On 7 March 2012 09:20, Remy Maucherat <rmaucher_at_redhat.com> wrote:
> On Wed, 2012-03-07 at 08:34 +1100, Greg Wilkins wrote:
> > This async IO only takes place on request that is suspended (after
> > returning after calling startAsync), so there will already be a
> > timeout on the request, which is the async timeout.
> >
> > Is there a need for a finer grained timeout? Ie if the AsyncContext
> > has a timeout of 300s, do we want to have a shorter timeout for idle
> > IO? I can kind of imagine uses... but if feels too complex for most
> > I think.
>
> Well, it doesn't need to be configurable per connection, but there
> should be *a* timeout at some point when using either Servlet 3.1 async
> IO or the upgraded async IO. It shouldn't be tied to either the read or
> write, but more about how long the connection hasn't been doing
> anything, for cleanup purposes.
>
Ah yes the upgrade mechanism would need a timeout.
However, if we add the timeout to the async IO, then for async servlets we
have the problem of having 2 timeouts.
hmmm difficult.