For most, not all, production servers, these two are the 'right' settings:
query_cache_type = OFF
query_cache_size = 0
Both are needed to avoid some code paths from being unnecessarily followed. (Maybe someday, that will be fixed, too.)
I recommend only 50M as the max for _size.
Here are some metrics to look at to see if the QC is worth having. (Of course, you have to run with it ON or DEMAND for a while to get values for these.)
Qcache_free_memory / query_cache_size -- good value..bad value: 0%,100%
-- Meaning: Pct Query Cache free -- What to do if 'bad': lower query_cache_size
Qcache_lowmem_prunes / Uptime -- good value..bad value: 0,15
-- Meaning: Query Cache spilling -- What to do if 'bad': increase query_cache_size
Qcache_not_cached / Uptime -- good value..bad value: 0,80
-- Meaning: SQL_CACHE attempted, but ignored -- What to do if 'bad': Rethink caching; tune qcache
Qcache_free_blocks * 4096 / query_cache_size -- good value..bad value: 0,1
-- Meaning: Fragmentation in qcache -- What to do if 'bad': decrease query_cache_min_res_unit
Qcache_hits / Qcache_inserts -- good value..bad value: 10,1
-- Meaning: Hit to insert ratio -- high is good
Qcache_hits / (Qcache_hits + Com_select) -- good value..bad value: 100%,25%
-- Meaning: Hit ratio -- What to do if 'bad': Use "_type=DEMAND" and use "SELECT SQL_NO_CACHE" more often
Qcache_total_blocks * query_cache_min_res_unit / Qcache_queries_in_cache
-- Meaning: estimate of query size -- What to do if 'bad': adjust query_cache_min_res_unit
-- Meaning: Queries cached
(Com_select + Qcache_hits) / (Com_insert + Com_update + Com_delete + Com_replace)
-- Meaning: Read to write ratio
> -----Original Message-----
> From: Johan De Meersman [mailto:vegivamp@stripped]
> Sent: Monday, July 15, 2013 11:53 PM
> To: shawn green; email@example.com
> Subject: Re: Mysql cache issues???
> I can't help but wonder wether that first paragraph means there are
> concrete plans to redo the qc?
> shawn green <shawn.l.green@stripped> wrote:
> >Hello Egoitz,
> >On 7/15/2013 1:35 PM, Egoitz Aurrekoetxea wrote:
> >> -----BEGIN PGP SIGNED MESSAGE-----
> >> Hash: SHA1
> >> On 15/07/13 17:27, Reindl Harald wrote:
> >>>... snip...
> >>> i would say my caches are working perfectly (not only the mysql
> >>>cache, also opcache etc.) since whe have generate times down to
> >>> 0.006 seconds for a typical CMS page here which runs in more than
> >>> 200 installations on the main machine, at high load mysqld is never
> >>>the problem
> >>> without the query cache the overall performance drops by 30-40%
> >> Hi,
> >> The query cache hit rate is near 90%.... so I assume it's doing all
> >> properly... now I'm using 1GB as cache.... but... I will do some
> >> tries... till I see some significant behavior either due to success
> >> failure... I was basically wondering what did you though about
> >> performance penalty due to the mysql cache... just that...
> >> Thank you very much then....
> >> ... signature snipped ...
> >Until we redesign the query cache, those stalls will remain. It is
> >unwize to keep so many sets of query results around if they are not
> >actually being used.
> >As has been covered already, the freeze required to perform the purge
> >of all results associated with a specific table can at times be
> >extended (durations of 20-30 minutes are not unusual with cache sizes
> >around 1GB). What you may find is that even if some of your results are
> >frequently for a short period of time, they are not reused at all
> >beyond a certain moment. This means you have hundreds or thousands of
> >sets of query results sitting idle in your cache. Reduce the size of
> >your cache until you start to see your reuse rate or efficiency rate
> >decline significantly. You may be surprised how small that is for your
> >To achieve scalability: customize your cache structures to your
> >workload (this may mean caching the results somewhere other than
> >MySQL), optimize your tables for efficient storage and retrieval, and
> >optimize your queries to be as efficient as practical. There are other
> >scalability options such as replication and sharding that can also be
> >introduced into your production environment to reduce the cost of
> >computation on each copy (or portion) of your data. However, this is a
> >topic best handled in a separate thread.
> Sent from Kaiten Mail. Please excuse my brevity.
> MySQL General Mailing List
> For list archives: http://lists.mysql.com/mysql
> To unsubscribe: http://lists.mysql.com/mysql