Query graph giving a Bad Gateway response

[saramsey]

At timestamp 2026-04-06T18:18:13.311 UTC, the ARAX CI/CD system (cicd.rtx.ai), while running the standard ARAX series of unit tests, ran the test test_ARAX_workflows.py::test_FET_example_3, which POSTed the following TRAPI message (with a timeout of 120)

{
  "message": {
    "query_graph": {
      "nodes": {
        "n02": {
          "ids": [
            "MONDO:0021661",
            "MONDO:0004995",
            "MONDO:0005311",
            "UMLS:C0011847",
            "MONDO:0005044"
          ],
          "categories": [
            "biolink:Disease",
            "biolink:PhenotypicFeature",
            "biolink:DiseaseOrPhenotypicFeature"
          ],
          "is_set": true
        },
        "n03": {
          "categories": [
            "biolink:Protein",
            "biolink:Gene"
          ],
          "is_set": true
        }
      },
      "edges": {
        "e02": {
          "predicates": [
            "biolink:related_to"
          ],
          "subject": "n02",
          "object": "n03",
          "exclude": false
        }
      }
    }
  },
  "submitter": "infores:arax",
  "parameters": {
    "tiers": [
      1
    ]
  }
}

to the API endpoint https://dev.retriever.biothings.io/query. After 17.45 seconds (well short of the timeout), the unit test received back from the infores:retriever API a 502 HTTP status code in the response, with the message "Bad Gateway". The ARAX team is wondering, is this query graph invalid in some way? This error seems to be inconsistent, i.e., sometimes the query graph gets a 200 response and sometimes it gets a 502 response.

[saramsey]

We are wondering if there were any maintenance or other issues going on with Retriever's dev endpoint, around the time of the above UTC timestamp?

[tokebe]

Nothing wrong with the query graph (aside from some things Retriever won't use like is_set and exclude, but those only produce warnings)

502 Bad Gateway indicates we may have had some dev instability. Retriever as an application never producedes this code, but nginx produces it when Retriever goes down.

I'll take a look into Retriever Dev's stability over the last couple days.

[saramsey]

Thanks, Willow. This is helpful info.

[tokebe]

Retriever didn't totally go down, but we've definitely had some workers die, I'll be investigating more.

[saramsey]

Thank you, Willow. I really appreciate the Retriever team's responsiveness.