Support for Direct Path Loading

[ceyhunkerti]

Hello, are there any plans to support direct path loading?

[cjbj]

@ceyhunkerti nothing in the immediate plans. In some discussion long ago there was a thought that it has some limited, targeted use cases, but we would need to evaluate this.

Can you provide info about what you want to use it for (data sizes & types, table types & indexes, etc etc)?

[ceyhunkerti]

@cjbj yes I searched the issues before and saw some discussion dating to several years ago. Also same question is asked in python/cx_oracle long ago.

Here is my thoughts and what I need it for If you bear with me:

• I am trying to create a data transfer utility in Zig and using odpi-c for oracle connection
• Will support different databases and just prototyped oracle side first.
• But from my prior experience sqlldr is the fastest way to load data in most of the cases (ETL loads) which uses direct path loading.
• So, I also want to enable users to opt in for direct path in my run configuration.
• I also tried to switch to ocilib since it has DPL support, although it starts smooth I had some troubles with it later on. Also the owner of ocilib repo addresses some of these in here to you.
• Lastly I've been using data transfer tools/methods for years and nowadays since python is very popular, tools in the singer family (there are many with the same logic) are using cx_Oracle hence odpi-c and they don't support DPL.

Thanks for reading so far :)

Now I have two options:

• I'll fallback to OCI which needs huge boilerplate.
• Skip the DPL support and move on.

I also have a third option in my mind but I don't know how to do it so far. Which is;

• Use ODPI and implement the DPL part in zig
  but didn't yet see how can I access lets say oci handles like env, conn, etc from dpi_struct.

And maybe

• Patch odpi-c in my repo to include DPL

[tgulacsi]

I'd interested in Direct Load Path it it's faster than OCIExecuteMany.
(csvload, import from Parquet)

[cjbj]

What sort of data size are you (all) using?

[ceyhunkerti]

For me;
I don't need it for a specific data size range. it'll be an option in my program for large data sizes let's say TB scale.