Streaming Google storage upload

[Fonsan]

I am trying to solve the case of uploading files in a streaming fashion without keeping the full contents in memory

I have previously implemented the same feature in aws/aws-sdk-ruby#1711

As identified in googleapis/google-cloud-ruby#1997
There are some limitations in google-api-client which hard codes the use of the content length header

I have written a minimal wrapper library that monkey patches it here which should serve as a good source of inspiration: https://github.com/Fonsan/gcs_stream_upload

require 'gcs_stream_upload'

storage = Google::Cloud::Storage.new(timeout: 5 * 60)
bucket = storage.bucket('bucket')
gcs_stream_upload = GCSStreamUpload.new(bucket)

gcs_stream_upload.upload('object') do |io|
 IO.copy_stream(IO.popen('yes | head -n 10'), io)
end
# => Written "y\n" * 10


gcs_stream_upload.upload('object') do |io|
 io << 'data'
 io << 'dat2'
end
# => Written "datadat2"

[quartzmo]

@Fonsan Thank you for your patience with this request! I finally got the chance today to run https://github.com/Fonsan/gcs_stream_upload and the examples work for me.

[quartzmo]

@Fonsan Can you see a backward-compatible way to add this functionality to googleapis/google-api-ruby-client? I think that might be the first step toward incorporating it into google-cloud-storage. Even then, however, I'm not sure if other considerations might prevent adding this feature, for example the use of #rewind as mentioned here by @blowmage.

[Fonsan]

@quartzmo Given the way request_header works in #send_start_command; it fails even if we would stub the #size method on objects that do not respond to size when passed and then stub #to_s we would still end up in a scenario where UPLOAD_CONTENT_LENGTH => nil. This still sets the header and the GCS backend service fails in that scenario currently.

What I am describing above is a bit of a hack but I would really rather see the change in

google-api-ruby-client/lib/google/apis/core/upload.rb

Line 173 in 711dfb8

def send_start_command(client)

Is there a specific reason why google-api-ruby-client could not be extended to allow for IO objects that do not respond to size? @blowmage

BTW @ https://www.kaiko.com/ we are running the gem above in production until a cleaner alternative comes around.

[quartzmo]

Is there a specific reason why google-api-ruby-client could not be extended to allow for IO objects that do not respond to size?

Should this conversation be moved to a new issue on googleapis/google-api-ruby-client ?

[quartzmo]

@Fonsan If it's OK with you, I will transfer this issue to googleapis/google-api-ruby-client for further discussion.

[Fonsan]

@quartzmo any efforts to further the progression would be excellent :)

[dazuma]

@quartzmo I'll let you investigate the feasibility of this.

[quartzmo]

@blowmage Any comment on this?

Is there a specific reason why google-api-ruby-client could not be extended to allow for IO objects that do not respond to size?

[blowmage]

I believe HttpClient is checking for size. But yes, google-api-client could be rewritten to use something other than HttpClient and not check File#size.

[ianks]

Judging from the documentation, Content-Length header is not required. Seems like it may be a straightforward fix. @quartzmo do you still plan on transferring the issue?

[quartzmo]

quartzmo transferred this issue from googleapis/google-cloud-ruby on Jan 29, 2019

[tinco]

hi @quartzmo are you actively working on the streaming support on this? I noticed this issue has no one assigned to it right now

[dazuma]

It's not being actively worked on right at this point. Likely this will require first completing #2348 (which is on my plate but I won't get to it for a few more weeks).

[simi]

No note this is related as well - googleapis/google-cloud-ruby#8235, any news? Is PR for #2348 welcomed?

[simi]

@dazuma is there anything I can help with? Would it be welcomed to push #2348 forward?

[simi]

:'( no progress? Anything I can help with here?

[simi]

@quartzmo @frankyn @dazuma It is 2023 and this is still not part of the official library if I understand it well. 😢