This section describes
InnoDB recovery. Topics
To recover an
InnoDB database to the present
from the time at which the physical backup was made, you must
run MySQL server with binary logging enabled, even before taking
the backup. To achieve point-in-time recovery after restoring a
backup, you can apply changes from the binary log that occurred
after the backup was made. See
Section 7.5, “Point-in-Time (Incremental) Recovery”.
If your database becomes corrupted or disk failure occurs, you must perform the recovery using a backup. In the case of corruption, first find a backup that is not corrupted. After restoring the base backup, do a point-in-time recovery from the binary log files using mysqlbinlog and mysql to restore the changes that occurred after the backup was made.
In some cases of database corruption, it is enough to dump,
drop, and re-create one or a few corrupt tables. You can use the
CHECK TABLE statement to check
whether a table is corrupt, although
TABLE naturally cannot detect every possible kind of
In some cases, apparent database page corruption is actually due
to the operating system corrupting its own file cache, and the
data on disk may be okay. It is best to try restarting the
computer first. Doing so may eliminate errors that appeared to
be database page corruption. If MySQL still has trouble starting
InnoDB consistency problems, see
Section 15.21.3, “Forcing InnoDB Recovery” for steps to start the
instance in recovery mode, which permits you to dump the data.
To recover from an unexpected MySQL server exit, the only
requirement is to restart the MySQL server.
InnoDB automatically checks the logs and
performs a roll-forward of the database to the present.
InnoDB automatically rolls back uncommitted
transactions that were present at the time of the crash.
consists of several steps:
Tablespace discovery is the process that
InnoDBuses to identify tablespaces that require redo log application. See Tablespace Discovery During Crash Recovery.
Redo log application
Redo log application is performed during initialization, before accepting any connections. If all changes are flushed from the buffer pool to the tablespaces (
*.ibdfiles) at the time of the shutdown or crash, redo log application is skipped.
InnoDBalso skips redo log application if redo log files are missing at startup.
The current maximum auto-increment counter value is written to the redo log each time the value changes, which makes it crash-safe. During recovery,
InnoDBscans the redo log to collect counter value changes and applies the changes to the in-memory table object.
For more information about how
InnoDBhandles auto-increment values, see Section 18.104.22.168, “AUTO_INCREMENT Handling in InnoDB”, and InnoDB AUTO_INCREMENT Counter Initialization.
When encountering index tree corruption,
InnoDBwrites a corruption flag to the redo log, which makes the corruption flag crash-safe.
InnoDBalso writes in-memory corruption flag data to an engine-private system table on each checkpoint. During recovery,
InnoDBreads corruption flags from both locations and merges results before marking in-memory table and index objects as corrupt.
Removing redo logs to speed up recovery is not recommended, even if some data loss is acceptable. Removing redo logs should only be considered after a clean shutdown, with
Incomplete transactions are any transactions that were active at the time of unexpected exit or fast shutdown. The time it takes to roll back an incomplete transaction can be three or four times the amount of time a transaction is active before it is interrupted, depending on server load.
You cannot cancel transactions that are being rolled back. In extreme cases, when rolling back transactions is expected to take an exceptionally long time, it may be faster to start
3or greater. See Section 15.21.3, “Forcing InnoDB Recovery”.
Change buffer merge
Applying changes from the change buffer (part of the system tablespace) to leaf pages of secondary indexes, as the index pages are read to the buffer pool.
Deleting delete-marked records that are no longer visible to active transactions.
The steps that follow redo log application do not depend on the redo log (other than for logging the writes) and are performed in parallel with normal processing. Of these, only rollback of incomplete transactions is special to crash recovery. The insert buffer merge and the purge are performed during normal processing.
After redo log application,
to accept connections as early as possible, to reduce downtime.
As part of crash recovery,
InnoDB rolls back
transactions that were not committed or in
PREPARE state when the server exited. The rollback is
performed by a background thread, executed in parallel with
transactions from new connections. Until the rollback operation
is completed, new connections may encounter locking conflicts
with recovered transactions.
In most situations, even if the MySQL server was killed
unexpectedly in the middle of heavy activity, the recovery
process happens automatically and no action is required of the
DBA. If a hardware failure or severe system error corrupted
InnoDB data, MySQL might refuse to start. In
this case, see Section 15.21.3, “Forcing InnoDB Recovery”.
For information about the binary log and
InnoDB crash recovery, see
Section 5.4.4, “The Binary Log”.
If, during recovery,
InnoDB encounters redo
logs written since the last checkpoint, the redo logs must be
applied to affected tablespaces. The process that identifies
affected tablespaces during recovery is referred to as
Tablespace discovery relies on the
which defines the directories to scan at startup for tablespace
default setting is NULL, but the directories defined by
datadir are always appended to
value when InnoDB builds a list of directories to scan at
startup. These directories are appended regardless of whether an
innodb_directories setting is
specified explicitly. Tablespace files defined with an absolute
path or that reside outside of the directories appended to the
should be added to the
Recovery is terminated if any tablespace file referenced in a
redo log has not been discovered previously.