Not really our fault, but we can still try to handle this case better.
This occurred with the NCI60 Drug Dataset pulling from Zenodo and getting a 503 Gateway error. The recently implemented retry method (on the build 2.2 branch) does not work for this case due to the fact that the PubChem script creates an ignore_chems list for all drugs that received an error. To fix this, I am updating the pubchem_retrieval script to only add drugs that receive a 404 error to that list. I'm also adding a delay on the build_all.py retries for 3 minutes and 10 minutes between the 2nd and 3rd attempts.