[Pulp-list] Download analytics from CDN?

[David Davis]

Hi Danny,

I don't know much about AWS logging but Pulp does set the filename in the
response-content-disposition[0]. Could that be used to determine the
filename for each request?

If not, I'm looking at the boto3 docs for get_object[1] to see if there's
another parameter we could set to help you track the filename in requests
but I'm seeing anything useful. My knowledge of s3 is a bit limited so if
you have a suggestion how we can construct a request to S3 that would help
you to track the filenames of requests to s3, I could probably look at how
we could support it in Pulp 3.

[0]
https://github.com/pulp/pulpcore/blob/f38f955425b185749b3c8d4d878a7e166cfc05b9/pulpcore/content/handler.py#L613-L614
[1]
https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/s3.html#S3.Client.get_object

David


On Tue, Mar 30, 2021 at 10:43 AM Danny Sauer <danny.sauer at konghq.com> wrote:

> I've got Pulp set up to serve all the content from S3 behind CloudFront.
> This works really well, except for a minor issue: the content URLs are all
> the UUIDs for artifacts, not, for example, the pretty name of the RPM being
> downloaded.  That's an issue in my situation because we'd really like to
> generate download analytics using off-the-shelf tools which consume the AWS
> CDN standard log format.
>
> My initial thought was that it might be easy to have the redirects include
> a query string in the generated URL which notes the original filename or
> relative path requested.  But I don't have sufficiently developed Django
> skills to know the easiest way to do that (or if it's even reasonable to
> think that's easy).  Using the content server's logs is another option, but
> I have some other content on the same S3 bucket which may not necessarily
> be reached solely through Pulp's content server, so that means two log
> locations, etc.  If it was easy to make Django / Gunicorn log to an S3
> bucket in a manner similar to Cloudfront, that might also be ok.
> Post-processing logs with a series of API calls to work out what artifact
> maps to what repository content would ideally be a last resort.
>
> Anyone have some great insights which might help me out here? :)  If it
> helps, I'm building my own Docker images which ultimately run in EKS.  So
> patches / extra modules are an option, but I'd prefer to stay as close to
> vanilla upstream as possible with environment variable-based config
> adjustments.
>
> Thanks.
> --Danny
> _______________________________________________
> Pulp-list mailing list
> Pulp-list at redhat.com
> https://listman.redhat.com/mailman/listinfo/pulp-list
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/pulp-list/attachments/20210406/e8e7523c/attachment.htm>