It seems that pap tracks a new pdp for each bounce in xacml-pdp and apex-pdp (as they come up with different name). It does not seem to timeout the older ones. 20+ minutes went by for example for the xacml one and still reports it with a GET on /policy/pap/v1/pdps .

"policies": [],
"currentInstanceCount": 2,
"desiredInstanceCount": 1,
"properties": {},
"pdpInstances": [
{
"instanceId": "apex_18",
"pdpState": "ACTIVE",
"healthy": "HEALTHY",
"message": "Pdp Heartbeat"
},
{
"instanceId": "apex_65",
"pdpState": "ACTIVE",
"healthy": "HEALTHY",
"message": "Pdp Heartbeat"
}

..

"policies": [],
"currentInstanceCount": 2,
"desiredInstanceCount": 1,
"properties": {},
"pdpInstances": [
{
"instanceId": "dev-policy-policy-xacml-pdp-548b6c6d7c-5t86t",
"pdpState": "ACTIVE",
"healthy": "HEALTHY"
},
{
"instanceId": "dev-policy-policy-xacml-pdp-548b6c6d7c-bkzsl",
"pdpState": "ACTIVE",
"healthy": "HEALTHY"
}

jorge.hernandez 10:44 AM
I think may be there's no side effects, don't you think @jim.hahn , that work could be postponed, I think?
I mean at runtime, the important thing, I don't think there's any bad interaction

runtime meaning, executing a guard or monitoring decisions ..

jim.hahn 10:46 AM
seems correct, because "new" PDPs will still be assigned and I think we only have one PDP group for Dublin
if a policy is deployed/undeployed, though, PAP should send an UPDATE to the old PDPs - any that fail to answer should eventually be timed out and removed from the DB (1-2 minutes at most)
so there is some clean-up that should occur

jorge.hernandez 10:52 AM
Ok, sounds good, no work on this release, and I'll create a ticket to revisit in El Alto.