Loading ...
Sorry, an error occurred while loading the content.

Re: [soaplite] Re: How to handle big soap attachments ?

Expand Messages
  • Paul Kulchenko
    Hi, Sebastian! That s true, but at the same time it s easy to imagine situation when you send something not directly, but thru the several different
    Message 1 of 13 , Mar 22, 2001
    • 0 Attachment
      Hi, Sebastian!

      That's true, but at the same time it's easy to imagine situation when
      you send something not directly, but thru the several different
      intermediaries and each of them will need to handle this huge
      request. If this piece is encoded as external reference then handler
      could be smart enough to get it only if it's required (yet I don't
      know about such smart handlers :)). Ideas, ideas...

      Ideally implementation should be flexible enough to handle both (and
      maybe man others) approaches, maybe with manual hints.

      Best wishes, Paul.

      --- sebaklu@... wrote:
      > Hi Petr.
      >
      > Since there exists a specification for soap messages with
      > attachements it should be possible to send the entire data
      > according
      > to the request at once. Why make it complicated with more requests
      > ?
      > My problem is that the client could send only data via HTTP and the
      >
      > server can receive it only via HTTP. The server themself can't send
      >
      > any request for requiered data to the client. Firewalls and
      > security
      > rules on both sides make other requests eg. via URLs or FTP
      > impossible .
      > I think SOAP is a good solution for this problem.
      >
      >
      > Sebastian
      >
      > --- In soaplite@y..., "Petr Janata" <petr.janata@i...> wrote:
      > > Hello,
      > >
      > > I would just like to remark that we had Don Box last week in
      > Prague
      > giving a
      > > SOAP talk and he said that SOAP is meant to work exactly like
      > this, i.e.
      > > not to handle large amounts of data. He also said that if you
      > need
      > to do
      > > that you can e.g. pass just the URL in a SOAP message and use
      > simple
      > > download (or SAX parser ) to handle the transfer.
      > >
      > > Petr Janata
      > >
      > > -----Original Message-----
      > > From: sentto-2738395-120-985195300-petr.janata=i.cz@r...
      > > [mailto:sentto-2738395-120-985195300-petr.janata=i.cz@r...]On
      > > Behalf Of allierogers@y...
      > > Sent: Wednesday, March 21, 2001 6:22 PM
      > > To: soaplite@y...
      > > Subject: [soaplite] Re: How to handle big soap attachments ?
      > >
      > >
      > > Paul,
      > >
      > > I have seen a similar problem with ALL SOAP (and XML-RPC)
      > > implementations. That is that they do not handle streams at all.
      > > Every implementation (that I know of) loads the entire HTTP
      > > request/response into memory (bad idea, what if it's a MPEG movie
      > or
      > > something), parses it into an in-memory tree, and then hands it
      > off
      > > for proper dispatch.
      > >
      > > Why do implementations assume that all methods accept 2 simple
      > > arguments and return 1 simple argument like all of the examples
      > (for
      > > instance, "getQuote")? The real world is not like that.
      > >
      > > In the real world, there may be many complicated arguments (like
      > a
      > > file upload) and the response may also be arbitrarily large
      > (e.g., a
      > > SQL result set of 100,000 rows).
      > >
      > > HTTP, SOAP, XML-RPC and XML can handle this fine, if the
      > > implementations were a bit smarter.
      > >
      > > - Parse using SAX or stream parsers, only (never DOM), and always
      > > parse "on the fly" without loading entire stream into RAM.
      > >
      > > - Always stream to/from HTTP, rather than loading entire
      > > request/response into RAM.
      > >
      > > Since HTTP is the most transport, do what good HTTP servers do.
      > When
      > > you browse to a site with streaming video, do you think the HTTP
      > > server loads the entire video file into RAM for each request? Of
      > > course not. It streams it off disk (or cache) directly back to
      > the
      > > client, chunk by chunk, so that only a small amount of a large
      > file
      > > is ever in memory at once. In that way, the HTTP server can
      > handle
      > > 100,000 simulatenous hits to that same streaming video.
      > >
      > > Am I wrong about this?
      > >
      > > Regards,
      > >
      > > Allie Rogers
      > >
      > > --- In soaplite@y..., Paul Kulchenko <paulclinger@y...> wrote:
      > > > Hi, Sebastian!
      > > >
      > > > It might, thanks for the tip. Won't make any promisses, but
      > > > definitely will take a look. Next version is about to be
      > released,
      > > > and I don't think will include any changes in this aspect
      > (unless
      > > > they are minimal, that is possible also), but I'll try to do it
      > > ASAP.
      > > > Thank you.
      > > >
      > > > Best wishes, Paul.
      > > >
      > > > --- sebaklu@y... wrote:
      > > > > Hi, Paul!
      > > > >
      > > > > Thanx for the answer.
      > > > >
      > > > > In fact the memory usage in the CGI package
      > > (SOAP::Transport::HTTP)
      > > > >
      > > > > is a problem for handling big SOAP requests because the
      > complete
      > > > > content is kept more than one times in the memory.
      > > > >
      > > > > extract from package SOAP::Transport::HTTP::CGI
      > > > > located in SOAP/Transport/HTTP.pm:
      > > > >
      > > > > my $content; read(STDIN,$content,$ENV{'CONTENT_LENGTH'} ||
      > 0);
      > > > > # ^^^^^^^^--- first time
      > > > >
      > > > > $self->request(HTTP::Request->new(
      > > > > $ENV{'REQUEST_METHOD'} || '' => $ENV{'SCRIPT_NAME'},
      > > > > HTTP::Headers->new(map {(/^HTTP_(.+)/i ? $1 : $_) => $ENV
      > > {$_}}
      > > > > keys %ENV),
      > > > > $content,
      > > > > # ^^^^^^^^- second time
      > > > > ));
      > > > >
      > > > > Will future version support stream-based handling directly
      > from
      > > > > STDIN?
      > > > >
      > > > > --- In soaplite@y..., Paul Kulchenko <paulclinger@y...>
      > wrote:
      > > > > > Hi, Sebastian!
      > > > > >
      > > > > > Yes, you may specify option for MIME::Parser to keep
      > temporary
      > > > > file
      > > > > > on disk during parsing, but presence it in memory will be
      > > > > required
      > > > > > later to parse the message, though it shouldn't take 200MB
      > to
      > > > > parse
      > > > > > it. As soon as it's one part you should be fine. Anyway,
      > there
      > > is
      > > > >
      > > > > not
      > > > > > much you can change (except options for MIME::Parser) and
      > I'll
      > > do
      > > > > my
      > > > > > tests to check it.
      > > > > >
      > > > > > Best wishes, Paul.
      > > > > >
      > > > > >
      > > > > >
      > > > > > __________________________________________________
      > > > > > Do You Yahoo!?
      > > > > > Get email at your own domain with Yahoo! Mail.
      > > > > > http://personal.mail.yahoo.com/
      > > > >
      > > > >
      > > > > ------------------------ Yahoo! Groups Sponsor
      > > > >
      > > > > To unsubscribe from this group, send an email to:
      > > > > soaplite-unsubscribe@y...
      > > > >
      > > > >
      > > > >
      > > > > Your use of Yahoo! Groups is subject to
      > > > > http://docs.yahoo.com/info/terms/
      > > > >
      > > > >
      > > >
      > > >
      > > > __________________________________________________
      > > > Do You Yahoo!?
      > > > Yahoo! Auctions - Buy the things you want at great prices.
      > > > http://auctions.yahoo.com/
      > >
      > >
      > >
      > > To unsubscribe from this group, send an email to:
      > > soaplite-unsubscribe@y...
      > >
      > >
      > >
      > > Your use of Yahoo! Groups is subject to
      > http://docs.yahoo.com/info/terms/
      >
      === message truncated ===


      __________________________________________________
      Do You Yahoo!?
      Get email at your own domain with Yahoo! Mail.
      http://personal.mail.yahoo.com/
    • allierogers@yahoo.com
      Paul, You make many good points, and I see the issues. Maybe we need to decompose the problem into smaller spaces. As I see it, there are 2 basic issues
      Message 2 of 13 , Mar 22, 2001
      • 0 Attachment
        Paul,

        You make many good points, and I see the issues. Maybe we need to
        decompose the problem into smaller spaces. As I see it, there are 2
        basic issues related to large SOAP RPC calls:

        1. SOAP request where arguments are large

        2. SOAP response where method result is large

        I thought the proper way to handle number 1 was to use the HTTP file
        upload capability where the SOAP argument references the data in the
        HTTP upload or via URL somehow. Maybe I am wrong about this because
        my particular implementations never make use of this feature. I
        realize this is an HTTP-centric approach, but the scalability issues
        really revolve around HTTP implementations and not so much in SMTP
        and FTP where disk rather than RAM storage is the default and
        dispatch and execution threads are not as memory sensitive as HTTP
        (for instance, an SMTP server is only handling one message at a time
        while an HTTP server may be handling 100,000 requests at once).

        For me, number 2 is more important. In this case, the method call is
        simple and its data is relatively small, so how you parse and handle
        the request on the server side is fine as it is. However, the
        response should be streamed in all cases. There is no issue of XML
        parsing on the server side in this case. But, the server should
        assume that the method return could be arbitrarily large, so it
        should not attempt to receive the entire method return before it
        starts passing it back to the client. The server should start, right
        away, streaming back the SOAP envelope, unbuffered, or at least allow
        this as a setting. For SOAP::Lite, you have the problem that you are
        dynamically trying to figure out what types are in the method
        response, rather than through static definition (e.g., a
        configuration file on the server to map method name, namespace, uri,
        arugments, return types, etc.). Without a static configuration,
        there is no way to stream as I would like. Maybe it could be a
        performance option?

        This is how we have solved the problem, here, both for SOAP and XML-
        RPC. However, we use SOAP::Lite servers only for some prototyping.
        In other cases, we have COM-based SOAP/XML-RPC servers (e.g., 4s4c
        from pocketsoap.com) and IIS configured for unbuffered response.

        Regards,

        Allie
      • allierogers@yahoo.com
        Petr, I know Don is an important figure, but I disagree with him. SOAP can be used for large data, just as all HTTP servers today handle large data. We use
        Message 3 of 13 , Mar 22, 2001
        • 0 Attachment
          Petr,

          I know Don is an important figure, but I disagree with him. SOAP can
          be used for large data, just as all HTTP servers today handle large
          data. We use it for this purpose here in all of our products and it
          works well. Maybe he did not intend this use of SOAP, but it does
          work. However, SOAP::Lite, currently, can not be used in this way.
          Perhaps that may change.

          Allie

          --- In soaplite@y..., "Petr Janata" <petr.janata@i...> wrote:
          > Hello,
          >
          > I would just like to remark that we had Don Box last week in Prague
          giving a
          > SOAP talk and he said that SOAP is meant to work exactly like
          this, i.e.
          > not to handle large amounts of data. He also said that if you need
          to do
          > that you can e.g. pass just the URL in a SOAP message and use simple
          > download (or SAX parser ) to handle the transfer.
          >
          > Petr Janata
        • sebaklu@yahoo.com
          Hi Paul, That s right, but the current version of SOAP::Lite should never expect large requests. The server will give no response and fill out the complete
          Message 4 of 13 , Mar 22, 2001
          • 0 Attachment
            Hi Paul,

            That's right, but the current version of SOAP::Lite should never
            expect large requests. The server will give no response and fill out
            the complete memory on the machine. Since it expect SOAP messages
            with attachements it should be able to handle large amount of data.

            However, it works fine with simple requests. But general for
            handling SOAP messages with attachments (the 7 MB attachment was an
            example, i had also problems to handle smaller attachments) should
            use stream mechanism. If not you should not read it into memory but
            reject the request. Maybe i'm wrong but it is a weak point and DOS
            attacks may use it.


            Sebastian

            --- In soaplite@y..., Paul Kulchenko <paulclinger@y...> wrote:
            > Hi, Sebastian!
            >
            > That's true, but at the same time it's easy to imagine situation
            when
            > you send something not directly, but thru the several different
            > intermediaries and each of them will need to handle this huge
            > request. If this piece is encoded as external reference then handler
            > could be smart enough to get it only if it's required (yet I don't
            > know about such smart handlers :)). Ideas, ideas...
            >
            > Ideally implementation should be flexible enough to handle both (and
            > maybe man others) approaches, maybe with manual hints.
            >
            > Best wishes, Paul.
          • Paul Kulchenko
            Hi, Sebastian! Absolutely agree. That s the reason why I want to introduce some additional transport options, like ACCEPTABLE_CONTENT_TYPE (if you want to
            Message 5 of 13 , Mar 22, 2001
            • 0 Attachment
              Hi, Sebastian!

              Absolutely agree. That's the reason why I want to introduce some
              additional transport options, like ACCEPTABLE_CONTENT_TYPE (if you
              want to accept ONLY text/xml or multipart/related) and
              MAX_CONTENT_SIZE that should take care about it and request will be
              rejected. As for DOS attack it could be introduced even with small
              request which has complex XML structure. Anyway, these options should
              make server side more robust.

              Best wishes, Paul.

              --- sebaklu@... wrote:
              > Hi Paul,
              >
              > That's right, but the current version of SOAP::Lite should never
              > expect large requests. The server will give no response and fill
              > out
              > the complete memory on the machine. Since it expect SOAP messages
              > with attachements it should be able to handle large amount of data.
              >
              > However, it works fine with simple requests. But general for
              > handling SOAP messages with attachments (the 7 MB attachment was
              > an
              > example, i had also problems to handle smaller attachments) should
              > use stream mechanism. If not you should not read it into memory but
              >
              > reject the request. Maybe i'm wrong but it is a weak point and DOS
              > attacks may use it.
              >
              >
              > Sebastian
              >
              > --- In soaplite@y..., Paul Kulchenko <paulclinger@y...> wrote:
              > > Hi, Sebastian!
              > >
              > > That's true, but at the same time it's easy to imagine situation
              > when
              > > you send something not directly, but thru the several different
              > > intermediaries and each of them will need to handle this huge
              > > request. If this piece is encoded as external reference then
              > handler
              > > could be smart enough to get it only if it's required (yet I
              > don't
              > > know about such smart handlers :)). Ideas, ideas...
              > >
              > > Ideally implementation should be flexible enough to handle both
              > (and
              > > maybe man others) approaches, maybe with manual hints.
              > >
              > > Best wishes, Paul.
              >
              >
              >
              > ------------------------ Yahoo! Groups Sponsor
              >
              > To unsubscribe from this group, send an email to:
              > soaplite-unsubscribe@yahoogroups.com
              >
              >
              >
              > Your use of Yahoo! Groups is subject to
              > http://docs.yahoo.com/info/terms/
              >
              >


              __________________________________________________
              Do You Yahoo!?
              Get email at your own domain with Yahoo! Mail.
              http://personal.mail.yahoo.com/
            Your message has been successfully submitted and would be delivered to recipients shortly.