See inline. Frits Hoogland http://fritshoogland.wordpress.com frits.hoogland@xxxxxxxxx Phone: +31 20 8946342 On 07 Jan 2014, at 23:33, keydana@xxxxxx wrote: > Hi, > > I'm trying to write some Haskell code for direct SGA access of - preferredly, > as of today - x$ksmsp (sql queries against this table being expensive, > possibly...), and for that I'm following the approach from Kyle Hailey, as > detailed in the Oracle Wait Interface book. > > So I need the struct's start address, but unfortunately, when I query x$ksmsp > several times in a row - even from the same sqlplus process - I get > different results every time... (Actually, this seems to be the case with > most fixed tables, apart from those "fixed" fixed ones as x$ksuse...) This seems very logical to me. This view externalises the shared pool administration (if memory serves me well), which has a lot of stuff going on all the time. You even might think twice about doing this on a real life, busy, system. I would have to look it up, but I think this view holds latches to get a more or less consistent view, which is the reason this view can take some time (acquiring the latches). The X$ views are essentially Oracle memory structures externalised as database tables. Because the data is so transient (you are watching memory areas, hash tables, arrays, etc. of a running program), the design of this views is the tables can be inconsistent (it would hang your system to lock everything down for a consistent version). This means that even if you dump the shared pool of a running system a few times from oradebug, you will see differences too, because all the processes keep on doing stuff. > > It'd be great if anyone had advice, or an idea, what I could do to get this > working all the same (and as to what the reason might be?) Becasue the X$ tables are not tables, but externalised views on memory structures of a running program. > It ought to be possible somehow to do this, as I've tried oradebug direct > access with x$ksmsp and it worked (of course, it runs in sqlplus, so I don't > know...) > > Thanks a lot in advance! > > Sigrid-- > //www.freelists.org/webpage/oracle-l > >