Brandon Ehle writes:
> >Removing fsync calls won't cause any failures for multi-user operation,
> >or for processes terminating unexpectedly.
> >
> >You will see failures for the case where the system loses power. Data
> >which was supposed to have been "committed" may be lost in the kernel's
> >write cache without the fsync calls.
>
> Assuming a stable kernel and a power backup running with DB_TXN_NOSYNC
> should be perfectly safe then?
Those are two big assumptions but assuming they are true then yes, you
are "reasonably" safe.
The problem with the first assumption is that a rogue process that
decides to eat all available memory, fork continuously, open a gazillion
file descriptors, or other such nonsense can crash even "stable"
kernels.
The problem with the second assumption is that your battery may be bad
and crap out on you over the weekend, your software responsible for
shutting down the machine may have problems (either through bugs or
other software upgrades on the machine that hosed the wrong file), or
some similar problem causes your machine to lose power without shutting
down properly.
On the other hand, crashes can and do happen and when they do you run
the risk of losing data. So you could just backup frequently and not
worry about it. That's my general policy.
Michael
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Thu Feb 27 19:08:58 2003