On 12/09/2011 08:54 PM, Greg Smith wrote:
> I decided about a year ago that further work on using Systemtap was a
> black hole: time goes in, nothing really usable on any production
> server seems to come out.
My off-list e-mail this weekend has, quite rightly, pointed out that
this cheap shot is unfair bordering on libel toward the hard working
Systemtap developers. I'd like to publicly apologize for that and
clarify my frustrated statement here (I'd *really* like this sort of
tool available more)
The main problem I've had with Systemtap is its reputation; I don't
actually have any real, informed gripes about its current state. But
the sort of customers I have are very risk-adverse. PostgreSQL does a
good job attracting that sort of user. I'm sure we have a
disproportionate number of them relative to your average open-source
program. Accordingly, unless a piece of software is very trusted, it's
hard for me to convince anyone to use it. (See "why Greg hates the
disclaimers around the PostgreSQL contrib modules") That makes it hard
for me to give Systemtap a serious spin on most of the production
servers I see my hardest problems on. That's the reason behind the
statement I made badly here--regardless of how much I know about it, I
can't seem to get Systemtap deployed in the places I spent the most time
working at.
True or false, the "Systemtap is more likely to crash your kernel than
DTrace" meme is out there. I think some of that is an unexpectedly bad
side-effect of its open-source development. DTrace had the luxury of
being hidden from the world at large until it was well formed. Whereas
a lot of people saw Systemtap in a really early state, formed opinions
several years ago, and the oldest of those are some of the highest
ranking pages when you search for information. I just searched again
today, and there's mounds of stuff from 2006 and 2007 that surely
doesn't reflect the current state of things coming back from that.
Systemtap didn't get a 1.0 release until September 2009.
As someone who has spent a lot of time at the wrong end of the
"PostgreSQL is slower than MySQL" meme, I shouldn't have just thrown
this sort of criticism out there without explaining the basis for my
statement. I hope this clears up what I meant. Ultimately I think we
need both more tools like Systemtap and DTrace, as well as more
instrumentation inside PostgreSQL, to cover all of the things people
would like visibility into.
--
Greg Smith 2ndQuadrant US greg@2ndQuadrant.com Baltimore, MD
PostgreSQL Training, Services, and 24x7 Support www.2ndQuadrant.us