On a reasonably big data set (e.g. one with 70,000 individual contacts), it's not hard to kill your server with a naive manual de-duplication attempt. For example, if you don't specify a group, that number of contacts generates about 5 billion comparisons.
Has anyone thought of a way to prevent unreasonable deduplication attempts?
I think Parvez's work on deduplication performance is a big piece of this puzzle: http://vedaconsulting.co.uk/civicrm-dedupe-workflow-ui (see step 1). Some of the answers here are also relevant to improving performance.
However, neither of these solves what's probably your most common use case - someone running a massive deduplication script. One solution might be to feed deduplication into a queueing system. Another might be to permission the dedupe screen - or even permission specific dedupe jobs, so that someone can run Parvez's smart dedupe but not a standard dedupe rule.
It occurs to me that when you can't find the appropriate screwdriver, there is of course the hammer.
Here's one version of a hammer: https://stackoverflow.com/questions/1189039/how-do-i-routinely-kill-mysql-queries-that-have-been-alive-for-too-long
It'd be nicer to have an extension that analyses the sql before executing it and catching/analysing it.
