maria-developers team mailing list archive

[Kristian Nielsen <knielsen@xxxxxxxxxxxxxxx>]

Sergey Petrunia <sergey@xxxxxxxxxxx> writes:

> * innodb_flush_log_at_trx_commit. Currently, Akonadi uses
> innodb_flush_log_at_trx_commit=2 which as far as I understand can cause up to 1
> second of changes to be lost. Using value of 1 (always flush on commit) should 
> be safer, but maybe it would put an unnecessary load on the system? Akonadi's 
> use case has a bit different idea of what load is - there is only one person 
> making changes, but the system is likely to be a laptop where one cares about 
> battery life, etc.

innodb_flush_log_at_trx_commit=2 can lose up to 1 seconds of changes, yes, but
this only happens if the kernel crashes (kernel bug or power failure). If
mysqld crashes, there is no data loss. Presumably power failure is less likely
on a laptop due to the battery.

But note that this is probably irrelevant for laptop usage. On a laptop, the
disk will almost invariably be configured with cache enabled. This means that
the disk will lie to the operating system about fsync(), claiming that data is
synced durably to disk when it is not. This will invalidate any durability
guarantees that InnoDB tries to make. In fact, it even destroys crash-safety
of InnoDB, meaning that a power failure can corrupt InnoDB data to the point
where recovery is not possible except by complete restore from backup.

 - Kristian.