[kde-freebsd] Akonadi/mysqld problems

[Alberto Villa]

On Wed, May 30, 2012 at 10:08 PM, Dwayne MacKinnon <dmk at ncf.ca> wrote:
> Done some further digging:

Thanks!

> It appears that something is going on with the db connections between
> mysqld and akonadi. The net_read_timeout value is being invoked,
> somehow. Basically, under the default value, 30, 30 seconds after you
> start up akonadi mysqld kills all the connections. Akonadi expects
> persistent connections... the conf file it uses resets wait_timeout to 1
> year.
>
> I'm currently experimenting with changing net_read_timeout, although
> obviously this is a hack fix. I've set it to 1 hr. We'll see how that goes.

I wonder if it has to do with this:
On Linux, the NO_ALARM build flag (-DNO_ALARM) modifies how the binary
treats both net_read_timeout and net_write_timeout. With this flag
enabled, neither timer cancels the current statement until after the
failing connection has been waited on an additional net_retry_count
times. This means that the effective timeout value becomes (timeout
setting) × (net_retry_count+1).

So, on Linux timeouts are actually 330 for read and 660 for write. I'm
experimenting with exactly those values. So far it's working, but it's
still syncing.
It might be that 30 seconds are really a too short timeout for
Akonadi, but no one cared because on Linux it's much longer. I'll ping
Akonadi developers on this.

> Couple of other things:
> When akonadi sets up its database, it doesn't create the mysql system
> database. This creates a bunch of errors when mysqld starts up. It may
> not do any actual harm, though. On linux the mysql system database is
> installed.

I have a fix for this.

> The reset of wait_timeout to 1 year seems to be ignored. On my system
> it stays at the default 8 hrs.

Why are you saying this? Consider that wait_timeout and
net_read_timeout have completely different meanings.
-- 
Alberto Villa, FreeBSD committer <avilla at FreeBSD.org>
http://people.FreeBSD.org/~avilla