Vernon Schryver
vjs@calcite.rhyolite.com
Thu, 19 Jan 2006 21:28:28 -0700 (MST)
> From: Gary Mills <mills@cc.umanitoba.ca>
> I just noticed dbclean running at an unusual time on one of my DCC
> servers (dcc1). The database seems quite large:
>
> -rw-r--r-- 1 daemon other 2109800448 Jan 19 21:24 dcc_db-new
> -rw-r--r-- 1 daemon other 2147483647 Jan 19 21:21 dcc_db
> -rw-r--r-- 1 daemon other 2147483647 Jan 19 18:28 dcc_db-old
>
> On the other one (dcc2), the files are smaller:
>
> -rw-r--r-- 1 daemon other 1368858624 Jan 19 21:24 dcc_db
> The filesystem is 50% full after the rebuild completed. Has the database
> hit a size limit? Has the filesystem run out of fragments? As I recall,
> I created it with a fragment size of 4096, so this should never happen.
Some systems really do not like mmap() regions larger than 2 GBytes.
The last few dozen versions of the DCC source have taken steps to limit
the size of the database to less than 2 GBytes even when there is enough
RAM to allow larger database.
Since about Dec. 10, the maximum size of the consensus database has
passed about 2 GBytes.
If your dcc2 has 2 or 3 GBytes while dcc1 has 4 GBytes or more,
the version you are using should try to have those sizes.
By far the best solution is to install the current version (perhaps
by running /var/dcc/libexec/updatedcc), but if you really must use
the old code, you might use something like
./configure --with-max-db-mem=2048
make install
Vernon Schryver vjs@rhyolite.com