[tarantool-patches] Re: [PATCH v2] replication: do not ignore replication_connect_quorum.

[Serge Petrenko]

> 9 авг. 2018 г., в 15:20, Vladimir Davydov <vdavydov.dev at gmail.com> написал(а):
> 
> On Thu, Aug 09, 2018 at 12:11:01PM +0300, Vladimir Davydov wrote:
>>>>> diff --git a/test/replication-py/master.lua b/test/replication-py/master.lua
>>>>> index 0f9f7a6f0..51283efdf 100644
>>>>> --- a/test/replication-py/master.lua
>>>>> +++ b/test/replication-py/master.lua
>>>>> @@ -3,6 +3,8 @@ os = require('os')
>>>>> box.cfg({
>>>>>    listen              = os.getenv("LISTEN"),
>>>>>    memtx_memory        = 107374182,
>>>>> +    replication_connect_timeout = 1.0,
>>>>> +    replication_timeout = 0.3
>>>> 
>>>> Why do you need to adjust the timeouts?
>>> 
>>> The timeouts set in all cfg files in all the tests had no effect, just like the
>>> replication_connect_quorum option, cos both of the options were ignored
>>> During bootstrap. We wanted every replica to connect, and the timeout was
>>> Set to TIMEOUT_INFINITY. Now when we actually start passing
>>> replication_connect_timeout, all these timeouts become too small.
>> 
>> Some tests now take too long because of this. E.g. replication/quorum
>> runs for 15 seconds for memtx+vinyl without this patch, which is already
>> long enough, but with this patch it takes more than 30 seconds. We need
>> to do something about that.
> 
> May be, we could pass replication_connect_timeout to these scripts
> in arguments and use different timeouts for bootstrap and subsequent
> restarts?
> 

Yeah, this makes sense. Will try.
I tried to fix replication/quorum, but could only
decrease the time to 22 seconds on memtx + vinyl, and the test wasn’t stable.