[Swift-devel] problem at end of workflow

[Michael Wilde]

On 4/24/09 1:46 PM, Mihael Hategan wrote:
> On Fri, 2009-04-24 at 13:29 -0500, Michael Wilde wrote:
>> Any pointers that might enable Zhao to reproduce these things in testing?
> 
> Yeah. Use Glen's proxy.

OK. Glen, when you get a chance, can you post how Zhao might test with a 
proxy similar to yours? Or even run the tests with you, under yours?

Im not following this thread close enough to know how the proxy figures 
into the problem, but sounds like it does.

My goal is that testing beat up the system and find bugs before users, 
so if there indeed strange aspects of various proxies (or user 
environments) we should try to understand that.

Thanks,

- Mike

>> On 4/24/09 1:09 PM, Mihael Hategan wrote:
>>> On Fri, 2009-04-24 at 01:17 -0500, Glen Hocky wrote:
>>>> I upgraded to newest cog and swift (Swift svn swift-r2880 cog-r2391)
>>>> I got new behavior this time. instead of just 16 jobs to the queue, i 
>>>> got up to 50 before I killed. So that's probably a good thing.
>>>>
>>>> The bad thing is that a job became active before any of the jobs in the 
>>>> queue started and i started getting error messages (error transfer log 
>>>> messages)
>>> How much time before?
>>>
>>>> Everything is in
>>>> /home/hockyg/oops/swift/output/rangeroutdir.3000
>>>>
>>>> and as promised, new ranger coaster logs in
>>>>> /home/hockyg/oops/swift/coaster_logs/ranger/coasters
>>>> sorry i don't have more time to stress test this tonight
>>> You seem to be hitting an old problem that is messy to reproduce because
>>> (I think) it involves certain proxies or credentials. I've put some code
>>> in (cog r2392) to alleviate some of the symptoms.
>>>
>>>
>>> _______________________________________________
>>> Swift-devel mailing list
>>> Swift-devel at ci.uchicago.edu
>>> http://mail.ci.uchicago.edu/mailman/listinfo/swift-devel
>