feat: Add Oracle Document Loader and Vector Store integration #8007
Conversation
dosubot
bot
added
the
size:XXL
|
The latest updates on your projects. Learn more about Vercel for Git ↗︎
1 Skipped Deployment
|
|
@cjbj can you please let us know if you have any more comments. We would be really grateful if we can merge this in. Several of our customers with a large JS presence would really like to use it as Langchain is the go to platform for everyone nowadays and Oracle DB is where all of their mission critical data is. |
docs/core_docs/docs/integrations/document_loaders/file_loaders/oracleai.mdx
Outdated
Show resolved
Hide resolved
docs/core_docs/docs/integrations/document_loaders/file_loaders/oracleai.mdx
Outdated
Show resolved
Hide resolved
docs/core_docs/docs/integrations/document_loaders/file_loaders/oracleai.mdx
Outdated
Show resolved
Hide resolved
| } | ||
| const result = await this.conn.execute( | ||
| <string>( | ||
| `select dbms_vector_chain.utl_to_text(t.${this.pref.colname}, :pref) text, dbms_vector_chain.utl_to_text(t.${this.pref.colname}, json('{"plaintext": "false"}')) metadata from ${this.pref.owner}.${this.pref.tablename} t` |
There was a problem hiding this comment.
Have the substituted values been filtered to avoid any SQL Injection issues?
There was a problem hiding this comment.
- You need to bind those values !
- Would it be nice to throw a custom error message if the check fails?
const cn = 'empno';
const ow = 'scott';
const tn = 'emp';
const qn = `${ow}.${tn}`;
const sql = `select sys.dbms_assert.simple_sql_name(:sn), sys.dbms_assert.qualified_sql_name(:qn) from dual`;
const binds = [cn, qn];
await connection.execute(sql, binds);
There was a problem hiding this comment.
Changed to bind values and added a custom error message as suggested
docs/core_docs/docs/integrations/document_loaders/file_loaders/oracleai.mdx
Outdated
Show resolved
Hide resolved
libs/langchain-community/src/vectorstores/tests/oraclevs_test.ts
Outdated
Show resolved
Hide resolved
| const embeddings = []; | ||
|
|
||
| if (oracledb.thin) { | ||
| // thin mode, can't use batching |
There was a problem hiding this comment.
Is something missing in node-oracledb Thin mode?
There was a problem hiding this comment.
The batching version of utl_to_embedding passes a collection called vector_array_t which is an array of clobs and I think thin mode didn't support this. Hence, in thin mode we embed each input at a time.
There was a problem hiding this comment.
Noted. We (node-oracledb) will revisit this.
hackerdave
changed the title
3 participants
Description
This transaction integrates Oracle Document Loader and Oracle Vector Search with LangChain.js. The integration enables the use of Oracle's advanced vector search capabilities within the LangChain.js framework.
Key Features
Fixes
Testing
Verified integration with unit tests.
Added relevant test cases to ensure proper functionality.
Request
Please let @skmishraoracle and @rohanaggarwal7997 know if there are any issues or areas that need improvement.