Size Change: +4.3 kB (0%)

Total Size: 1.2 MB

_{compressed-size-action}

This looks great to me!

Chunks processing seems to be one step too far as they make the entire transaction not atomic.

Could you expand what this means?

Chunks processing seems to be one step too far as they make the entire transaction not atomic.
Could you expand what this means?

@TimothyBJacobs sure! The words "transaction" and "commit" usually characterize an atomic operation that either goes through or fails in its entirety. In this PR, however, a transaction may consists of multiple chunks which means it may partially fail so I'm not entirely comfortable with this vocabulary.

I'm not saying we shouldn't do chunks - that's a useful automation. If the caller needs to save 21 entities and a single batch may only handle 20, then it's going to send two batch requests anyway. Handling that logic manually would be repetitive and prone to errors so it's good to have it baked in. And there transactions out there that may end up in a partially failed state too (MyISAM - I'm looking at you).

I think I would be more comfortable with this if the chunks were called "transactions" and transactions were called something else. This way each transaction would be atomic, but the collection of transactions wouldn't.

There we go, I updated the nomenclature to "batch" and "transaction".

One issue I'm running into is that the batch we return to the caller takes the ordered list of items in a queue, and returns them as an object. Object insertion order isn't guaranteed. This means that if the caller doesn't have the itemId returned by enqueueItem, which in nearly all cases it won't. They can't match up their apiFetch calls with the results returned by processBatch.

Keying the results and errors by the itemId makes a lot of operations convenient, so maybe instead of reverting that to an array, we maintain a property on the batch of the itemIds in insertion order?

@adamziel would you mind explaining the public API of the proposed package in the description or the README?