Hello All --
I have a simple queuing application written on top of postgres which
I'm trying to squeeze some more performance out of.
The setup is relatively simple: there is a central queue table in
postgres. Worker daemons do a bounded, ordered, limited SELECT to
grab a row, which they lock by setting a value in the queue.status
column. When the task is complete, results are written back to the
row. The system is designed to allow multiple concurrent daemons to
access a queue. At any one time, we expect 1-5M active items on the
queue.
Now this design is never going to win any performance awards against a
true queuing system like Active/Rabbit/Zero MQ, but it's tolerably
fast for our applications. Fetch/mark times are about 1ms,
independent of the number of items on the queue. This is acceptable
considering that our tasks take ~50ms to run.
However, the writing of results back to the row takes ~5ms, which is
slower than I'd like. It seems that this is because I need to to do
an index scan on the queue table to find the row I just fetched.
My question is this: is there some way that I can keep a cursor /
pointer / reference / whatever to the row I fetched originally, so
that I don't have to search for it again when I'm ready to write
results?
Thanks in advance for any pointers you can provide.
Brian
