openstack/devstack
Code Issues Proposed changes

Merge "Revert "put mysql on a memory diet""

Browse Source
This commit is contained in:
Jenkins 2017-03-16 03:55:54 +00:00 committed by Gerrit Code Review
commit 07d612e460
1 changed files with 1 additions and 182 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

183
lib/databases/mysql
View File

@ -96,191 +96,10 @@ function configure_database_mysql {
iniset -sudo $my_conf mysqld bind-address "$SERVICE_LISTEN_ADDRESS"
iniset -sudo $my_conf mysqld sql_mode TRADITIONAL
iniset -sudo $my_conf mysqld default-storage-engine InnoDB
# the number of connections has been throttled to 256. In the
# event that the gate jobs report "Too many connections" it is
# indicative of a problem that could be the result of one of many
# things. For more details about debugging this error, refer
# https://dev.mysql.com/doc/refman/5.5/en/too-many-connections.html.
# Note that the problem may not ONLY be an issue with MySQL
# connections. If the number of fd's at the OS is too low, you
# could see errors manifest as MySQL "too many connections".
iniset -sudo $my_conf mysqld max_connections 256
iniset -sudo $my_conf mysqld max_connections 1024
iniset -sudo $my_conf mysqld query_cache_type OFF
iniset -sudo $my_conf mysqld query_cache_size 0
# Additional settings to put MySQL on a memory diet. These
# settings are used in conjunction with the cap on max_connections
# as the total memory used by MySQL can be simply viewed as
# fixed-allocations + max_connections * variable-allocations. A
# nifty tool to help with this is
# http://www.mysqlcalculator.com/. A short description of each of
# the settings follows.
# binlog_cache_size, determines the size of cache to hold changes
# to the binary log during a transaction, for each connection. For
# more details, refer
# https://dev.mysql.com/doc/refman/5.6/en/replication-options-binary-log.html#sysvar_binlog_cache_size
# When binary logging is enabled, a smaller binlog cache could
# result in more frequent flushes to the disk and a larger value
# would result in less flushes to the disk but higher memory
# usage. This however only has to do with large transactions; if
# you have a small transaction the binlog cache is necessarily
# flushed on a transaction commit. This is a per-connection cache.
iniset -sudo $my_conf mysqld binlog_cache_size 4K
# binlog_stmt_cache_size determines the size of cache to hold non
# transactional statements in the binary log. For more details,
# refer
# https://dev.mysql.com/doc/refman/5.6/en/replication-options-binary-log.html#sysvar_binlog_stmt_cache_size
# This cache holds changes to non-transactional tables (read:
# MyISAM) or any non-transactional statements which cause
# modifications to data (truncate is an example). These are
# written to disk immediately on completion of the statement or
# when the cache is full. If the cache is too small, you get
# frequent writes to the disk (flush) and if the cache is too
# large, it takes up more memory. This is a per-connection cache.
iniset -sudo $my_conf mysqld binlog_stmt_cache_size 4K
# bulk_insert_buffer_size for MyISAM tables that use a special
# cache for insert statements and load statements, this cache is
# used to optimize writes to the disk. If the value is set to 0,
# the optimization is disabled. For more details refer
# https://dev.mysql.com/doc/refman/5.7/en/server-system-variables.html#sysvar_bulk_insert_buffer_size
# We set this to 0 which could result in higher disk I/O (I/O on
# each insert block completion).
iniset -sudo $my_conf mysqld bulk_insert_buffer_size 0
# host_cache_size controls a DNS lookup optimization. For more
# details refer
# https://dev.mysql.com/doc/refman/5.6/en/host-cache.html
iniset -sudo $my_conf mysqld host_cache_size 0
# innodb_buffer_pool_size This is the size of the server wide
# buffer pool. It is the cache for all data blocks being used by
# the server and is managed as a LRU chain. Dirty blocks either
# age off the list or are forced off when the list is
# full. Setting this to 5MB (default 128MB) reduces the amount of
# memory used by the server and this will result in more disk I/O
# in cases where (a) there is considerable write activity that
# overwhelms the allocated cache, or (b) there is considerable
# read activity on a data set that exceeds the allocated
# cache. For more details, refer
# https://dev.mysql.com/doc/refman/5.6/en/innodb-parameters.html#sysvar_innodb_buffer_pool_size
iniset -sudo $my_conf mysqld innodb_buffer_pool_size 5M
# innodb_ft_cache_size and innodb_ft_total_cache_size control the
# per-connection full text search cache and the server wide
# maximum full text search cache. We should not be using full text
# search and the value is set to the minimum allowable. The former
# is a per-connection cache size and the latter is server
# wide. For more details, refer
# https://dev.mysql.com/doc/refman/5.7/en/innodb-parameters.html#sysvar_innodb_ft_cache_size
# and
# https://dev.mysql.com/doc/refman/5.7/en/innodb-parameters.html#sysvar_innodb_ft_total_cache_size
iniset -sudo $my_conf mysqld innodb_ft_cache_size 1600000
iniset -sudo $my_conf mysqld innodb_ft_total_cache_size 32000000
# innodb_log_buffer_size This buffer is used to buffer
# transactions in-memory before writing them to the innodb
# internal transaction log. Large transactions, or high amounts of
# concurrency, will cause the system to fill this faster and thus
# make the system more disk-bound. For more details, refer
# https://dev.mysql.com/doc/refman/5.7/en/innodb-parameters.html#sysvar_innodb_log_buffer_size
iniset -sudo $my_conf mysqld innodb_log_buffer_size 256K
# innodb_sort_buffer_size, This buffer is used for sorting when
# InnoDB is creating indexes. Could cause that to be slower, but
# only if tables are large. This is a per-connection setting. For
# more details, refer
# https://dev.mysql.com/doc/refman/5.7/en/innodb-parameters.html#sysvar_innodb_sort_buffer_size
iniset -sudo $my_conf mysqld innodb_sort_buffer_size 64K
# join_buffer_size, This buffer makes table and index scans
# faster. So this setting could make some queries more disk
# bound. This is a per-connection setting. For more details refer
# https://dev.mysql.com/doc/refman/5.7/en/server-system-variables.html#sysvar_join_buffer_size.
iniset -sudo $my_conf mysqld join_buffer_size 128
# key_buffer_size defines the index blocks used for MyISAM tables
# and shared between threads. This is a server wide setting. For
# more details see
# https://dev.mysql.com/doc/refman/5.7/en/server-system-variables.html#sysvar_key_buffer_size
iniset -sudo $my_conf mysqld key_buffer_size 8
# max_heap_table_size sets the maximum amount of memory for MEMORY
# tables (which we don't use). The value is set to 16k, the
# minimum allowed. For more details, see
# https://dev.mysql.com/doc/refman/5.7/en/server-system-variables.html#sysvar_max_heap_table_size
iniset -sudo $my_conf mysqld max_heap_table_size 16K
# net_buffer_length Each client has a buffer for incoming and
# outgoing data, both start with a size of net_buffer_length and
# can grow (in steps of 2x) upto a size of max_allowed_packet. For
# more details see
# https://dev.mysql.com/doc/refman/5.7/en/server-system-variables.html#sysvar_net_buffer_length
iniset -sudo $my_conf mysqld net_buffer_length 1K
# read_buffer_size, read_rnd_buffer_size are per-thread buffer
# used for scans on MyISAM tables. It is a per-connection setting
# and so we set it to the minimum value allowable. Same for
# read_rnd_buffer_size. For more details refer
# https://dev.mysql.com/doc/refman/5.6/en/server-system-variables.html#sysvar_read_buffer_size
# and
# https://dev.mysql.com/doc/refman/5.6/en/server-system-variables.html#sysvar_read_rnd_buffer_size
iniset -sudo $my_conf mysqld read_buffer_size 8200
iniset -sudo $my_conf mysqld read_rnd_buffer_size 8200
# sort_buffer_size when a sort is requested, it will be performed
# in memory in a buffer of this size (allocated per connection)
# and if the data exceeds this size it will spill to disk. The
# innodb and myisam variables are used in computing indices for
# tables using the specified storage engine. Since we don't
# dynamically reindex (except during upgrade) these values should
# never be material. Obviously performance of disk based sorts is
# worse than in memory sorts and therefore a high value here will
# improve sort performance for large data. For more details,
# refer:
# https://dev.mysql.com/doc/refman/5.7/en/server-system-variables.html#sysvar_sort_buffer_size
# and
# https://dev.mysql.com/doc/refman/5.7/en/innodb-parameters.html#sysvar_innodb_sort_buffer_size
# and
# https://dev.mysql.com/doc/refman/5.7/en/server-system-variables.html#sysvar_myisam_sort_buffer_size
iniset -sudo $my_conf mysqld sort_buffer_size 32K
iniset -sudo $my_conf mysqld innodb_sort_buffer_size 64K
iniset -sudo $my_conf mysqld myisam_sort_buffer_size 4K
# thread_cache_size specifies how many internal threads to cache
# for use with incoming connections. We set this to 0 whic means
# that each connection will cause a new thread to be created. This
# could cause connections to take marginally longer on os'es with
# slow pthread_create calls. For more details, refer
# https://dev.mysql.com/doc/refman/5.7/en/server-system-variables.html#sysvar_thread_cache_size
iniset -sudo $my_conf mysqld thread_cache_size 0
# thread_stack is the per connection stack size, the minimum is
# 128k and the default is 192k on 32bit and 256k on 64bit
# systems. We set this to 192k. Complex queries which require
# recursion, stored procedures or other memory intensive
# operations could exhaust this and generate a very characteristic
# failure ("stack overflow") which is cleanly detected and the
# query is killed. For more details see
# https://dev.mysql.com/doc/refman/5.6/en/server-system-variables.html#sysvar_thread_stack
iniset -sudo $my_conf mysqld thread_stack 196608
# tmp_table_size is the maximum size of an in-memory temporary
# table. Temporary tables are created by MySQL as part of a
# multi-step query plan. The actual size of the temp table will be
# the lesser of tmp_table_size and max_heap_table_size. If a
# temporary table exceeds this size, it will be spooled to disk
# using the internal_tmp_disk_storage_engine (default
# MyISAM). Queries that often generate in-memory temporary tables
# include queries that have sorts, distinct, or group by
# operations, also queries that perform IN joins. For more details
# see
# https://dev.mysql.com/doc/refman/5.7/en/server-system-variables.html#sysvar_tmp_table_size
iniset -sudo $my_conf mysqld tmp_table_size 1K
if [[ "$DATABASE_QUERY_LOGGING" == "True" ]]; then
echo_summary "Enabling MySQL query logging"
if is_fedora; then