Re: [Firebird-devel] Fix for CORE-3034 causes server collapse under load

[Vlad K. <hv...@us...>]

> Hello, All!
> 
> The fix effectively disables JRD_reschedule calls. For example in btr.cpp:scan() JRD_reschedule will 
> never yield control to another thread because it is always called with at least one latch.
> 
> Since AST processing in Firebird 2.5 and later requires yield of control this means that other 
> connections will have to wait until long index scan finishes.
> 
> In our case this means that long SELECT statement blocks all other writing connections for 15 
> minutes or more effectively causing server collapse.
> New connections are also impossible during that time as they require AST processing to function.
> 
> I privately discussed this issue with Dmitry and would like to confirm here that backing out the 
> patch indeed fixes the problem.

    Dmitry already discussed it with me and we are looking for another solution for CORE-3034.

Regards,
Vlad