 |
 |
If a s3 plugin to XrdHTTP is a better architectural choice (seems nobody disagree), this is OK for us. Keeping up with the s3 development is a concern. I think we need to agree among ourselves that at this stage, we are only talking about basic functions, and low hanging fruits. In addition to Rubin, I see the python ecosystem in analysis facility side is picking up momentum. They inevitably run into object stores. Currently they are deploying a MinIO. This is a thing we can help them eliminating if they can write the Xrootd, EOS and dCache via s3. -- Wei ________________________________________ From: Andrew Hanushevsky <[log in to unmask]> Sent: Monday, April 11, 2022 12:35 PM To: Fabrizio Furano Cc: Yang, Wei; xrootd-l Subject: Re: Feature request on XrdHTTP I agree with Fabrizio. XrdHttp should really be focused on http, though we've alreay extended that to some degree but only on the request side. Responses are still totally http. Any further extensions beyond that should be relegated to a plugin. The code is complicated enough and anything further would likely make it unmaintainable. Andy On Mon, 11 Apr 2022, Fabrizio Furano wrote: > Hi Wei, > > yes, I see, and I have a good understanding of S3, coming > from the evolution of Davix. > > S3 is built on HTTP, yes, however it adds many bits and pieces, as > you discovered. > > IMO the best solution to this is to make an XrdHttp plugin > in order to override its behaviour towards being S3-ish. Making > the plugin is pretty easy, the interface is rather simple, > what would become heavier IMO is following up with > the feature requests that inevitably will come when advertising > S3 pragmatic compatibility. > > Fabrizio > > Il 11/04/2022 10:52, Yang, Wei ha scritto: >> Hi Fabrizio, >> >> I am trying to understand how far away is XrdHTTP from a S3 object store. I >> know XrdHttp will work with 'requests', etc. but by far the most popular >> object store is s3. Rubin is one of the experiments that widely uses s3 >> (they claim they support webdav as well). They don't use all the features >> of s3 but a subset of basic functions. Things like ACL and multipart upload >> are mentioned but not in high priority. >> >> I am not sure I fully understand the difference between HTTP(s) and S3. I >> thought that S3 is built on top of HTTPs. >> >> So I am trying boto3 against XrdHTTP. It both surprises and not surprise >> to me that some of the things are actually working. e.g.download_file(), >> upload_file() (if I remove the emoji from the XrdHTTP response), >> get_object(), delete_object(), delete_bucket(). etc. >> >> There are things that doesn't work or need twisting. AA is one, which I >> worked around by using a macaroon token. Listing objects/buckets is another >> one because boto3 expect some kind of XML (probably specified in S3 >> protocol). >> >> XrdHTTP can identify access via boto3 api by checking the user-agent. The >> following is what I see in the log file: >> >> http_Protocol: rc:91 got hdr line: User-Agent: Boto3/1.12.49 Python/3.6.8 >> Linux/3.10.0-1160.49.1.el7.x86_64 Botocore/1.15.49 >> >> I don't know how far we can and should go along this route. I am assessing >> what is possible, and what is not. >> >> ?-- >> Wei? >> >> ________________________________________ >> From: Fabrizio Furano <[log in to unmask]> >> Sent: Monday, April 11, 2022 12:54 AM >> To: Yang, Wei; xrootd-l >> Subject: Re: Feature request on XrdHTTP >> >> Hi Wei, >> >> could you please be more specific about the requests that should get >> an XML? >> >> I would prefer to keep the browsers working against XrdHttp >> (which I think it's a great feature), so I would not just substitute >> HTML with XML >> >> Can these requests that need XML be recognized through their headers? >> >> Also, boto3 uses the AWS protocol, while XrdHttp is ... HTTP. Does not >> seem the perfect match to me. >> >> Maybe we can think at a plugin that provides basic S3 >> compliance instead than adding tweaks and special cases? >> >> >> Cheers >> Fabrizio >> >> >> >> Il 09/04/2022 00:00, Yang, Wei ha scritto: >>> I am playing with boto3 with XrdHTTP. After a few hack to pass a Macaroon >>> to xrootd I can get boto3 to download a file from xrootd (I guess I can >>> use scitoken too). But list_objects() and list_buckets() do not appreciate >>> the HTML returned by the servers. They want XML. Could this be possible? >>> >>> ?-- >>> Wei? >>> >>> ######################################################################## >>> Use REPLY-ALL to reply to list >>> >>> To unsubscribe from the XROOTD-L list, click the following link: >>> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-L&A=1 > > ######################################################################## > Use REPLY-ALL to reply to list > > To unsubscribe from the XROOTD-L list, click the following link: > https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-L&A=1 > ######################################################################## Use REPLY-ALL to reply to list To unsubscribe from the XROOTD-L list, click the following link: https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-L&A=1