haohuaijin commented on issue #6363: URL: https://github.com/apache/arrow-rs/issues/6363#issuecomment-2366788331
Hi @alamb, I think I found the reason after reading the code. While decoding the `RecordBatch`, we first construct an `ArrayReader`. Then, we call the `next_buffer` method in `ArrayReader` to get the `Buffer` needed to construct each `Array`. In `next_buffer`, we use `slice_with_length` to obtain a slice of the total `Buffer`. The code shows that the `data` in `Buffer` is shared among all [`Buffer`](https://github.com/apache/arrow-rs/blob/d05cf6d5e74e79ddcacaa4a68bddaba230b0f163/arrow-buffer/src/buffer/immutable.rs#L233) instances. When `get_array_memory_size()` is called on a `RecordBatch`, it calls `get_array_memory_size()` for each `Array`. Each `Array` then calls `capacity()` on each `Buffer`. Since `capacity()` refers to the total capacity of the `data`(the size of `RecordBatch`) rather than the capacity used by this `Array`, this has caused `get_array_memory_size()` to be very large(especially when there are many fields). https://github.com/apache/arrow-rs/blob/d05cf6d5e74e79ddcacaa4a68bddaba230b0f163/arrow-ipc/src/reader.rs#L562-L569 https://github.com/apache/arrow-rs/blob/d05cf6d5e74e79ddcacaa4a68bddaba230b0f163/arrow-ipc/src/reader.rs#L404-L406 https://github.com/apache/arrow-rs/blob/d05cf6d5e74e79ddcacaa4a68bddaba230b0f163/arrow-ipc/src/reader.rs#L51-L63 https://github.com/apache/arrow-rs/blob/d05cf6d5e74e79ddcacaa4a68bddaba230b0f163/arrow-buffer/src/buffer/immutable.rs#L223-L237 https://github.com/apache/arrow-rs/blob/d05cf6d5e74e79ddcacaa4a68bddaba230b0f163/arrow-buffer/src/buffer/immutable.rs#L166-L168 -- 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]
