bugfix:PK constraint name isn't the same as the unique index name which is belong to PK

[Musknine]

Ⅰ. Describe what this PR did

In some case,the table first created the unique index, and then created the PK but didn't give a explicit name the same as unique index name.

Ⅱ. Does this pull request fix one issue?

fixes #3987

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 #4005 (d58e373) into develop (64c9598) will decrease coverage by 0.03%.
The diff coverage is 18.18%.

@@              Coverage Diff              @@
##             develop    #4005      +/-   ##
=============================================
- Coverage      49.85%   49.81%   -0.04%     
+ Complexity      3761     3759       -2     
=============================================
  Files            698      698              
  Lines          23478    23495      +17     
  Branches        2902     2908       +6     
=============================================
  Hits           11705    11705              
- Misses         10594    10609      +15     
- Partials        1179     1181       +2     

Impacted Files	Coverage Δ
...asource/sql/struct/cache/OracleTableMetaCache.java	65.68% <18.18%> (-10.79%)	⬇️
...torage/file/store/FileTransactionStoreManager.java	56.12% <0.00%> (-0.65%)	⬇️

[l81893521]

How about change it to Primary directly if pkcol constains col.getColumnName?

[Musknine]

i thought we need to check the situation that PK is multiple column PK (i was not sure if the seata will surpport composite primary key). another situation is the PK only include one column,but the unique index was composite key, so there need to compare it exactly.

[l81893521]

I have a idea that is when rsPrimary is not empty and put a new IndexMeta instead of change it. We separate the Unique Key and Primary Key. How do you think about this?

[l81893521]

We change the Unique Key to Primary key before because when we create the primary key and Oracle will create a unique key automatically but now look like the solution of change the key type is now that good.

[Musknine]

for oracle will create a unique key automatically, it's not happend in all the situation. only ,if there is no unique index,it would create one automatically.but i thought my PR logic is right.if the index is unique and the column is exacly same as PK column. it means the unique index is the PK. I couldn't agree with you more. it's a good idea that use a individual IndexMeta to indicate PK,actually,i had thought this solution before. but i didn't check all the code of the seata,i have no idea how many places need to update for. IN some other Oracle like database, example, 达梦,the defalut unique index name that was created by when we created the PK always was not the same as PK name.

[l81893521]

You do not need to change all the code in seata, just need to keep the number of primary key and index type is right that s ok. You can check out the main code here, it will no problem if you keep the logic as same as before.

io.seata.rm.datasource.sql.struct.TableMeta#getPrimaryKeyOnlyName
io.seata.rm.datasource.sql.struct.TableMeta#getPrimaryKeyMap

[Musknine]

I got what you meaning is.but it's not easy to determine how can we get the indexMeta property.actually,the PK is constraint,the actual unique index was created by PK is stored in the view of all_indexes.it means we need to determine which unique index was a part of the PK(PK include constraint name and index name).

indeed,at step 2,it's not a indexname,it's a constraint name. maybe the current PR is good. i thought the key logic is accoding the PK constriant name to determine which unique index name was a part of the PK.

[l81893521]

OK. You idea was right. You still have to add some comments and let other developer understand . For example:

// when we create a primary key constraint oracle will uses and existing unique index.
// if we create a unique index before create a primary constraint in the same column will cause the problem that primary key constraint name was different from the unique name.
// a table can have only one primary key
List<String> pkcol = new ArrayList<>();

...
//save the columns that constraint primary key name was different from unique index name
pkcol.add(rsPrimary.getString("COLUMN_NAME"));

...
//find the index that belong to the primary key constraint
for (Map.Entry<String, IndexMeta> entry : tm.getAllIndexes().entrySet())

[Musknine]

thanks for your advice. I'll add some necessary comments later.

[slievrly]

don't move imports

[Musknine]

@a364176773 @slievrly @l81893521
1, I moved back some imports that i moved down before.
2,I have added some comments.

[l81893521]

LGTM

[jsbxyyx]

LGTM