-
Notifications
You must be signed in to change notification settings - Fork 0
Home
The behaviour of ProxySQL can be tweaked using global variables. These can be configured in 2 ways:
- at runtime, using the admin interface (preferred)
- at startup, using the dedicated section in the configuration file
ProxySQL supports maximal uptime by allowing most variables to change at runtime and take effect immediately, without having to restart the daemon. There are only 2 variables that cannot be changed at runtime - mysql-threads and mysql-stacksize.
Also, there are 2 types of global variables, depending on which part of ProxySQL they control:
- admin variables, which control the behaviour of the admin interface. Their names begin with the token "admin-"
- mysql variables, which control the MySQL functionality of the proxy. Their names begin with the token "mysql-"
These global variables are stored in a per-thread fashion inside of the proxy in order to speed up access to them, as they are used extremely frequently. They control the behaviour of the proxy in terms of memory footprint or the number of connections accepted, and other essential aspects. Whenever a LOAD MYSQL VARIABLES TO RUNTIME command is issued, all the threads using the variables are notified that they have to update their values.
To change the value of a global variable either use an UPDATE statement:
UPDATE global_variables SET variable_value=1900 WHERE variable_name='admin-refresh_interval';
or the shorter SET statement, similar to MySQL's:
SET admin-refresh_interval = 1700;
SET admin-version = '1.1.1beta8';
Next, we're going to explain each type of variable in detail.
This is a list of semi-colon separated user:password pairs, that can be used to authenticate to the admin interface with read-write rights. For read-only credentials that can be used to connect to the admin, see the variable admin-stats_credentials. Note that the admin interface listens on a separate port from the main ProxySQL thread. This port is controlled through the variable admin-mysql_ifaces.
Note: users in admin-admin_credentials cannot be used also in mysql_users table.
| span 2 cols | span 2 rows | |
| stuff | stuff | |
Semicolon-separated list of hostname:port entries for interfaces on which the admin interface should listen on. Note that this also supports UNIX domain sockets for the cases where the connection is done from an application on the same machine.
Default value: 127.0.0.1:6032.
Example: SET admin-mysql_ifaces='127.0.0.1:6032;/tmp/proxysql_admin.sock'
When this variable is set to true and loaded at runtime, the Admin module does not accept write anymore. This is useful to ensure that ProxySQL is not reconfigured.
When admin-read_only=true, the only way to revert it to false at runtime (and make the Admin module writable again) is to run the command PROXYSQL READWRITE.
Default value: false
The refresh interval (in microseconds) for updates to the query rules statistics and commands counters statistics. Be careful about tweaking this to a value that is:
- too low, because it might affect the overall performance of the proxy
- too high, because it might affect the correctness of the results
Default value: 2000 (microseconds)
This is a list of semi-colon separated user:password pairs that defines the read-only credentials for connecting to the admin interface. These are not allowed updates to internal data structures such as the list of MySQL backend servers (or hostgroups), query rules, etc. They only allow readings from the statistics and monitoring tables (the other tables are not only even visible).
Note: users in admin-stats_credentials cannot be used also in mysql_users table.
Default value: stats:stats
Not currently used (planned usage in a future version).
Not currently used (planned usage in a future version).
This variable displays ProxySQL version. This variable is read only.
When set to true, client flag CLIENT_FOUND_ROWS is set when connecting to MySQL backends.
Default value: true
Enable per-command MySQL query statistics. A command is a type of SQL query that is being executed. Some examples are: SELECT, INSERT or ALTER TABLE.
Default value: true
The delay (in milliseconds) before trying to reconnect after a failed attempt to a backend MySQL server. Failed attempts can take place due to numerous reasons: too busy, timed out for the current attempt, etc. This will be retried for mysql-connect_retries_on_failure times.
Default value: 1 (milliseconds)
The number of times for which a reconnect should be attempted in case of an error, timeout, or any other event that led to an unsuccessful connection to a backend MySQL server. After the number of attempts is depleted, if a connection still couldn't be established, an error is returned. The error returned is either the last connection attempt error or a generic error ("Max connect failure while reaching hostgroup" with error code 28000).
Be careful about tweaking this parameter - a value that is too high can significantly increase the latency which with an unresponsive hostgroup is reported to the MySQL client.
Default value: 5 (reconnect times)
The timeout for a single attempt at connecting to a backend server from the proxy. If this fails, according to the other parameters, the attempt will be retried until too many errors per second are generated (and the server is automatically shunned) or until the final cut-off is reached and an error is returned to the client (see mysql-connect_timeout_server_max).
Default value: 1000 (milliseconds, the equivalent of 1 second)
The timeout for connecting to a backend server from the proxy. When this timeout is reached, an error is returned to the client with code #28000 and the message "Max connect timeout reached while reaching hostgroup...". See also mysql-shun_recovery_time_sec
Default value: 10000 (milliseconds, the equivalent of 10 seconds)
When mysql-connection_max_age_ms is set to a value greater than 0, inactive connections in the connection pool (therefore not currently used by any session) are closed if they were created more than mysql-connection_max_age_ms milliseconds ago. By default, connections aren't closed based on their age.
Default value: 0 (milliseconds)
The default server charset to be used in the communication with the MySQL clients. Note that this is the defult for client connections, not for backend connections.
Default value: utf8
ProxySQL uses a mechanism to automatically ignore hosts if their latency is excessive. Note that hosts are not disabled, but only ignored: in other words, ProxySQL will prefer hosts with a smaller latency. It is possible to configure the maximum latency for each backend from mysql_users table, column max_latency_ms. If mysql_users.max_latency_ms is 0, the default value mysql-default_max_latency_ms applies.
Default: 1000 (millisecond)
Note: due to a limitation in SSL implementation, it is recommended to increase mysql-default_max_latency_ms if using SSL.
Simple throttling mechanism for queries to the backends. Setting this variable to a non-zero value (in miliseconds) will delay the execution of all queries, globally. There is a more fine-grained throttling mechanism in the admin table mysql_query_rules, where for each rule there can be one delay that is applied to all queries matching the rule. That extra delay is added on top of the default, global one.
Default value: 0 (miliseconds)
Mechanism for specifying the maximal duration of queries to the backend MySQL servers until ProxySQL should return an error to the MySQL client. Whenever ProxySQL detects that a query has timed out, it will spawn a separate thread that runs a KILL query against the specific MySQL backend in order to stop the query from running in the backend. Because the query is killed, an error will be returned to the MySQL client.
Default value: 86400000 (miliseconds - the equivalent of 1 day = 24 hours)
Not used for now.
The default schema to be used for incoming MySQL client connections which do not specify a schema name. This is required because ProxySQL doesn't allow connection without a schema.
Default value: information_schema
ProxySQL tracks the status of autocommit as specified by the client, and ensures that autocommit is set correct on backend connections. This implementation is problematic if a client starts a transaction using autocommit, and read/write split is implemented via query rules. In fact, if a read is sent to a slave and ProxySQL sets autocommit=0 on slave, this will result in 2 transactions (one on master and one on slave). To prevent this to happen, if mysql-enforce_autocommit_on_reads=false (the default), ProxySQL won't change the value of autocommit on backend connections for SELECT stataments.
Default value: false
If this variable is set, ProxySQL will log all traffic to the specified filename. Note that the log file is not a text file, but a binary log with encodided traffic. The value of this variable can be set to an absolute pathname (e.g. "/data/events_log/events_log" or else a filename (e.g. "events_log") will be written to the defined data directory. A sequential number will always be suffixed in the file's extension (e.g. "events_log.00000001").
Default value: empty string, not set
This variable specifies the maximum size of files created by ProxySQL logger as specified in mysql-eventslog_filename. When the maximum size is reached, the file is rotated.
Default value: 104857600 (100MB)
ProxySQL uses a connection pool to connect to backend servers.
Connections to backend are never pre-allocated if there is no need, so at start up there will 0 connections to the backend.
When application starts sending traffic to proxysql, this identifies to which backend if needs to send traffic. If there is a connection in the connection pool for that backend, that connection is used, otherwise a new connection is created.
When the connection completes serving the client's request, it is sent back to the the Hostgroup Manager. If the Hostgroup Manager determines that the connection is safe to share and the connection pool isn't full, it will place it in the connection pool. Although, not all the unused connections are kept in the connection pool.
This variable controls the percentage of open idle connections from the total maximum number of connections for a specific server in a hostgroup.
For each hostgroup/backend pair, the Hostgroup Manager will keep in the connection pool up to mysql-free_connections_pct * mysql_servers.max_connections / 100 connections . Connections are kept open with periodic pings.
A connection is idle if it hasn't used since the last round of pings. The time interval between two such rounds of pings for idle connections is controlled by the variable mysql-ping_interval_server_msec.
Default value: 10 (percent)
Currently unused.
Semicolon-separated list of hostname:port entries for interfaces for incoming MySQL traffic. Note that this also supports UNIX domain sockets for the cases where the connection is done from an application on the same machine.
Default value: 0.0.0.0:6033;/tmp/proxysql.sock
Threshold for counting queries passing through the proxy as 'slow'. The total number of slow queries can be found in the stats_mysql_global table, in the variable named Slow_queries (each row in that table represents one variable).
Default value: 1000 (ms)
mysql-max_allowed_packet defines the maximum size of a single packet/command received by the client.
It mimic the behavior of mysqld's max_allowed_packet
Default value: 4194304 (4MB)
Minimum value: 8192 (8KB)
Maximum value: 1073741824 (1GB)
The maximal number of TCP connections that the proxy can handle. After this number is reached, new connections will be rejected with the #HY000 error, and the error message Too many connections.
Default value: 10000 (connections)
Sessions with active transactions running more than this timeout are killed.
Default value: 14400000 (miliseconds - the equivalent of 4 hours)
The interval at which the Monitor module of the proxy will try to connect to all the MySQL servers in order to check whether they are available or not.
Default value: 120000 (miliseconds - the equivalent of 2 minutes)
Connection timeout in milliseconds. The current implementation rounds this value to an integer number of seconds less or equal to the original interval, with 1 second as minimum. This lazy rouding is done because SSL connections are blocking calls.
Default value: 200 (miliseconds)
It enables or disables MySQL Monitor.
Defaut value: true
The duration for which the events for the checks made by the Monitor module are kept. Such events include connecting to backend servers (to check for connectivity issues), querying them with a simple query (in order to check that they are running correctly) or checking their replication lag. These logs are kept in the following admin tables:
mysql_server_connect_logmysql_server_ping_logmysql_server_replication_lag_log
Default value: 600000 (miliseconds - the equivalent of 100 seconds)
Specifies the username and password that the Monitor module will use to connect to the backends. The user needs only USAGE privileges to connect, ping and check read_only. The user needs also REPLICATION CLIENT if it needs to monitor replication lag.
Default values: monitor / monitor
The interval at which the Monitor module should ping the backend servers by using the mysql_ping API.
Default value: 60000 (miliseconds, the equivalent of 1 minute)
How long the Monitor module will wait for a ping reply.
Default value: 1000 (milliseconds)
Currently unused. Will be used by the Monitor module in order to collect data about the global status of the backend servers.
Default value: 60000 (miliseconds, the equivalent of 1 minute)
Currently unused. Will be used by the Monitor module in order to collect data about the global status of the backend servers.
Default value: SELECT * FROM INFORMATION_SCHEMA.GLOBAL_STATUS
Currently unused. Will be used by the Monitor module in order to collect data about the global status of the backend servers.
Default value: 100 (milliseconds)
Currently unused. Will be used by the Monitor module in order to collect data about the global status of the backend servers.
Default value: SELECT * FROM INFORMATION_SCHEMA.GLOBAL_VARIABLES
The interval at which the proxy should connect to the backend servers in order to monitor the replication lag between those that are slaves and their masters. Slaves can be temporarily shunned if the replication lag is too large. This setting is controlled by the mysql_servers.max_replication_lag column in the admin interface, at a per-hostgroup level.
Default value: 10000 (milliseconds, the equivalent of 10 seconds)
How long the Monitor module will wait for the output of SHOW SLAVE STATUS
When replication check returns that Seconds_Behind_Master=NULL , the value of mysql-monitor_slave_lag_when_null (in seconds) is assumed to be the current replication lag. This allow to either shun or keep online a server where replication is broken/stopped.
Default value: 60 (seconds)
Minimum value: 100 (up to 1.3.1) , 0 (from 1.3.2)
Maximum value: 604800 , equivalent to 1 week
This variable controls whether ProxySQL should use a cached (and less accurate) value of wall clock time, or not. The actual API used for this is described here.
Default value: true
The interval at which the proxy should ping backend connections in order to maintain them alive, even though there is no outgoing traffic. The purpose here is to keep some connections alive in order to reduce the latency of new queries towards a less frequently used destination backend server.
Default value: 10000 (milliseconds, the equivalent of 10 seconds)
The proxy internally pings the connections it has opened in order to keep them alive. This elliminates the cost of opening a new connection towards a hostgroup when a query needs to be routed, at the cost of additional memory footprint inside the proxy and some extra traffic. This is the timeout allowed for those pings to succeed.
Default value: 200 (milliseconds)
The minimal timeout used by the proxy in order to detect incoming/outgoing traffic via the poll() system call. If the proxy determines that it should stick to a higher timeout because of its internal computations, it will use that one, but it will never use a value less than this one.
Default value: 2000 (milliseconds, the equivalent of 2 seconds)
The timeout used in order to detect incoming/outgoing traffic after a connection error has occured. The proxy automatically tweaks its timeout to a lower value in such an event in order to be able to quickly respond with a valid connection.
Default value: 100 (milliseconds)
When this variable is set to true, the proxy analyzes the queries passing through it and divides them into classes of queries having different values for the same parameters. It computes a couple of metrics for these classes of queries, all found in the stats_mysql_query_digest table. For more details, please refer to the admin tables documentation.
It is also very important to note that query digest is required to determine when multiplexing needs to be disabled, for example in case of TEMPORARY tables, SQL_CALC_FOUND_ROWS , GET_LOCK, etc.
Do not disabled mysql-query_digests unless you are really sure it won't break your application.
Default value: true (query digests are enabled)
When this variable is set to true, query digest is automatically converted to lowercase.
Default value: false (query digests are are not converted to lowercase)
Defines the maximum length of digest_text as then reported in stats_mysql_query_digest
Default value: 2048
Minimum value: 16
Maximum value: 1048576
Introduced in: 1.3.2
Before 1.3.2 there was no explicit limit, but due to other constraints was limited to 65000
Defines the maximum query length processed when computing query's digest and digest_text
Default value: 65000
Minimum value: 16
Maximum value: 16777216
Introduced in 1.3.2
Before 1.3.2 the limit 65000 was hardcoded
If mysql_query_rules.flagOUT is set and mysql-query_processor_iterations is greater than 0, a matching rule will set flagIN and starts processing rules from the beginning up to mysql-query_processor_iterations iterations.
Therefore, mysql-query_processor_iterations allows to jump back to previous mysql_query_rules.
Default value: 0 (for backward compatibility)
This variable defines which regex engine to use:
Before version v1.4.0, only RE2 was available, CASELESS was always enabled, and GLOBAL was always disabled.
Starting from v1.4.0, both PCRE and RE2 are available. Now both PCRE and RE2 support CASELESS and GLOBAL using re_modifiers.
Although, RE2 doesn't support both CASELESS and GLOBAL at the same time if they are both configured in re_modifiers. For this reason, the default regex engine was change to PCRE.
Default value: 1 (uses PCRE)
In case of failures while running a query, the same can be retries mysql-query_retries_on_failure times.
Default value: 1 (retries just once)
The bitmask of MySQL capabilities (encoded as bits) with which the proxy will respond to clients connecting to it.
This is useful in order to prevent certain features from being used, although it is planned to be deprecated in the future.
The default capabilities are:
server_capabilities = CLIENT_FOUND_ROWS | CLIENT_PROTOCOL_41 | CLIENT_IGNORE_SIGPIPE | CLIENT_TRANSACTIONS | CLIENT_SECURE_CONNECTION | CLIENT_CONNECT_WITH_DB | CLIENT_SSL;More details about server capabilities in the official documentation.
Default value: 47626
The server version with which the proxy will respond to the clients. Note that regardless of the versions of the backend servers, the proxy will respond with this.
Default value: 5.5.30
Currently unused. Will be removed in a future version.
Default value: true
Currently unused. Will be removed in a future version.
Default value: true
Starting from v1.3.0 , each MySQL_Thread has an auxiliary thread that is responsible to handle idle sessions (client connections). mysql-session_idle_ms defines when a session is idle and passed from the main thread to the auxiliary thread.
Default value: 1000 (1 seconds)
mysql-session_idle_show_processlist defines if in idle session (as defined by mysql-session_idle_ms) should be listed in SHOW PROCESSLIST (or in general, in stats_mysql_processlist table). For performance reason, idle sessions are not listed by default.
Default value: false
Sessions are conversations between a MySQL client and a backend server in the proxy. Sessions are generally processed in a stable order but in certain scenarios (like using a transaction workload, which makes sessions bind to certain MySQL connections from the pool), processing them in the same order leads to starvation.
This variable controls whether sessions should be processed in the order of waiting time, in order to have a more balanced distribution of traffic among sessions.
Default value: true
The number of connection errors tolerated to the same server within an interval of 1 second until it is automatically shunned temporarily. For now, this can not be disabled by setting it to a special value, so if you want to do that, you can increase it to a very large value.
Default value: 5
A backend server that has been automatically shunned will be recovered after at least this amount of time.
Note that if ProxySQL isn't handling client traffic, there is no actual hard guarantee of the exact timing, but in practice it shouldn't exceed this value by more than a couple of seconds.
Self tuning:
-
mysql-shun_recovery_time_secshould always be less thanmysql-connect_timeout_server_max/1000, in order to prevent that a server is taken out for so long that an error is returned to the client. Ifmysql-shun_recovery_time_sec>mysql-connect_timeout_server_max/1000, the smaller of the two is used. (see #530) - if only one server is present in a hostgroup and
mysql-shun_recovery_time_sec > 1, the server is automatically brought back online after 1 second
Default value: 10 (seconds)
The stack size to be used with the background threads that the proxy uses to handle MySQL traffic and connect to the backends. Note that changing this value has no effect at runtime, if you need to change it you have to restart the proxy.
Default value: 1048576 (bytes, the equivalent of 1 MB)
The number of background threads that ProxySQL uses in order to process MySQL traffic. Note that there are other "administrative" threads on top of these, such as:
- the admin interface thread
- the monitoring module threads that interact with the backend servers (one for monitoring connectivity, one for pinging the servers and one for monitoring the replication lag)
- occasional temporary threads created in order to kill long running queries that have become unresponsive
- background threads used by the libmariadbclient library in order to make certain interactions with MySQL servers async
Note that changing this value has no effect at runtime, if you need to change it you have to restart the proxy.
Default value: 4 (background threads to process MySQL traffic)
The maximal size of an incoming SQL query to the proxy that will mark the background MySQL connection as non-reusable. This will force the proxy to open a new connection to the backend server, in order to make sure that the memory footprint of the server stays within reasonable limits.
More details about it here: https://dev.mysql.com/doc/refman/5.6/en/memory-use.html
Relevant quote from the mysqld documentation: "The connection buffer and result buffer each begin with a size equal to net_buffer_length bytes, but are dynamically enlarged up to max_allowed_packet bytes as needed. The result buffer shrinks to net_buffer_length bytes after each SQL statement."
Default value: 524288 (bytes, the equivalent of 0.5 MB)
If a resultset returned by a backend server is bigger than this, proxysql will start sending the result to the MySQL client that was requesting the result in order to limit its memory footprint.
Default value: 4194304 (bytes, the equivalent of 4 MB)
If a proxy session (which is a conversation between a MySQL client and a backend MySQL server) has been idle for more than this threshold, the proxy will kill the session.
Default value: 28800000 (miliseconds, the equivalent of 8 hours)