[Date Prev][Date Next]
[Chronological]
[Thread]
[Top]
Re: Syncrepl and chain Overlay
- To: "Ralf Haferkamp" <rhafer@suse.de>
- Subject: Re: Syncrepl and chain Overlay
- From: "Pierangelo Masarati" <ando@sys-net.it>
- Date: Fri, 10 Dec 2004 17:18:11 +0100 (CET)
- Cc: openldap-devel@OpenLDAP.org
- Importance: Normal
- In-reply-to: <200412101633.34790.rhafer@suse.de>
- References: <200412092029.02810.rhafer@suse.de> <200412101512.52316.rhafer@suse.de> <54456.131.175.154.56.1102689207.squirrel@131.175.154.56> <200412101633.34790.rhafer@suse.de>
- User-agent: SquirrelMail/1.4.3a-1
[moved from -software to -devel]
>> mmmmh, that global overlay stuff needs some review; I mean, for write
>> operations I suspect some of the data consolidation should occur earlier
>> so that regular overlays can be injected as global.
>
> Ok I didn't know that, thanks for clarifying.
Note that I'm mostly concerned about write operations.
>> I'v ebeen playing with the chain overlay recently (see my fixes this
>> week)
>> and it worked well alone. There might be some isses in cooperating with
>> the syncrepl. Did you try reversing the overlay stacking order?
>
> I am not using any other overlays than chain. I might be wrong here but
> isn't
> the syncprov overlay only needed on the provider side?
correct.
> Anyway, I've removed all the syncrepl related stuff from my config and
> inserted a normal referral into the database. With that the chain overlay
> seems to work fine. So seems to be indeed some integration issue with
> syncrepl.
The chain overlay responds when the server is __returning__ a referral, so
there shouldn't be any interaction with syncrepl. I suspect there might
be some if syncrepl registers any callback that by chance call the whole
stack instead of directly the underlying database.
In any case, what I note is that the syncrepl consumer (without chain),
when it finds a reference in the subtree it's shadowing (at least in
refreshOnly the first time it imports the provider) it (erroneously, I
assume) chases it, so it ends up shadowing both providers (I'm playing
with a scenario where two providers host two halves of a tree, with
cross-referrals for the other half).
I'll investigate this further.
p.
--
Pierangelo Masarati
mailto:pierangelo.masarati@sys-net.it
SysNet - via Dossi,8 27100 Pavia Tel: +390382573859 Fax: +390382476497