Edge case/failure handling in /search

[jsstevenson]

Currently, the /search endpoint takes an accession ID and a start/end and returns variants contained within (overlapping or fully contained, i can't remember -- that's a separate issue).

First, if accession_id doesn't start with ga4gh:, then we try to use seqrepo to translate it to a known GA4GH accession ID (i.e. something starting with ga4gh:SQ.. If the arg DOES start with ga4gh:, then we just proceed with a search for all variants where the refgetAccession ID equals that value. This is from main.py on the storage refactor branch:

anyvar/src/anyvar/restapi/main.py

Lines 606 to 626 in fb12671

	try:
	if accession.startswith("ga4gh:"):
	ga4gh_id = accession
	else:
	ga4gh_id = av.translator.get_sequence_id(accession)
	except KeyError as e:
	raise HTTPException(
	status_code=HTTPStatus.NOT_FOUND,
	detail="Unable to dereference provided accession ID",
	) from e
	alleles = []
	if ga4gh_id:
	try:
	refget_accession = ga4gh_id.split("ga4gh:")[-1]
	alleles = av.object_store.search_alleles(refget_accession, start, end)
	except NotImplementedError as e:
	raise HTTPException(
	status_code=HTTPStatus.NOT_IMPLEMENTED,
	detail="Search not implemented for current storage backend",
	) from e

Some behavior I've noticed, which varies between correct, weird, and probably wrong:

• Using a bogus or unrecognized sequence alias, such as "zzzzz" or "SQ.zzzzz" returns a 404 NOT FOUND with {"detail": "Unable to dereference provided accession ID"}. I think that's not quite right, it would be one thing to return 404 when a requested resource is not found, but here the 404 seemingly indicates that GET /search doesn't exist, which is not true. ❌
• Using a correct sequence ID but forgetting "ga4gh:" returns the same 404 thing btw 🙄
• Using a known, GA4GH-prefixed sequence ID or otherwise using a known sequence alias (e.g. "ga4gh:SQ.8_liLu1aycC0tPQPFmUaGXJLDs5SbPZ5", or "GRCh38.p1:Y") returns a 200 OK with body {"variations": []}. ✅
• Using a valid GA4GH-prefixed sequence ID that isn't actually a real sequence (ie, try changing a single character on a known ID) returns the 200 OK response. 🤔
• Using an invalid GA4gh sequence ID, such as "ga4gh:ZZZZZ" returns a 200 OK with body {"variations": []}. 🤔

I think there are a few distinct kinds of failure modes (or ways to handle flawed inputs) and it'd be nice to communicate them to the user or handle them appropriately:

• Unknown sequence alias (e.g. "GRCh38383838.p1:Y")
• Correct sequence ID but forgot ga4gh: (e.g. "SQ.8_liLu1aycC0tPQPFmUaGXJLDs5SbPZ5")
• Invalid ga4gh sequence ID (e.g. wrong # of characters) (e.g. "ga4gh:SQ.12345")
• Valid-appearing, but unrecognized sequence ID (e.g. "SQ.8_liLu1aZZZZZZZZZZZZZZZZZZZZZZZZ")
• Valid-appearing and recognized sequence ID, but no variants in provided interval (i.e. you just searched for a specific interval that's immediately adjacent to, but not quite inclusive of a variant stored in the DB)