Hey All,

I propose adding *scan-planning-mode *to loadTable API, which is an
optional value in the loadTable config section, which when present clients
MUST use it to decide which mode of scan planning they wanna do, server
side (using IRC scan planning API) or client side (client reading the
manifest and then figuring out FileScan Tasks).

For details please check :
 - PR : https://github.com/apache/iceberg/pull/14867

Some summary on background discussion :
We debated a lot offline on what does MUST means to the client, as if does
the client has a liberty to fail fast if they have configured something in
their client side config which is orthogonal to what server is suggesting
and it feels like we had 2 options from the client end, either fail fast or
let the server override the client side config, it seemed like server
overriding the client side config with the client logging this as a warning
is what i have implemented mostly from pov what's done *today* for other
configs.
 I do think we should think a bit more about how server side overrides go
along with the client side configs (I understand this is more *client side*
implementation details than directly related directly to server)  and plan
to start a thread discussing this more in depth. I wanted to share a
summary of this discussion (which is captured in pr as well [here
<https://github.com/apache/iceberg/pull/14867#discussion_r2799750448>]) to
keep the wider community aware.

Please vote in the next 72 hours:

[ ] +1 Add these changes to the spec
[ ] +0
[ ] -1 I have questions and/or concerns

Best,
Prashant Singh

Reply via email to