Kiran Ayyagari wrote:
hello guys,
As a second step in getting the CiDIT [1] am currently
implementing a new partition which uses
LDIF files as backend, however as I go deeper in
implementation(atm the partition supports add operation)
I asked a couple of questions to myself which I would like to
share and seek opinion on
1. Instead of creating a whole new partition impl for storing
CiDIT can we use the existing (and well tested)
JDBM partition?
Well, at first sight, I don't think it's a good idea. The initial target
was to allow an admin to manipulate the conf without having to use a
tool to change the config. Being able to modify the conf with
vi/emacs/notepad is always a good thing compared to having a broken conf
in an opac file. But ...
The JDBM based partition stores data in binary format but in
order to make the configuration editable using
a simple text editor I propose the following steps
a. After initial startup of server the default configuration
will be dumped to the file system as a single
LDIF file or as a set of directories and LDIF files(
mimicking the DIT structure)
What if the JDBM file is broken ?
b. A watchdog thread will keep listening on this file(s) for
changes
i. any change of type update/delete done to the content
of file(s) will be reflected in
the CiDIT accordingly, but
ii. If the file itself gets deleted we *don't* delete the
configuration data in DIT but simply dump
that data again to file system. To delete the
configuration data completely the corresponding
partition needs to be physically removed from the
disk in which case the system restores the
default configuration during the next startup
c. similarly this file(s) gets updated if the configuration
data DIT gets updated
thanks for your time and I appreciate your comments :)
Ok, I see where you want to go : basically, guarantee that the config
can be read as we know that if it's JDBM packed, the partition will
work. I would say that as a first step, to debug the config itself (ie,
to check that all the parameters are OK), we can do that. But at the
end, I really think we must go for a text based config.
P.S:- Though implementing a partition is not difficult it is
certainly not trivial, am thinking in terms of
maintaining code and the amount of configuration required for
each new partition impl to be used in server
and also considering the time we have to implement replication
support in 2.0
IMO, writing a new partition bascked by ldif files should not really
take much more than 2 weeks. But as we really need to have some config
element for the replication subsystem, and if going with JDBM backed
config will help, I would say : go for it.
Thanks !
[1]
http://cwiki.apache.org/confluence/display/DIRxSRVx20/ADS+2.0+configuration
Kiran Ayyagari
--
--
cordialement, regards,
Emmanuel Lécharny
www.iktek.com
directory.apache.org
