint?

我个人的理解上，如果在 autocommit=false 的时候，如果要释放本地锁然后重试意味着要 重新执行对应的DML以及业务操作 ,用savepoint或者记录到context只能保证重新执行对应的DML，没有执行对应的业务操作。现在没能找到一个方案可以实现重新执行对应的DML以及业务操作，所以才采用了这么一个临时方案。或者云嵩哥有没有什么更好的建议？
对于修改配置的定义，云嵩哥觉得改成什么好？

I think biz op is something bigger than things here, the global lock keys are also from dml and db values, we can consider this configuration only in at mode so far. Under this consideration, what about savepoint?
As for the configuration naming, let's think about it together, maybe more discussions. When the configuration changed, the code logic may have to be changed.

[slievrly]

LGTM, the retry functionality is fine, but the meaning of the configuration item has changed and may require further consideration.

[funky-eyes]

LGTM