On 21 September 2015 at 12:53, Ivan Zhakov <ivan_at_visualsvn.com> wrote:
> On 21 September 2015 at 12:47, Branko Čibej <brane_at_apache.org> wrote:
>> On 20.09.2015 09:40, Ivan Zhakov wrote:
>>> On 20 September 2015 at 00:53, Branko Čibej <brane_at_apache.org> wrote:
>>>> On 19.09.2015 19:20, Ivan Zhakov wrote:
>>>>> On 19 September 2015 at 17:24, Ivan Zhakov <ivan_at_visualsvn.com> wrote:
>>>>>> On 19 September 2015 at 14:03, Branko Čibej <brane_at_apache.org> wrote:
>>>>>>> On 19.09.2015 13:12, Ivan Zhakov wrote:
>>>>>>>> On 18 September 2015 at 12:49, Stefan Sperling <stsp_at_elego.de> wrote:
>>>>>>>>> On Thu, Sep 17, 2015 at 05:41:41PM +0200, Ivan Zhakov wrote:
>>>>>>>>>> That branch is complete and ready for merging, but I'm still not sure
>>>>>>>>>> whether we should merge it or not.
>>>>>>>>> I think we should merge it to trunk now.
>>>>>>>>>
>>>>>>>>> I don't think this branch can improve much further unless we start
>>>>>>>>> exercising the code ourselves to see how well it's working for us.
>>>>>>>>>
>>>>>>>>> ANd I believe it's hard to tell whether these changes provide an
>>>>>>>>> actual benefit in practice without running the code for a while.
>>>>>>>>>
>>>>>>>>> I like the debug and profiling functionality.
>>>>>>>>> This should make it easy to tune the system going forward.
>>>>>>>> Ok. It seems people here generally support 'reuse-ra-session' branch
>>>>>>>> concept. So I'm going to merge this branch to trunk and see how it
>>>>>>>> will work.
>>>>>>> +1
>>>>>> Merged to trunk in r1704048.
>>>>>>
>>>>> The r1704048 broke JavaHL tests.
>>>>>
>>>>> This happens because JavaHL bindings changes content of AUTH_BATON
>>>>> field in svn_client_ctx_t between diferent svn_client_*() invocations.
>>>>> While RA session in RA session pool references AUTH_BATON from first
>>>>> invocation.
>>>>>
>>>>> The most interesting question is it allowed by our API or not?
>>>> Unfortunately, I'd say it is because none of the API docs say otherwise.
>>>> Or at least I can't find any.
>>>>
>>> I also couldn't find any of them :(
>>>
>>>>> If it's not allowed we just need to fix JavaHL to use the same AUTH_BATON.
>>>> I think the most interesting question here is: why is JavaHL doing this
>>>> in the first place? I have to confess I've no idea, offhand.
>>>>
>>> I also have no idea, but I was sort of hoping that you could provide
>>> some insight on this part :)
>>>
>>> Anyway, as we are doing this kind of things ourselves in JavaHL, there
>>> could be other API users that are also doing it, and we will probably
>>> break them unless we revv the API.
>>
>> Here's one more data point that has nothing to do with JavaHL: The
>> svn-x64-macosx-bdb builder started failing after the merge:
>>
>> https://ci.apache.org/builders/svn-x64-macosx-bdb/builds/253/steps/Test%20ra_local%2Bbdb/logs/faillog
>>
> Yup, I've seen it. This one seems to be caused by too many BDB handles
> opened due parallel run or something. Anyway I'm currently working on
> revertng r1704048: it's clear that code is not ready for prime time
> and it's not obvious how to fix even known problems.
>
I've reverted r1704048 in r1704255. I'll revive 'reuse-ra-session'
branch and attempt to fix these problems there someday.
--
Ivan Zhakov
Received on 2015-09-21 12:15:43 CEST