Actions

Incoherent CDR during switchover

Revision as of 10:06, 12 July 2022 by Luc Morissette (talk | contribs) (Added title)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)


It is worth noting that some CDR records can be lost during transition from active to standby following a system fault. Consequently, a CDR analysis script must handle few "corner" cases:

A "Start" CDR entry without corresponding "End" entry

This happens if a call was terminated during the switchover period.

==> In that case, billing the call is not possible, the "End" CDR information was lost.

A "End" CDR entry without corresponding "Start" entry

This happens if a call was answered just before the HA switchover occurred, and the CDR entry was not yet flushed to disk.

==> In that case, billing can still be done using the "End" entry's "end time" versus "connected time" (unless connected time is 0, meaning the call was never answered)

A call with two "End" CDR entries

This case may happen after some partial HA switchover of the Toolpack system:

  • The CDR generating application (Gateway) remains alive, but looses it's connection with toolpack_engine
  • After a timeout, it destroys it's call contexts, and thus writes CDR "End" entries.
  • Later, connection with toolpack_engine is re-established, and some calls were still valid and connected
  • The Gateway application re-synchronizes with these calls. These call continue normally until they're hung-up
  • When hung-up, another "End" CDR entry is written

==> In that case, billing can be done by using the "end time" of the second CDR entry, minus the "connected time" of the first CDR entry.