mysqltuner output, recommendations for settings? - mysql

Question

I do not have 250 GB of RAM (way our of budget), I currently have 70GB and it fills to 57GB after an hour of usage (mysql buffers).
I am mostly unsure about:
temporary table buffers
query cache prunes
join_buffer and query_cache size
The innodb buffer is probably not too small, a large part of those 240GB are rarely accessed. The mostly access stuff is all buffered I think.
Some of those values are already at the edge where I read that it's not beneficial (like the 128MB tmp_table or 1MB join_buffer)
[--] Up for: 9h 51m 30s (78M q [2K qps], 921K conn, TX: 110B, RX: 26B)
[--] Reads / Writes: 49% / 51%
[--] Total buffers: 46.6G global + 1.9M per thread (1500 max threads)
[OK] Maximum possible memory usage: 49.3G (82% of installed RAM)
[OK] Slow queries: 0% (1K/78M)
[OK] Highest usage of available connections: 28% (428/1500)
[OK] Key buffer size / total MyISAM indexes: 256.0M/1.6G
[OK] Key buffer hit rate: 99.9% (105M cached / 55K reads)
[OK] Query cache efficiency: 26.9% (11M cached / 44M selects)
[!!] Query cache prunes per day: 4482061
[OK] Sorts requiring temporary tables: 0% (0 temp sorts / 10M sorts)
[!!] Joins performed without indexes: 603
[!!] Temporary tables created on disk: 49% (1M on disk / 2M total)
[OK] Thread cache hit rate: 99% (6K created / 921K connections)
[OK] Table cache hit rate: 83% (1K open / 2K opened)
[OK] Open file limit used: 12% (633/5K)
[OK] Table locks acquired immediately: 99% (9M immediate / 9M locks)
[!!] InnoDB buffer pool / data size: 46.0G/240.6G
[OK] InnoDB log waits: 0
-------- Recommendations -----------------------------------------------------
General recommendations:
Run OPTIMIZE TABLE to defragment tables for better performance
MySQL started within last 24 hours - recommendations may be inaccurate
Increasing the query_cache size over 128M may reduce performance
Adjust your join queries to always utilize indexes
When making adjustments, make tmp_table_size/max_heap_table_size equal
Reduce your SELECT DISTINCT queries without LIMIT clauses
Variables to adjust:
query_cache_size (> 180M) [see warning above]
join_buffer_size (> 1.0M, or always use indexes with joins)
tmp_table_size (> 128M)
max_heap_table_size (> 128M)
innodb_buffer_pool_size (>= 240G)
config:
max_allowed_packet = 64M
thread_stack = 256K
thread_cache_size = 16
max_connections = 1500
max_user_connections = 850
query_cache_limit = 3M
query_cache_size = 180M
query_cache_type = 1
table_open_cache = 2500
key_buffer_size = 256M # index in memory for myisam
innodb_buffer_pool_size = 46G
innodb_log_file_size = 256M
tmp_table_size = 128M
max_heap_table_size = 128M
join_buffer_size = 1M

Answer 1

Let's try to work within the 70GB of RAM. "InnoDB buffer pool / data size: 46.0G/240.6G" is fine if you are not thrashing too badly. Usually improving indexes and/or queries can help.
What I see of the Query cache specs is not too bad. But how big is it? What is the value of query_cache_size? If it is more than 50M, it is too big because the prunes are costly.
Don't change the other settings that mysqltuner recommends.
You have about 50 query_cache_prunes per second, which is too high.
So, although the hit rate is high, I suspect the QC is not well tuned. Recommendations:
Discover which queries are likely to be repeated exactly, and which are not. Change the frequently used ones to SELECT SQL_CACHE ... and change the infrequent ones ones to SELECT SQL_NO_CACHE .... And change to query_cache_type = DEMAND.
To dig deeper, and to fix some slow queries, please follow the suggestions here.

Answer 2

In your configuration file things to do:
thread_cache_size=100 # from 16 for V8 suggested CAP on busy system
innodb_buffer_pool_instances=8 # from default 1 to reduce mutex contention
query_cache_min_res_unit=512 # from 4K to increase capacity of QC
query_cache_limit=1M # from 3M - 180M QC could only hold 60 of them
query_prealloc_size=32768 # from def 8192 to reduce malloc load
max_write_lock_count=2 # from a huge number to give RD opportunity every 2
max_connections=500 # from 1500 will leave 20% growth room
thread_concurrency=30 # to prevent saturation
key_age_threshold=64800 # why discard a key at 5 minutes? keep keep 18 hrs
key_cache_division_limit=50 # from 100% for warm cache to manage
key_cache_block_size=32768 # from 1024 to reduce CPU overhead