#opendaylight-group-policy: GBP ARCH May9
Meeting started by alagalah at 18:22:16 UTC
(full logs).
Meeting summary
- Mickey to discuss subscription from whiteboard (alagalah, 18:22:37)
- phone 1-866-432-9903 (dconde,
18:23:04)
- access code 204 699 499 (dconde,
18:23:15)
- recording webex (alagalah,
18:24:28)
- dvorkini_ drawing on whiteboard re: EP
attach/mutation (alagalah,
18:24:57)
- Can attach conditions to EPGs as well as EPs,
and EPG can be scoped to enforcement domain (alagalah,
18:25:30)
- Can extend this concept to the tenant
(alagalah,
18:25:43)
- EPG has conditions and can also belong to an
"enforcement domain" which is a renderer or set of renderers who
share the domain-ness (alagalah,
18:26:31)
- Can rely on EP attachment and there's context
of which EPG is belongs to and domain it belongs to and resolution
of EPG and things on EPG depends on and what things it depends on,
and we resolve those things (alagalah,
18:27:06)
- Tenant is generalised concept of EPG
scope (alagalah,
18:27:44)
- Mickey asks if Enforcement domain comes from
EPG and tenant, but dvorkini_ says not necessarily, can be other
things (alagalah,
18:28:13)
- So Enforcement Domain can be fully reactive or
it can be pre-populated (alagalah,
18:28:33)
- pre-population: we know which policies are in
scope, from EPG/tenant, and can apply the same logic to the EPs and
anything that is in scope for your EnfDom you bring in scope
(alagalah,
18:29:10)
- Need to consume different models, because say
we are in an OF controller ... and alagalah lost the train of
thought (see recording) (alagalah,
18:29:45)
- raghu67_ asks if a new EPG gets created, how is
the scope created. dvorkini_ so either an EP attaches (and rules are
applied based on EP) or someone writes the rules for the EPG
(alagalah,
18:30:41)
- dvorkini_ says we can a load of contracts that
aren't instantiated as they only come into scope at various times of
day (ie nightly batch jobs etc) (alagalah,
18:31:29)
- dvorkini_ says that once we get into OpFlex
etc, then we need to be reactive, as others are adding things
dynamically to the system. (alagalah,
18:32:50)
- Up to the renderer what gets resolved or either
you can resolve things that are "within your circle" (alagalah,
18:33:26)
- ie one mode, is you retrieve the group and all
its surrounding contracts, and once EPs appear, you already know,
and once its peer EP comes online you retrieve its'
information (alagalah,
18:34:04)
- mickey says this is where all the questions
come about (alagalah,
18:34:34)
- Question was do I grab anything proactively (ie
if I have consumer, do I try and proactively get provider
information ahead of time) (alagalah,
18:34:58)
- dvorkini_ shows EP->EPG-> contract
(single side, say provider), and we have another side that consumes
it (and shows selectors within EPG) and shows retrieval of
everything from Provider side through to Consumer selector *(which
has the consumer requirements) so we can make the match
(accord) (alagalah,
18:36:02)
- point is that there are multiple modes
(alagalah,
18:36:09)
- dvorkini_ points out that if we don't know
something, we can still reactively retrieve information.
(alagalah,
18:36:45)
- reactivity vs proactivity again resolves to the
renderer, as for instance, OpFlex renderer being reactive may not be
as expensive as say an OF renderer (alagalah,
18:37:21)
- dvorkini_ says for ACI, the way we are thinking
of this is to subscribe to tenants and policies (alagalah,
18:37:49)
- readams points out that at the initial scales
reactive will be fine, but if we want to scale to "ridiculous
levels" we may need to look at alternate sharding etc...
(alagalah,
18:38:33)
- some comment about who's who in the zoo ie does
Jan's MD-SAL team have to own this and dvorkini_ points out that
they shouldn't care (alagalah,
18:38:55)
- net of the last 5min. Renderer commons, gets
EP, contracts etc, does some work, once we get closer to
clauses/subjects/rules, we are using the specific implementation
renderer (this is active rendering), passive rendering is where we
are pretty much just passing stuff through (think APIC subtended to
ODL, and some slight munging of policy semantics might be necessary
but its ostensibly passthru) (alagalah,
18:46:43)
- alagalah will augment these minutes with
mickey's notes (alagalah,
18:54:51)
- Dave Lenrow brings up that we need to model and
be cogniscent of the Renderer performance, and readams points out
that this is a complex situation, espec. when we have an outage,
recovery, and a flood of messaging (alagalah,
19:00:31)
- mickey: Jan maybe interested in how we
subscribe to selection relators, but rob points out that we need to
actually have the yang model get into the datastore, but its a bit
premature (alagalah,
19:04:04)
- We need a better phone/microphone for the
conference room where you host. I'll contribute. It's brutal trying
to follow the audio in the room (lenrow,
19:04:33)
- need to define exactly what we need to
subscribe to. (alagalah,
19:04:54)
- ACTION: Mickey/Rob by
May23 to build up a list of specific parts of the UML model that the
MD-SAL has to support subscription services for (alagalah,
19:06:31)
Meeting ended at 19:06:37 UTC
(full logs).
Action items
- Mickey/Rob by May23 to build up a list of specific parts of the UML model that the MD-SAL has to support subscription services for
People present (lines said)
- alagalah (39)
- regXboi (4)
- odl_meetbot (4)
- dconde (3)
- cloudtoad (2)
- lenrow (1)
Generated by MeetBot 0.1.4.