When changing query cache size at runtime, we got a lot of disk reads. How to explain this phenomenon?

Question

At runtime we issued a query cache size change. We reduced it significantly.

This process run for about 4 minutes.

On the graphs we seen during this period increased disk reads, approximately 18MB/s.

We can't explain why this change needed disk reads, and why so much? As what we expected is that it will prune the cache from the RAM, and nothing else. Anyway we did not expected disk reads at this size.

How do you explain?

Answer 1

Because changing the size most flushes the whole cache, which means that MySQL needed to run a lot of queries that would otherwise have been in the cache? After a few minutes the cache was hot enough that disk reads went down again.