@jamshale it's still in progress (I can't reopen it) -- still need to make a PR for connection / cred ex / presentation records. And adding support for ordering in askar

Potential endpoints that may need pagination as well:

• listing created rev regs, in the revocation API
• listing transaction records, in the endorse_transaction API
• listing mediation requests, in coordinate_mediation

Those are the only other ones I can see, with potentially large responses

I've discovered a bug with pagination + additional post-filter query parameters.

e.g. for fetching connection records, specifying limit=1 and alias="some alias", then it will return 0 records, despite that alias appearing in some records (i.e. it'll return relevant records when limit is set higher). Post-filtering is being done after the limited fetch, which messes up expected output

This impacts:

• fetching v1 or v2 credential or presentation exchange records, with query params: connection_id, role, state
• fetching connection records with: alias, state, their_role, connection_protocol

These are all the post-filter query params for the routes that now have paginated query support. So, limit + offset will not work as expected when using these query params

A workaround is to ignore limit/offset when these query params exist, and to only apply limit/offset after the post filtering ... this would resolve the expected behavior, but there will still be a performance/scalability cost when there are potentially 1000s of records.

@jamshale Do you think this is a reasonable workaround to implement? Behavior <= 0.12.1 was anyway fetching all records. So, while it's not ideal to apply limit/offset after post-filtering, it's still better than having no pagination support ...

@ff137 Yes, I think that's the correct way forward for now. I've never understood why these post filters have to be applied the way they do but I think that's good enough as is and we can try and address it in the future.

Agreed. It's of course highly inefficient -- loading each record result as json and post filtering on fields (edit: I see the json deserialization is necessary either way, nvm). I'll do the workaround for now, but will also take a look at askar library for how this may be improved

I'm working on the following issue in Askar to add support for custom ordering: openwallet-foundation/askar#290

@swcurran @jamshale I think this would be an essential feature for the 1.0.0 release, so that paginating through storage records guarantees that missing or duplicate records aren't returned across pages.

Thanks for taking this on. This is an important feature.

My only concern about having it in the 1.0.0 release is that it keeps getting delayed. We really want to get it out and have a more steady release cadence.

This pagination still works correctly, correct? It's just extremely inefficient with the post filtering?

That's all good. Writing tests for the new feature might take some time too. Is the idea to release 1.0.0 ASAP? If there's a rough date set, then I can try get this ready by then. If not, I'll just see how far I get!

Also yes, the pagination works -- I've just run into trouble when trying to assert that each page returns distinct results, with no duplicates or missing records. I managed to create 100'000 wallets, and fetch 1000 at a time, asserting each page has distinct results. So that was successful. But then when fetching connection records with post-filter query params, I couldn't get the same guarantees. Creating new connections, or new wallets, might cause some shuffling. Difficult to pin down exactly what the cause or frequency of the problem is. But all I can say is there's currently no "guarantee" that scanning pages, by increasing the offset, will always return distinct results. Having an ordering option will provide guarantee / more control for the end user.

We don't have a date. We want to release ASAP.

I think this problem is acceptable to have in the release. It would make a good release milestone once it's completed.