Attention is currently required from: fixeria, laforge, neels.
1 comment:
File src/libosmo-pfcp/pfcp_cp_peer.c:
unsigned int tval_rx_heartbeat_s =
osmo_tdef_get(cp_peer->ep->cfg.tdefs, OSMO_PFCP_TIMER_HEARTBEAT_RESP, OSMO_TDEF_S, -1);
o
I understand that the user can configure the HB interval and response timeout separately - this is f […]
Yes, because you are expected to set "HB response timeout" to be > "HB interval".
The tval_rx_heartbeat_s is expected to be set so that it waits up to N Heartbeat Requests lost/non-answered when it triggers.
So usually you'd set eg. tval_tx_heartbeat_s=3, tval_rx_heartbeat_s=10 to trigger after 3 hearbeat request went unanswered.
Unfortunately the PFCP spec doesn't really provide detailed information on how the Heartbeat procedure is supposed to figure out assoc timeout, so we are basically doing something similar to what is done iirc in GTP here.
What I realize now is that the values set by default in the tdef are both 15s for both tval_tx_heartbeat_s and tval_rx_heartbeat_s, which may be too tight. I'll update tval_rx_heartbeat_s to 25 or so. This values were not used so far since this feature was not implemented, and value "15s" doesn't really show up in the spec in any way afaict.
To view, visit change 42120. To unsubscribe, or for help writing mail filters, visit settings.