@miconda, thank you for the detailed follow up. I am looking at what it would take on my end to package up an install based on master to provide better feedback on this.
In the meantime, I do think it's interesting that the acc_cdrs (on the 5.6 branch with tip ebda78c92c) contains the correct call duration, and that the duration is duplicated into the start_time and end_time -- wondering if it's related to 60a62d6f7df233d61d4ccecc39504e77e25421b8.
The following are the only changes to the acc or dialog modules between the 5.6.1 release (which works) and ebda78c92c (which doesn't).
`$ git log --oneline 5.6.1..ebda78c92c -- src/modules/{acc,dialog}/` ad753f4657 modules: readme files regenerated - modules ... [skip ci] 644938a6be dialog: remove strlen() checks when setting a variable 2427f1951a dialog: wrapper function to set a dlg var with unsigned integer parameter 38cbf04afa dialog: fetch directly the int value for dlg variable used in dlg_cseq_update() 363a6267c4 dialog: safer version for getting dlg variable value 948f066718 dialog: updated to get the int value for cseq update operations 60a62d6f7d acc: duplicate dlg var values used in extra2strar_dlg_only() 5bcb685d19 dialog: dont reset dlg_db_mode in POSTCHILDINIT for DB_MODE_SHUTDOWN e8d797cb26 acc: add some explanation that dialogs need to be tracked for CDR accounting
I'm thinking I can try to revert these one-at-a-time and see what happens (probably starting with 60a62d6f7d). If there isn't a 5.7 version planned for some time, perhaps a 5.6 revert and/or other fix is necessary (assuming the master changes are too invasive to backport directly).