bug / known issue: inserts may be disordered even on one session

[fuyufjh]

Describe the bug

Here a random write channel will be picked.

risingwave/src/source/src/table.rs

Lines 82 to 87 in 314fcc7

	let (index, tx) = core
	.changes_txs
	.iter()
	.enumerate()
	.choose(&mut rand::thread_rng())
	.context("no available table reader in streaming source executors")?;

As a result, even in one session, DML statements may be reordered regarding the time that they leave the write channel.

To Reproduce

Just a technical analysis.

Expected behavior

Writes in one session should happen in order.

Additional context

No response

[lmatz]

#7134,

And another question is that whether we ignore an insert with the same pk
or let it overwrite the previous one
or both make sense so user decide

[fuyufjh]

#7134,

And another question is that whether we ignore an insert with the same pk or let it overwrite the previous one or both make sense so user decide

I once discussed with @st1page.

@st1page was considering adding a table-level option to control whether a new update should override. (upsert or ignore)

My idea is to always do upsert, similar to Flink's updatable stream.

[github-actions]

This issue has been open for 60 days with no activity. Could you please update the status? Feel free to continue discussion or close as not planned.

[st1page]

FYI: risingwavelabs/rfcs#48

[xxchan]

fixed by #14380