Tom Lane writes:
Keep in mind though that the COPY process is going to involve several working copies of that data (at least four that I can think of --- line input buffer, field input buffer, constructed text object, and constructed tuple).
Will this be for the shared_buffers memory?
I'm also not clear on whether the 512MB limit you refer to will count the PG shared memory area
The OS limit is set to 1.6GB. I increased the shared_buffers to 450MB and it still failed.
hundred meg off the top of what a backend can allocate as temporary workspace.
Is there anything I can change in my log settings so I can produce something which will help you narrow down this problem?
So it seems entirely likely to me that you'd need a ulimit above 512MB to push around 84MB fields.
The issue I am trying to figure is which limit.. the OS limit is set to 1.6GB. I am now trying to increase my shared_buffers. So far have them at 450MB and it is still failing.
Will also try the setting Vivek suggested although for that may need to restart the machine.
