Re: proposal for 9.5: monitoring lock time for slow queries

From: "Pavel Stehule" <pavel(dot)stehule(at)gmail(dot)com>
> isn't it too heavy?

Are you concerned about the impactof collection overhead on the queries
diagnosed? Maybe not light, but I'm optimistic. Oracle has the track
record of long use, and MySQL provides performance schema starting from 5.6.

> I have just terrible negative experience with Vertica, where this design
> is
> used - almost all information about queries are available, but any query
> to
> related tables are terrible slow, so I am inclined to more simple design
> oriented to log based solution. Table based solutions is not practical
> when
> you exec billions queries per day. I understand to motivation, but I
> afraid
> so it can be very expensive and slow on highly load servers.

Which do you mean by "query related to tables", the queries from
applications being diagnosed, or the queries that diagnose the performance
using statistics views?

Could you elaborate on your experience with Vertica? That trouble may be
just because Vertica's implementation is not refined.

I understand the feeling of being inclined to log based solution for its
implementation simplicity. However, the server log is difficult or
impossible to access using SQL queries. This prevents the development of
performance diagnostics functionality in GUI administration tools. Also,
statistics views allow for easy access on PAAS like Amazon RDS and Heroku.

Regards
MauMau