Re: [Boost-testing] Test results processing transition..

Subject: Re: [Boost-testing] Test results processing transition..
From: Rene Rivera (grafikrobot_at_[hidden])
Date: 2015-01-14 23:28:03

On Wed, Jan 14, 2015 at 8:11 PM, Tom Kent <lists_at_[hidden]> wrote:

>
>
> On Thu, Jan 8, 2015 at 8:27 AM, Rene Rivera <grafikrobot_at_[hidden]> wrote:
>
>> On Thu, Jan 8, 2015 at 7:07 AM, Tom Kent <lists_at_[hidden]> wrote:
>>
>>> I had my machine go down last week, and I'm only now getting back to
>>> this. I just tried a couple runs and had two issues:
>>>
>>> 1) It takes a *long* time to run >30min....I think the previous version
>>> was <5min on the same machine. Did it use to be an incremental process?
>>> Most of results don't change between runs, so does it need to re-process
>>> everything? Can it just check timestamp (or maybe checksum) of the result
>>> sets?
>>>
>>
>> It shouldn't be longer, except for the first time. As it's running the
>> exact same commands, just wrapped in a shell script. Can you tell which
>> part is taking so long?
>>
>
> I've added some timing outputs before each step in the script (and
> manually gaged some with a stopwatch)...here's what I found:
>
> After cleaning out my directory (getting everything from scratch) and
> running (roughly adding the numbers to the left manually) here's what I saw:
> $ cat times_first_run.log
> build_setup Fri Jan 9 22:26:11 UTC 2015 --- 0s
> git_update boost_root Fri Jan 9 22:26:11 UTC 2015 --- 10.5m
> git_update boost_regresssion Fri Jan 9 22:36:50 UTC 2015 --- 1s
> git_update boost_bb Fri Jan 9 22:36:51 UTC 2015 --- 3s
> update_tools Fri Jan 9 22:36:54 UTC 2015 --- 0s
> bootstrap Fri Jan 9 22:36:54 UTC 2015 --- 30s
> build bb Fri Jan 9 22:37:27 UTC 2015 --- 5m
> build_results develop Fri Jan 9 22:42:31 UTC 2015 --- 17m
> build_results master Fri Jan 9 22:59:35 UTC 2015 --- 13m
> upload_results develop Fri Jan 9 23:12:48 UTC 2015 --- 30s
> upload_results master Fri Jan 9 23:13:20 UTC 2015 --- 30s
> complete Fri Jan 9 23:13:56 UTC 2015
>

I don't have exact numbers.. But the first run took about 25 minutes total.
Including me typing in the password twice to do the upload.

Immediately running aging at the end of the first output resulted in:
> $ cat times_second_run.log
> build_setup Mon Jan 12 14:09:14 UTC 2015 --- 0s
> git_update boost_root Mon Jan 12 14:09:14 UTC 2015 --- 2.5m
> git_update boost_regresssion Mon Jan 12 14:11:47 UTC 2015 --- 1s
> git_update boost_bb Mon Jan 12 14:11:48 UTC 2015 --- 1s
> update_tools Mon Jan 12 14:11:49 UTC 2015 --- 0s
> bootstrap Mon Jan 12 14:11:49 UTC 2015 --- 30s
> build bb Mon Jan 12 14:12:25 UTC 2015 --- 2.5m
> build_results develop Mon Jan 12 14:15:01 UTC 2015 --- 13m
>  - downloading + unzipping files (2 downloaded) --- 30s
>  - reading xml, merging results, generating links --- 11.5m
>  - writing documents --- 1m
> build_results master Mon Jan 12 14:28:11 UTC 2015 --- 9.5m
>  - downloading + unzipping files (0 downloaded) --- 5s
>  - reading xml, merging results, generating links --- 9m
>  - writing documents --- 30s
> upload_results develop Mon Jan 12 14:37:38 UTC 2015 --- 30s
> upload_results master Mon Jan 12 14:38:12 UTC 2015 --- 30s
> complete Mon Jan 12 14:38:54 UTC 2015
>

Second run took about 13 minutes total (although I didn't do the upload
part).

Are those times similar to what you're seeing? This was a mult-core machine
> with 3.5GB of RAM, so nothing too tiny.
>

Yes, and no :-\

You second run seems considerably longer than mine. I'm doing this on a
MacBook Pro with 4GB of RAM. With a bunch of other stuff already running
which takes up just about all the physical RAM. So it does spend some
effort swapping. Especially when reading in Marshall's results which take
up 1.5GB just to load.

-- 
-- Rene Rivera
-- Grafik - Don't Assume Anything
-- Robot Dreams - http://robot-dreams.net
-- rrivera/acm.org (msn) - grafikrobot/aim,yahoo,skype,efnet,gmail