Items for Friday

[Brent Putman]

On 1/30/17 12:47 PM, Cantor, Scott wrote:
>> Haven't looked at IDP-1115 yet in detail, 

Mostly I responded in the issue, but:
> And the batch case has to handle large files that can take 30+ seconds in unusual cases to download, the dynamic one doesn't.

True, but: as I mentioned in one of the (way too many) linked issues on
this, HttpClient doesn't have any hard wall clock type of timeout over
the whole request operation.  The 3 timeouts are for 1) request
connection from pool 2) initial socket connection 3) socket read timeout
(interval between consecutive packets).  Assuming success on 1 and 2, it
will then happily AFAIK spend hours downloading a multi-TB file, as long
as the data keeps flowing and you don't trip the socket read timeout.

If you want a hard timeout over the whole HTTP request, you have to
build that yourself, like here #7, using the request abort() and a Timer
thread or some other aync means:

http://www.baeldung.com/httpclient-timeout

>
> Probably if one or the other is relying on the HttpClient default it might be better to have both metadata resolver parsers setting it explicitly, just so there's parity in handling it at that later, but that's style.

Specifically it's relying on our HttpClientBuilder default.  If's fine
with me to handle both explicitly for symmetry and style, but as I
mention in IDP-1115 it would be advisable to not repeat the actual
default values for batch from the HCB if we can avoid it.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://shibboleth.net/pipermail/dev/attachments/20170130/c904855e/attachment.html>