[PATCH] Re: Proposed ldb 1.1.30 and tdb 1.3.13 (improve AD DC search performance, make multi-process)

[Stefan Metzmacher]

Hi Andrew,

>> Here's an update on current master.
>>
>> I've added the run-fcntl-deadlock as test.
>>
>> I still want to run the standalone make test of tdb and ldb on
>> solaris,
>> before we push this and have a closer look at the ldb changes.
> 
> Thanks!
> 
> Finally, this is may also be a 'data corruption' issue:  
> 
> My theory is that because there is no read lock held, the index might
> refer to different records compared with the data.  Sometimes that will
> fall back to a traverse (we do that on errors), but sometimes that will
> mean we don't return all the records, as we essentially check twice,
> both in the index and in the final search filter.  
> 
> If a search took a long time, a modify could happen between those two
> points, and while the modify would be atomic, the search would not be
> atomic.

I don't understand what you're trying to say...

Do you mean we need to call ltdb_lock_read() in a wider window?
E.g. when starting the ldb_search() from the top level module stack
until end of that search? Currently only searches are only atomic
within ltdb_search(), but not ldb_search().

> On the flip side, with this patch reads will block all writes for
> longer.

And here...?

> BTW, A test for the LDB changes is in the cmocka thread.  Because LDB
> is buggy it passes right now, but fails if we do have the ldb fix but
> don't have the tdb fix. 

And here...

Be explicit please and refer to specific commits and tests.

Thanks!
metze

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: OpenPGP digital signature
URL: <http://lists.samba.org/pipermail/samba-technical/attachments/20170411/5e032ba3/signature.sig>