Обсуждение: Bgwriter--- BufferSync() and StrategyStatsDump()

wht StrategyStatsDump prints is
ARC clean buffers at LRU of T1 and T2....
now lets say i have a dirty buffer at position 31st
from LRU and the next one is at 3500th...

in cases like this... t1_clean and t2_clean are of no
use....

a better option would be to have a function like
StrategyDirtyBufferList() which will scan till it
finds max_buffers dirty buffers or end of the list...
but a function like this would have a high
overhead....

so a even better method is to print the no. of total
buffers it had to scan each time bgwriter wrote...

something like.. to find 100 dirty buffers it had to
scan 15000 buffers....

this would greatly help ppl in determining the
appropriate value of bgwriter parameters.... it would
require a simple patch to written which will add two
else statements in StrategyDirtyBufferList() and
returning a struct instead of int...
also a line in function BufferSync()=>
elog(DEBUG1,......

i would really love to see something like this to be
added to postgres... as then we wont have to spend one
whole day trying to determine bgwriter values and
finding in the end there is nothing much....

if there is one... plz tell...(i have already used
iowait time... but its results are not really
encouraging)...

Regards
Himanshu


__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around
http://mail.yahoo.com

Himanshu Baweja <himanshubaweja@yahoo.com> writes:
> this would greatly help ppl in determining the
> appropriate value of bgwriter parameters.... it would
> require a simple patch to written which will add two
> else statements in StrategyDirtyBufferList() and
> returning a struct instead of int...
> also a line in function BufferSync()=>
> elog(DEBUG1,......

That whole algorithm is gone in CVS tip, so there's no point in
creating such a patch ...

            regards, tom lane