[konlue]

Fixes #7724.

Summary

InstanceInfoMapper.selectOneByQuery has no Oracle-specific mapping: the shared statement ends with LIMIT 1, which Oracle rejects. InstanceInfoServiceImpl#createOrUpdate calls this lookup before every insert/update, so instance identity persistence cannot work on an Oracle deployment.

The same mapper already demonstrates the supported pattern in its existed statement (databaseId="oracle" + ROWNUM = 1).

Modifications

• Add a databaseId="oracle" variant of selectOneByQuery that keeps every identity/namespace predicate and ends with AND ROWNUM = 1.
• Add a statement-shape regression test that parses the mapper with databaseId="oracle" and checks the resolved SQL.

Verifying this change

• New InstanceInfoMapperOracleSqlTest: with databaseId="oracle", selectOneByQuery resolves to SQL containing ROWNUM and no LIMIT, with all four filters (instance_ip, namespace_id, instance_port, instance_type) intact.

• Red-green: without the Oracle variant the test fails and reports the generic ... limit 1 statement; with it the test passes.

• Targeted tests pass:

  ./mvnw -B -ntp -pl shenyu-admin test -Dtest=InstanceInfoMapperOracleSqlTest

  1/1 green.

• Checkstyle passes:

  ./mvnw -B -ntp -pl shenyu-admin checkstyle:check

  clean.

Notes

AND ROWNUM = 1 applies after the filters and preserves the single-row semantics of LIMIT 1. The test pins MyBatis dialect selection locally; execution against a real Oracle instance is exercised by the repository's SQL matrix.

[Aias00]

LGTM. The variant is a faithful copy of the generic statement plus a valid Oracle row limit.

Evidence

• The five predicates (instance_ip, namespace_id, instance_port, instance_type, id) match the generic statement one for one — I diffed them, nothing was dropped and no bind name changed.
• AND ROWNUM = 1 after the filters preserves LIMIT 1 semantics: ROWNUM is assigned to rows that already satisfy the WHERE clause, so the first matching row is returned. WHERE 1=1 keeps the dynamic predicates uniform.
• The same pattern already exists in this mapper (existed, databaseId="oracle"), so the DatabaseIdProvider resolution is established, and the test pins which statement MyBatis picks.
• The red-green is real: on master the resolved statement ends with LIMIT 1, so assertFalse(sql.contains("limit")) fails.
• createOrUpdate builds its query from ip/port/type/namespace only, so the id filter is a bonus and its absence in the test does not weaken the regression.

Non-blocking

1. Two statements now have to be kept in sync by hand. If you touch this again, a shared <sql> fragment for the predicate block with only the tail (LIMIT 1 vs AND ROWNUM = 1) differing would remove the drift risk.
2. SElECT casing is inherited from the five existing statements in the file, so the new one is internally consistent — just noting that a case-sensitive SQL lint would flag all six.
3. The test asserts four of the five filters; adding instanceId would cover the one predicate the caller never sets.