Ok, partial day results. Looks like my changes have not solved the problem, just spread it out a little more (as would
beexpected based on your responses). The delays are now shorter (about half) but occur more frequently (maybe 1x /
minute). The params I used are:
bgwriter_lru_percent = 5.0
bgwriter_lru_maxpages = 100
bgwriter_all_percent = 5.0
bgwriter_all_maxpages = 100
wal_buffers = 20
checkpoint_segments = 10
shared_buffers = 196608
I'm going to do some reading today/tonight with the docs/links you have provided and give it another shot on Monday.
Rightnow I'm leaning towards lowering my shared_buffers param as you suggested as third possibility. I will also take
alook at the buffer cache data to get a better feel.
It seems to me that the real solution is for me to stop using the database as an IPC system to pass somewhat
time-criticaldata between processes. Given the time constraints I'm working under this unfortunately was the quickest
route.
Again, I can't say thanks enough for the great info you have provided. I'll continue to post any changes/results I
make/see.
Marc
