Improved InnoDB I/O Scalability¶
Because InnoDB is a complex storage engine it must be configured properly in order to perform at its best. Some points are not configurable in standard InnoDB. The goal of this feature is to provide a more exhaustive set of options for XtraDB, like ability to change the log block size.
Version Specific Information¶
Command Line: Yes Config File: Yes Scope: Global Dynamic: Yes Type: Boolean Default Value: True Range: True/False
This variable enables or disables the effect of the per-session value of the innodb_flush_log_at_trx_commit variable.
If the global variable innodb_use_global_flush_log_at_trx_commit is
1, the session will always use the current
global value of innodb_flush_log_at_trx_commit. This is the
upstream compatible mode. If the user attempts to change the
innodb_flush_log_at_trx_commit value for a
session, the session value is ignored.
If the global variable innodb_use_global_flush_log_at_trx_commit is set to
0, a user can modify the
innodb_flush_log_at_trx_commit per-session using the following command:
SET SESSION innodb_flush_log_at_trx_commit=0
This modification only affects the transactions in that session. Other sessions, if they have not been individually modified, continue to use the global innodb_use_flush_log_at_trx_commit value.
Command Line: Yes Config File: Yes Scope: Global Dynamic: No Variable Type: Numeric Default Value: 512 Units: Bytes
This variable changes the size of transaction log records. The default size of 512 bytes is good in most situations. However, setting it to 4096 may be a good optimization with SSD cards. While settings other than 512 and 4096 are possible, as a practical matter these are really the only two that it makes sense to use. Clean restart and removal of the old logs is needed for the variable
innodb_log_block_size to be changed. Note: This feature implementation is considered Experimental quality.
5.6.13-61.0- Ported from Percona Server for MySQL 5.5
This is an existing MySQL 5.6 system variable that has a new allowed value
ALL_O_DIRECT. It determines the method InnoDB uses to flush its data and log files. (See
innodb_flush_method in the MySQL 5.6 Reference Manual).
The following values are allowed:
fsync()to flush both the data and log files.
O_SYNC: use O_SYNC to open and flush the log files; use
fsync()to flush the data files.
O_DIRECT: use O_DIRECT to open the data files and fsync() system call to flush both the data and log files.
O_DIRECT_NO_FSYNC: use O_DIRECT to open the data files and parallel doublewrite files, but does not use the
fsync()system call to flush the data files, log files, and parallel doublewrite files. This option isn’t suitable for XFS file system.
ALL_O_DIRECT: use O_DIRECT to open data files, log files, and parallel doublewrite files and use
fsync()to flush the data files but not the log files or parallel doublewrite files. This option is recommended when InnoDB log files are big (more than 8GB), otherwise, there may be performance degradation. Note: When using this option on ext4 filesystem variable
innodb_log_block_sizeshould be set to 4096 (default log-block-size in ext4) in order to avoid the
5.6.14-62.0- Variable introduced
This variable is used to specify how log checksums are generated and verified. Behavior of
innodb_log_checksum_algorithm depending on its value is mostly identical to
innodb_checksum_algorithm, except that the former applies to log rather than page checksums. NOTE: this feature is currently considered experimental.
The following values are allowed:
none: means that a constant value will be written to log blocks instead of calculated checksum values and no checksum validation will be performed on InnoDB/XtraBackup recovery, or changed page tracking (if enabled).
innodb: (the default) means the default InnoDB behavior – a custom and inefficient algorithm is used to calculate log checksums, but logs created with this option are compatible with upstream MySQL and earlier Percona Server for MySQL or Percona XtraBackup versions that do not support other log checksum algorithms.
crc32: will use CRC32 for log block checksums. Checksums will also benefit from hardware acceleration provided by recent Intel CPUs.
strict_*: Normally, XtraDB or Percona XtraBackup will tolerate checksums created with other algorithms than is currently specified with the
innodb_log_checksum_algorithmoption. That is, if checksums don’t match when reading the redo log on recovery, the block is considered corrupted only if no algorithm produces the value matching the checksum stored in the log block header. This can be disabled by prepending the value with the
strict_innodbwill only accept checksums created using the corresponding algorithms, but not the other ones. To ensure that any log data written using the previous algorithm is fully overwritten before strictness becomes effective following migration procedure to a strict log block checksum should be used:
- note the current LSN;
- set the log block checksum algorithm to the non-strict version of the desired algorithm;
- wait until the current LSN advances to at least the previous LSN + log capacity;
- set the log block checksum algorithm to the strict version of the desired algorithm.
The following information has been added to
SHOW ENGINE INNODB STATUS to confirm the checkpointing activity:
The max checkpoint age The current checkpoint age target The current age of the oldest page modification which has not been flushed to disk yet. The current age of the last checkpoint ... --- LOG --- Log sequence number 0 1059494372 Log flushed up to 0 1059494372 Last checkpoint at 0 1055251010 Max checkpoint age 162361775 Checkpoint age target 104630090 Modified age 4092465 Checkpoint age 4243362 0 pending log writes, 0 pending chkp writes ...