I'm opening this ticket to generate discussion...

We've hit issues again recently with overflow items, specifically overflow keys in WT-2119. Tracking overflow items across reconciliation calls for a page is painful, and overflow keys constrain what pages can be evicted during checkpoints (to avoid freelists becoming inconsistent before a checkpoint completes).

Some time back, we changed the handling of overflow items so that we could store values much larger than the configured "page_max" setting on a page. In other words, we made "page_max" a soft limit instead of a hard limit. FTR, MongoDB sets leaf_value_max=64MB because of a performance cliff with overflow values.

What if we change WiredTiger to treat *_key_max = *_value_max = infinite? In other words, what if we stopped writing overflow records at all?