[CalendarServer-users] Connection aborted - took too long to close

[Andre LaBranche]

Hi,

In testing 5.4-dev, I came across this error two times while throwing around recurring events with a bunch of exceptions.

[twext.web2.channel.http#error] Connection aborted - took too long to close …

However, I can’t reproduce it after trying a bunch. Also it doesn’t seem to have adversely affected anything - all my tests are producing the expected results.

One thing that might help debug in 5.4-dev is to edit twext/web2/channel/http.py, line 895. Uncomment it and change log.info to log.error, so the line looks like this:

log.error("Timing out client: %s" % str(self.transport.getPeer()))

… then restart the service, or HUP the master process (the one with Combined in its name).

Another thing you could try is to tcpdump from both the client side and the server side, and make sure both sides agree regarding how the connection is being closed.

-dre


> On Feb 9, 2015, at 11:39 AM, Jacques Distler <distler at golem.ph.utexas.edu> wrote:
> 
> 
>> On Feb 9, 2015, at 12:02 PM, Andre LaBranche <dre at apple.com> wrote:
>> 
>> Hi,
>> 
>> That is really weird, and I don’t think I’ve seen this symptom before.
>> 
>> A tcpdump of an affected connection might help.
>> 
>> Can you reproduce this problem when accessing the server locally? It would help to rule out any sort of network-level influence, such as a firewall.
> 
> Yes. It's not a firewall problem.
> 
> As I said, the page starts to load (and since browsers do progressive rendering, I can SEE the data as it loads). Then, 20 seconds later, the server closes the connection, and the browser says
> 
>  Connection Interrupted
>  The document contains no data.
>  The network link was interrupted while negotiating a connection. Please try again.
> 
> I've timed it (with a stopwatch), and the timeout is pretty consistently 20 seconds (I don't think I can reliably eliminate 19 or 21, but definitely not 30 seconds).
> 
> JD