2011/12/08 21:53, Tomas Vondra wrote:
>> - performance monitoring and diagnostics. It's way harder to find out
>> what's causing load on a busy Pg server or report on frequent/expensive
>> queries etc. Tooling is limited and fairly primitive. It's find, but
>> nowhere near as powerful and easy as some if the other DBs.
>
> True. Greg Smith actually mentioned this as one of the frequently asked
> features in his post about two weeks ago
> (http://blog.2ndquadrant.com/en/2011/11/global-trends-in-deploying-pos.html).
> I've started to build my own tool and got it somehow working for my needs,
> and there are other tools available, but none of them is really a complete
> solution. Would be nice to form a dev group that would work on this.
Seems a good point. I'm trying to build "a complete solution". :)
Anyway, one of the reasons of such difficulties to build "a complete solution"
is based on necessity of the support from the *entire* core code. Without the
core support, a complete solution would never be built. Obtaining LWLock
statistics or write I/O operations is actually pretty tough work for
"non-experienced" PostgreSQL DBA, like me. :)
For examples, I've been working on investigating PostgreSQL LWLock behaviors
precisely for a few weeks, and it could not be obtained within PostgreSQL
itself, therefore, I picked up SystemTap. However, SystemTap could not be
used in a production system, because it often kills the target processes. :(
How can I observe LWLocks in the production system?
There are several tools to monitor system behaviors around operating systems,
but it is far from understanding PostgreSQL behavior. And DBAs coming from
other RDBMSes, in particular proprietary RDBMSes, need it, because they've
already been using such facilities (or tools) in their RDBMSes.
That's the reason why we need more facilities to observe inside PostgreSQL.
In addition, one more reason of the difficulties is that experienced
PostgreSQL DBAs (or hackers) do not need such facilities in general,
because they can imagine how PostgreSQL works in such particular situation.
I still think we can implement (or enhance) for those facilities if we
focus on it, but I sometimes feel it's like "a chicken and egg situation".
Regards,
--
NAGAYASU Satoshi <satoshi.nagayasu@gmail.com>