chore(deps): Bump lapis-silo to 0.2.2 to allow empty input in silo-preprocessing

[corneliusroemer]

GenSpectrum/LAPIS-SILO#433

https://empty-input.loculus.org

Works!

Let's get it merged @Taepper soon, congrats!

[Taepper]

You should retest now :)

[corneliusroemer]

Retriggering

[theosanderson]

You might want to delete the preview and recreate it

[theosanderson]

I still see 503 from https://lapis-empty-input.loculus.org/dummy-organism/sample/details is that what you expect because we haven't adjusted the thing that prepares the data to import?

[corneliusroemer]

My gut feeling was that this was expected? What works now is silo preprocessing, it no longer crashes. But you're right @theosanderson, this problem doesn't look entirely solved. SILO/LAPIS shouldn't 503 when there's no data available. It should simply do what one expects when there's no data. Return nothing.

The PR mentions:

In particular the ingest with an empty ndjson (which does not have keys anymore) should not fail. This was challenging, because we never want to read the ndjson files into a table as a whole, but only the non-sequence columns of it, and the sequence columns should be compressed while reading them

@Taepper can you enlighten us?

[theosanderson]

Are we actually preparing an empty dataset for SILO to import? My idea was that maybe we weren't, but if you say it was crashing before then I guess we were

[corneliusroemer]

Yeah it crashed and as a result we didn't update the output but discarded. Now it doesn't crash, so we use the output. But now SILO-proper crashes. I guess we've pushed the error down one step, but still way to go to LAPIS 🙃 First SILO, then LAPIS. It's possible that SILO can actually deal with the empty input and it's now a LAPIS bug. I don't know. @chaoran-chen / @Taepper might be able to tell :)

[theosanderson]

Approving as it seems ok to bump, regardless of this Q