Details
-
Type:
Task
-
Status: Open
-
Priority:
Trivial
-
Resolution: Unresolved
-
Fix Version/s: None
-
Component/s: Query Cache
-
Labels:
Description
instead of removing queries from cache with today method, we could create a new method (news methods)
query_cache_clean_method=DEFAULT / PROCEDURE
DEFAULT is today implementation
PROCEDURE will execute some mysql schema stored procedure
since query cache can be veeery big (today i'm using 100MB), maybe procedure will lock query cache a lot, and queries will not hit query cache...
well in this case we could continue with query cache locked, or timeout and get back to default method
add a new global variable: query_cache_clean_timeout= 999999 in seconds, if set to "0" don't timeout
well that's just an idea =) query cache is a nice feature, and, yes, it's not the main function of database / application development/runtime, but it can speed up many things very easly, in other words, a good cache (with very high hit rate) is a performace boost in any system
in coments i will put some ideas from others MDEV (sorry about many comments in others MDEV), that could be procedures or method to remove queries from cache when lowmem
Gliffy Diagrams
Attachments
Activity
- All
- Comments
- Work Log
- History
- Activity
- Transitions
from MDEV-4581
maybe should consider:
1) mean period between hits lower than expend time
2) mean period between hits higher than expend time
3) global mean period between hits (today implemented)
i'm seeing many queries that could be removed because the hit period is bigger than the expend time
in other words... if my query cache remove this query i will have near no user reporting 'system is slow now'
please don't close this query cache...
maybe it's a nice 'low memory' solution about query cache
it's not related to prune what will be inserted (label = querycache_insert)
it's not related to control query cache (label = querycache_control)
it's not related to select / query cache hit (label = querycache_result)
it's not related to query cache at client side (label = querycache_client)
it's related to what query cache will do when it get low mem