Re: bdb backend: memp_trickle

[Eric Irrgang <erici@motown.cc.utexas.edu>]

On Tue, 15 Feb 2005, Quanah Gibson-Mount wrote:

>In OL 2.3, the checkpoint option in slapd.conf works correctly in that it
>will checkpoint every X amount of data change or Y minutes, whereas right
>now, it will only checkpoint if a new change has come in, and X or Y has
>elapsed.
>
>I assume you are setting the checkpoint command in your slapd.conf file so
>that the DB is regularly checkpointed.

Yes, but I had made the observation about the checkpointing in OL 2.2
working as you say.  I was setting the data size checkpoint parameter to
something big enough that a checkpoint wouldn't happen during the actual
data load, but then the time parameter did not cause slapd to checkpoint
hours later when there were no writes.  Then, BOOM, the next nightly load
comes along and immediately triggers a checkpoint, bogging down the
database when I need it the most.  :)  I didn't realize this had been
recognized as a bug... I was willing to count it as accepted behavior
because I assumed the developers had.  It's a nice surprise when coders
decide to make their code match a spec rather than the other way around...

Don't know if I'm ready to run something labeled Alpha in a production
environment yet.  I'll just trigger checkpoints from outside of slapd.

Thanks!

-- 
Eric Irrgang - UT Austin ITS Unix Systems - (512)475-9342