Concurrency sizing

Higher query_mem gives each query more memory but leaves room for fewer queries at once, so how many foreign table queries you can run concurrently depends on instance memory and per-query query_mem.

As a rough guide, with the default shared_buffers, keep the combined query_mem of concurrent queries within about 30 percent of instance memory (peak memory). You can estimate this as (instance memory x 0.3) / query_mem per query.

Rising query failure rates, memory usage above 90 percent, or query latencies that climb sharply as more queries run are signs that you are reaching this limit.

To stay within it, lower query_mem per query, move to a larger instance class, or limit how many foreign table queries run at the same time so they queue instead of competing for memory (for example, with a bounded worker pool or a concurrency limit in your application).