Re: How do web servers handle dropped connections?

[Philip Rowlands <phr () DOC IC AC UK>]

Walter Hop wrote:
> looking at the stars last night I was wondering: if I request a scripted
> page /blah.pl from a webserver, the server starts the perl interpreter and
> runs the script. What happens when the browser breaks the connection
> with the web server before the script has finished? Is the script allowed
> to execute fully and is the output being discarded afterwards, or is the
> process killed immediately? In other words, can server side scripts be
> aborted prematurely in this way - and for example corrupt open files or
> leave inconsistencies in a database?

http://hoohoo.ncsa.uiuc.edu/cgi/interface.html
It's not mentioned in the CGI spec (URL above), so one could say that
the behavior is undefined, and left up to the server implementor. There
is a draft RFC forthcoming at <http://web.golux.com/coar/cgi/>, but I
can't find any mention in there either.

In practice, I believe webservers will allow a spawned CGI process to
continue to completion. The rationale (although I can't remember where I
read this) is that Perl scripts, etc. might have files, sockets,
database connections open and should expect to be able to complete
whatever operation they are performing.

Consider the converse; you'd have to write ugly signal handling code
into every CGI, to take account of a closed connection at any moment.
You'd have to implement some sort of transaction rollback if you were in
the middle of rewriting a text file etc... not pleasant.


Cheers,

Phil