On Friday 20 November 2009 14:04:35 Harald Welte wrote:
Hi zecke,
On Wed, Nov 18, 2009 at 10:19:27AM +0100, Holger Freyther wrote:
Hi all,
I have written this small patch yesterday. It is removing the "auto
release" after timeout from lchan and will directly release the channel
ones the ref count drops to zero.
thanks for investigating this.
The reason why we have this long timeout is mainly due to MO SMS.
What happens in the MO SMS case is that we get a CM SERVICE REQUEST is that
the channel stays active for a very long time (especially if you have a
very long sms) before we get the actual CP-DATA.
If our timeout is too low, then the channel will simply timeout while the
SMS is being transmitted.
So what do we do here in order to solve the issue ? Handle CM SERVICE
REQUEST as a special case and manually increase it or start some special
timer?
Hi LaF0rge,
I'm not really following here. I have removed the lchan "auto release" timer
for now and directly release when the refcount drops down to zero. The only
timer that is active is the RSL timer we start in CHan ACTivate (or such).
For the CM Service Request... In the future this, Locationg Updating Request
and Paging Response (some more according to 08.08) would start the transaction
and we would keep the channel open as long as the transaction is alive.
We will introduce a "channel" leak only in the case where we will leak the
transaction. But this is a bug.
In any case I will try sending a real long SMS from my phone and try to
observe what is happening.
regards
holger