In hopes that nobody else suffers critical slowdowns in their production
database in the middle of the day or night, I am sharing what we have
recently learned about some of the automatic memory management "features"
(aka, bugs).
Apparently, the algorithms that decide when and how fast to grow the
shared pool are a bit less than ideal. I do not know all the details yet,
but we have stumbled upon the following bugs in 10.2.0.1:
4466399
4472338 (duplicate of the above 4466399)
4507532 (turned out to be a bad diagnosis, but the bug is real)
4920199
It is my understanding that only 4507532 is fixed in 10.2.0.2 - too bad we
did not hit that bug. =) 4920199 is currently being worked on, and I was
told that 4466399 was only available in v11. And since there is an ingenious
work-around, no one-off patch is planned. What is this work-around? Simple,
turn off the automatic tuning features by setting sga_target = 0.
