499 errors on pending.api production for semmeddb/query endpoint

[newgene]

We see quite some 499 errors on production nginx access logs, seems all come from the semmeddb/query endpoint:

 x.x.x.x - - [11/Mar/2024:07:01:50 +0000] "POST /semmeddb/query?with_total=true&fields=object.ncbigene,predication.pmid,pred
ication.sentence,subject.name,object.name&filter=predicate:PART_OF+AND+object.semantic_type_abbreviation:aapp+AND+_exists_:objec
t.ncbigene+AND+pmid_count:%3E3&size=1000 HTTP/1.1" 499 0 "-" "BTE/prod Node/v16.20.0 linux"

Let's investigate if anything we can do to reduce such errors

[colleenXu]

Not sure how relevant this is, but here's some info I recall from the on-site week:

• there are some records that are very large (lots of PMIDs, sentences) and it seems to take more time to retrieve them. Ex: retrieving 2 records with > 10000 PMIDs takes ~ 1 sec.
• I'm not sure that Translator/their UI effort is using the sentence-level info we provide (just asked in Translator Slack). I wonder if removing this data or cutting/reformatting it will improve performance by decreasing the size of records.

[ctrl-schaff]

@tokebe Any updates or insights into this?