api.get_dataframe (and therefore DataFrame.from_deeporigin) can take extremely long because:
2 calls to list_database_rows and describe_row take place in series when they could take place in parallel (this is a problem of 700ms vs 300ms)
(more serious problem). there is a place where N DescriebFile calls are made in series, where N is the number of files in the database. this means that the time it takes to fetch data grows with the number of files linearly, and the time for each file resolution is ~350ms.
proposed solution in this PR
using a new ListFiles endpoint that allows us to resolve all file IDs and names in a single API call (which typically takes 300ms no matter how many files)
screenshot
from cold start, this is how long it takes to display a dataframe. in the old version, this took ~10s
problem description
api.get_dataframe
(and thereforeDataFrame.from_deeporigin
) can take extremely long because:list_database_rows
anddescribe_row
take place in series when they could take place in parallel (this is a problem of 700ms vs 300ms)DescriebFile
calls are made in series, where N is the number of files in the database. this means that the time it takes to fetch data grows with the number of files linearly, and the time for each file resolution is ~350ms.proposed solution in this PR
using a new
ListFiles
endpoint that allows us to resolve all file IDs and names in a single API call (which typically takes 300ms no matter how many files)screenshot
from cold start, this is how long it takes to display a dataframe. in the old version, this took ~10s
rejected alternative
using async methods. this approach was tried in this PR: https://github.com/deeporiginbio/deeporigin-client/pull/73
it was rejected because it's much slower, and complicates testing dramatically