December 10, 2005
View 57857a94b6

- tm migrated to the new timers (tm timers completely re-written). - tm structures re-ordered & cleaned up for better "packing" on 64 bit machines => size reduction (~800 bytes less per transaction on a 64 bit machine) - params: - retr_timer1p1, retr_timer1p2, retr_timer1p3 removed and replaced by retr_timer1 and retr_timer2 - all timer values are now expressed in milliseconds (they were in seconds before). Affected params: fr_timer, fr_inv_timer, wt_timer, delete_timer, retr_timer1, retr_timer2 - retr_timer1 (first retransmission) changed to 500 ms - delete_timer changed to 200 ms - functions: - new t_set_fr(timeout_fr_inv, timeout_fr) -- allows changing the transaction timer from script, even if the transaction was already created (see tm docs for more). - fr_inv_timer/fr_timer can be changed now without any performance penalty - more precise retransmissions timing - overall improved tm speed an memory footprint (e.g. on a 64 bit 2*Xeon 2.8Ghz + HT, optimized ser, NO_DEBUG => ~ 6370 cps new version, ~6000 cps old version + rel_0_9_0 tm inser_timer fix; memory usage @ 6000cps: 490Mb new version, 616Mb old version). - NEWS updated

Andrei Pelinescu-Onciul authored on 10/12/2005 16:04:53
December 1, 2005
View 31d48be07e

- updated to latest avp api

Jan Janak authored on 01/12/2005 10:56:48
November 17, 2005
View 063356f7d4

- Aligned to new prototypes of functions in usr_avp.h - clone missing attributes in to_body

Jan Janak authored on 17/11/2005 13:20:25
September 19, 2005
View 9648ac53ad

- spelling corrected

Jan Janak authored on 19/09/2005 17:15:27
View fe04f09564

- obsolete code removed for the sake of clarity - CANCEL terminates retransmission timers properly

Jan Janak authored on 19/09/2005 16:27:33
August 1, 2005
View a4c2553b80

- free AVP list to avoid memory leaks when tm runs out of memory while creating a new cell (fixed by Michal Matyska)

Jan Janak authored on 01/08/2005 20:52:41
July 4, 2005
View 3c41d76d19

- free shm-cloned via lumps bug fixed (triggered by forwarding statlessly a message, then forwarding it statefully and setting a failure_route). Closes: SER-55

Andrei Pelinescu-Onciul authored on 04/07/2005 15:52:11
February 12, 2005
View c10a942f00

- current processed reply made available in failure route - reinforce the checking on transaction destroy to avoid mixture between SHM and PKG uac replies (this can happend only on shutdown)

Bogdan-Andrei Iancu authored on 12/02/2005 23:13:09
August 24, 2004
View 84d8e16586

- Spelling checked - READMEs updated

Jan Janak authored on 24/08/2004 08:58:23
August 23, 2004
View 2be21e31d9

- avp core replaced with .14 version - avp DB support removed from core (later to be added as separate module) - tm support for avps (avp list saved into transactions and made available into tm callbacks, in failure and reply routes).

Bogdan-Andrei Iancu authored on 23/08/2004 20:47:51
March 3, 2004
View 91c18bf342

fifo_hash_unx -> unixsock_hash (better name) missing header field added

Jan Janak authored on 03/03/2004 16:26:37
View fcd6abd768

t_hash equivalent for the unix domain sockets

Jan Janak authored on 03/03/2004 16:16:30
February 23, 2004
View d9f8744f20

some enhancement related to failure handlers processing: - lumps (add_rm and body_lumps) are not any more duplicated, used and removed -> are used direclty the ones from shm memory and at the end all non shm lumps are removed. - after failure handlers are executed, all non null hdr->parsed are checked if points into cloned request's memory chunk. If not, it means the parse was added by failure handlers (some function did some parsing there) and it will be freed as pkg mem. -> It will be safe to call from failure route any functions that does parsing without any additional modifications.

Bogdan-Andrei Iancu authored on 23/02/2004 16:47:43
February 13, 2004
View 0be6158b28

static part of a transaction (struct cell) reduced by 216 bytes (for current MAX_BRANCHES) : - timer_link structer has no more payload field (calculated on the fly) -> 208 bytes less - local, is_invite, noisy_ctimer merged as flags -> 8 bytes less

Bogdan-Andrei Iancu authored on 13/02/2004 18:21:33
February 11, 2004
View c2ea965c63

(hopefuly successful) merger of Uli's CANCEL/FIFO (thanks!), jiri's hash calculation for UAC transactions and current development branch

Jiri Kuthan authored on 11/02/2004 03:38:48