reading from (public bucket S3)

[womta]

I have a link in a public repository in AWS S3
https://s3-eu-west-1.amazonaws.com/stratominerdata/developer.stratominer.com/processedData/testingS3/default/testingS3_raw_plate_10_rep_1.fst which I can download anywhere

but when I try to load it in to memory directly from the link it fails... why?
read.fst('https://s3-eu-west-1.amazonaws.com/stratominerdata/developer.stratominer.com/processedData/testingS3/default/testingS3_raw_plate_10_rep_1.fst')

ERROR:
Error opening fst file for reading, please check access rights and file availability

I have an equivalent file in csv
which can be read in directly by
data.table::fread('https://s3-eu-west-1.amazonaws.com/stratominerdata/developer.stratominer.com/rawData/testingS3/default/file01.txt')

[MarcusKlik]

Hi @womta, thanks for reporting the issue!

Yes, fst cannot easily access an online file directly, mostly because it requires random access during reading and it will fail on trying to open the file (in the C++ backend) when a https URL is specified.

The data.table package is able to open the file because it first downloads the file to a temp location and then reads it from there (see this code). The curl::curl_download method is used to download the file. You could simulate this with your fst file:

tmp_file <- tempfile()
fst_url <- "https://s3-eu-west-1.amazonaws.com/stratominerdata/developer.stratominer.com/processedData/testingS3/default/testingS3_raw_plate_10_rep_1.fst"
curl::curl_download(fst_url, tmp_file, mode="wb")
ft <- fst::read_fst(tmp_file)

Obviously, you won't have random access, but you can still get the data.

I think a better error message would be useful when the user specifies a https URL as the path. Or, for complete loads, the same method could be used as with data.table, so leverage a full download with the curl package, what do you think?

[womta]

Hi @MarcusKlik

Thank you for the fast reply.
Is random access still required when the arguments 'columns', 'from' and 'to' are not specified in read_fst() i.e. it reads the full file anyway
Or does it still use random access?

Implementing the retrieval of the file before trying it and failing seems to be a nice improvement in the package...

Right now I use the aws.s3 package to store the file locally with save_object()

I noticed that s3:// links do work with fst but somehow with larger files it sometimes gives the exact same error...

[MarcusKlik]

Hi @womta, you're right, when all columns and rows are loaded, the fst file is read completely sequentially, so it could be downloaded in the same manner as fread() does. That would be a nice improvement to the package.

It's interesting to see that objects in s3 buckets can be accessed randomly using the Range specifier (see this code). That means that a cloud based fst file could in principle be read with random row- and column access by leveraging the aws.s3 package. Perhaps that's something to look at for a future enhancement, thanks!

[HugoGit39]

fread

Is this still the only method to read an url with a .fst file?