Il 14/11/2013 11:27, Timo Hatakka ha scritto:
Hi,
we have tried to use event / timestamp based synchronization from an
external user database to repository. It seems that Syncope calls sync
and getLastSyncToken connector framework operations in a following
manner when time goes on:
t1.1 sync, token = null
t1.2 getLastSyncToken
t2.1 sync, token = value got from t1.2
t2.2 getLastSyncToken
t3.1 sync, token = value got from t2.2
t3.2 getLastSyncToken
in ti.j i means task execution iteration number and j means called ICF
operation number in current task execution.
Why getLastSyncToken is used at all? Could it just continue from the
time which was processed in an earlier run, or use getLastSyncToken to
test if there is something newer and then call sync only if something
is available?
There seems to be quite a serious problem in the current
implementation. Let's suppose that we are going to process step t2.1
and there are three new events e1, e2 and e3 available. And let us
further suppose that two new items e4 and e5 will happen during t2.1.
In t2.2 token will be set to e5 which means that in the step t3.1 we
will continue from e5 and e4 will be missed!
We have tested this with database table connector and scripted sql
connector and the problem exists in both implementations.
Hi Timo, yes, you are right, there is a big problem.
I'm quite sure it wasn't in the past: order of calls should be inverted.
I mean:
1. get new sync token
2. execute sync with the oldest one
3. replace the old one whit the new one
In this case we risk to perform twise the same operation but, at least,
it shouldn be a problem.
Does it make sense?
Why don't you open a new issue for this problem? The fix will be
released into the 1.1.5 and 1.2.0.
Best regards,
F.
Thanks!
Timo Hatakka
--
Fabio Martelli
Tirasa - Open Source Excellence
http://www.tirasa.net/
Apache Syncope PMC
http://people.apache.org/~fmartelli/
