alamb commented on issue #7744: URL: https://github.com/apache/arrow-datafusion/issues/7744#issuecomment-1749435395
I am not sure about extending the `RecordBatchStream` -- given we don't know
what partition column values are in the input it will not be possible to know
how many output files (one for each partition that has data) are needed before
the plan is run.
I think actually writing to a partitioned datasource will require a more
dynamic approach, with something similar to a `RepartitionExec` that
dynamically creates a FileSink whenever new values are seen. I tried to sketch
out what this might look like here:
```
rows from the
input batch
that belong to
partition 1 ─ ─ ─ ─ ─ ┐
│
▼
┌──────┐ Partition 1
┌──────▶│Batch │ (date=2023-10-01)
│
└──────┘
│
┌──────┐
┌────────────┐
├──────▶│Batch │ Partition 2
┌───────────┐ ┌────────────┐ │ │▒ │
└──────┘ (date=2023-10-02)
│ batch N │────▶...────▶ Batch 1 │────────▶│ Demux ? │──────┤ ...
└───────────┘ └────────────┘ │ │▒ │
└────────────┘▒ └──────▶
▲
▒▒▒▒▒▒▒▒▒▒▒▒▒▒
└ ─ ─ ─ ─ ─
│
Input batches are repartitioned
│
according to partition column's
As new partition values are
values with a "Demux" operator (like
seen, new output streams are
RepartitionExec)
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
