#opendaylight-ovsdb: ovsdb_weekly_call
Meeting started by tbachman at 20:00:07 UTC
(full logs).
Meeting summary
- agenda (tbachman, 20:00:23)
- https://meetings.opendaylight.org/opendaylight-ovsdb/2015/ovsdb_weekly_call/opendaylight-ovsdb-ovsdb_weekly_call.2015-01-27-20.00.html
Last recorded meeting minutes (tbachman,
20:00:42)
- Trello Board (tbachman, 20:03:21)
- https://trello.com/b/ddIvDQE0/ovs-openstack
Trello Board for the OVSDB project (tbachman,
20:03:59)
- shague says the NSF migration is waiting on
go-ahead from michal_rehak (tbachman,
20:05:48)
- vishnoianil and shague have pushed patches to
remove AD-SAL pieces (tbachman,
20:06:21)
- shague and vishnoianil working on replacing
AD-SAL Node with MD-SAL Node (tbachman,
20:06:42)
- shague eventually replace activation with the
config subsystem (tbachman,
20:06:54)
- edwarnicke offers his help with the config
subsystem (tbachman,
20:07:05)
- The Neutron Services is eligible for a creation
review at this Thursday’s TSC meeting (tbachman,
20:07:39)
- colindixon says it’s hard for him to imagine
the Creation Review not going through (tbachman,
20:07:53)
- flaviof says a lot of progress has been made on
networking-odl plugin (tbachman,
20:09:07)
- the openstack folks want to remove everything
that’s not pure openstack and turn it into a plugin (tbachman,
20:09:29)
- flaviof says there’s still work to be done to
connect things like LBaaS, FW, and L3 (tbachman,
20:09:52)
- flaviof says the critical piece is the
integration with gerrit and tempest testing (tbachman,
20:10:04)
- Swami asks what the services work is — pulling
from a repo? (tbachman,
20:10:53)
- flaviof says the first one is a final gerrit to
remove all the ODL stuff from the devstack repo (tbachman,
20:11:07)
- flaviof says plaurin_ is working on the “entry
points” in the config, and is looking for help from Swami and
mestery, to take the out of tree work and plug it in (i.e. not use
the L3 agent way, but now pass this responsibility to ODL)
(tbachman,
20:11:56)
- shague says that stuff is important for
supporting the back-end tempest testing so that OpenStack and ODL
can integrate (tbachman,
20:12:20)
- LuisGomez asks shague if they’re trying to get
this testing going in the LF infrastructure (tbachman,
20:12:45)
- flaviof says yes (tbachman,
20:12:48)
- LuisGomez asks if they’ve gotten any input from
the LF infrastructure folks — he’s been looking for the same things
to support the needs of the integration team (tbachman,
20:13:15)
- flaviof says the task right now is to make sure
that openstack doesn’t break ODL — i.e. a changing openstack vs. a
stable ODL release (tbachman,
20:13:43)
- flaviof says that LuisGomez is probably looking
for the opposite — ODL changes affecting external projects
(tbachman,
20:14:03)
- LuisGomez says that’s the case (tbachman,
20:14:07)
- LuisGomez asks if flaviof has had any issue
with Vagrant, etc.? (tbachman,
20:14:29)
- flaviof says there’s a jenkins job in OVSDB
(not yet in JJB) that spawns a special VM and does an all-in-one
openstack deployment (tbachman,
20:16:12)
- flaviof says that tempest is one of the things
they’d do there (tbachman,
20:16:33)
- LuisGomez asks if Vagrant is used in this
setup (tbachman,
20:16:42)
- flaviof says no, not in this one, and has heard
from LF folks that doing things in Vagrant is a nightmare
(tbachman,
20:16:58)
- shague says the goal is a multi-node
deployment (tbachman,
20:17:45)
- edwarnicke has submitted a bunch of new patches
for the MD-SAL OVSDB southbound (tbachman,
20:19:33)
- https://git.opendaylight.org/gerrit/#/q/status:open+project:ovsdb+branch:master+topic:mdsal_ovsdb_sba
<- ovsdb mdsal sb patches (edwarnicke,
20:19:59)
- You can do both active and passive mode
connecting (tbachman,
20:20:19)
- snackewm provided a patch so that the
passive-node delete works (tbachman,
20:20:30)
- There are patches that bring in the infa for
the monitor requests (tbachman,
20:20:48)
- There’s also a patch that listens in the config
store for the OVDSB bridge nodes and retrieves the client
(tbachman,
20:21:08)
- edwarnicke says we need to get some of the data
in the operational data store before pushing things down; would be
nice to be able to look up th bridge to determine handles for
talking to different elements (tbachman,
20:21:57)
- edwarnicke says it’s the translation between
the model and the row/column table objects (tbachman,
20:22:09)
- vishnoianil says taht will be done by the
12th (tbachman,
20:22:20)
- shague asks if there’s any UT for this
(tbachman,
20:23:10)
- edwarnicke says he’ll feel much better as more
tests are put in place (tbachman,
20:23:20)
- edwarnicke asks about the docker-based
testing (tbachman,
20:24:01)
- flaviof says the docker stuff was useful for
version testing against different revs of OVS (tbachman,
20:24:45)
- edwarnicke asks if there’s a version in the
schema (tbachman,
20:24:51)
- shague says that’s a field in the tables
(tbachman,
20:24:57)
- edwarnicke asks for some examples in the ways
that schema shifts between versions (tbachman,
20:25:47)
- ACTION: flaviof to
come up with concrete examples of schema shifts (tbachman,
20:26:05)
- shague says the nicira extensions for tunnels
are one example; used to be part of the port, now are
flow-based (tbachman,
20:26:57)
- clustering (tbachman, 20:27:51)
- ed asks colin to be nice (flaviof,
20:28:36)
- notes Colin is usually nice, if perhaps
sardonic ;) (edwarnicke,
20:28:53)
- colindixon says that clustering comes down to 3
things that are hard in the current model (flaviof,
20:29:40)
- colindixon says you need to figure out what the
state is, then you have to model it, and then you have to make your
data change notification listeners to work correctly (tbachman,
20:29:43)
- shague says with the OVSDB use case, it pulls
in the DB, and when this goes to clustering, how does this map in
ODL (tbachman,
20:31:38)
- colindixon points out that clustering does not
address failover — just recovery of state (tbachman,
20:32:21)
- For deep discussion on clustering and failover
behaviors, please see the webex recording at about 30 minutes
in (tbachman,
20:34:51)
- vishnoianil says all the connection caching is
at the library level (tbachman,
20:36:01)
- vishnoianil says in a clustered configuration,
it’s okay if a node goes down; but what happens if a heartbeat is
lost and a new leader is elected (i.e. connection is stil there,
even though the cluster leader went down) (tbachman,
20:36:38)
- colindixon says that has to be handled somehow
out of band (tbachman,
20:37:02)
- colindixon says it is possible for a 3-node
cluster for one node to be operating in the past, yet still have
connections to local things (tbachman,
20:37:34)
- cdub says this is a split brain scenario, and
it needs to be fenced in some way (tbachman,
20:37:51)
- colindixon says he doesn’t know if OVSDB
provides any such mechanism (tbachman,
20:38:22)
- vishnoianil says there are two connections
involved — OpenFlow plugin and OVSDB (tbachman,
20:38:42)
- colindixon says one way to deal with this is to
piggy-back on openflow (whoever OpenFlow controller is, have OVSDB
follow) (tbachman,
20:39:07)
- colindixon asks if OVSDB has any way to deal
with this (tbachman,
20:39:31)
- shague says you can have multpiple
managers (tbachman,
20:39:40)
- colindixon asks if you can ask an OVSDB manager
who it’s managers are (tbachman,
20:39:51)
- shague says there is a table for this
(tbachman,
20:39:57)
- colindixon says that this can be used — where
the manager writes themselves into the table (tbachman,
20:40:11)
- shague says you can also be a listener
(tbachman,
20:40:19)
- colindixon asks if this state is kept in the
table (tbachman,
20:40:25)
- shague says he doesn’t believe so (tbachman,
20:40:35)
- colindixon says that if someone’s going to take
over node management, you would have at most 1 instance of the OSVDB
plugin attached to the switch at any given time (tbachman,
20:41:05)
- you can have all the nodes in the cluster
connected, but only one node desginating themselves as the master
somehow (tbachman,
20:41:43)
- edwarnicke asks about the use of external_ids
in the OVSDB tables (tbachman,
20:43:30)
- edwarnicke asks if we can write an external_id
for whatever we consider to be the master (tbachman,
20:43:42)
- shague says external_ids are used so the node
can identify itself to the management piece (tbachman,
20:44:16)
- cdub says the other_config can be used
(tbachman,
20:44:22)
- edwarnicke says it feels like doing this allows
us to determine the rightful owner, so that stonith can be
performed (tbachman,
20:45:06)
- edwarnicke says one way to discover the master
is gone by listening to updates on the managers table (tbachman,
20:46:06)
- colindixon says that only works if every
cluster instance is connected to every node — otherwise when the
master goes down, you stop getting updates (tbachman,
20:46:30)
- the advantage of the complete connectivity
model is that failures in the cluster, it’s all internal
(tbachman,
20:47:59)
- colindixon says that load balancers are
sometimes deployed between teh devices connecting to the controller
and the controller (tbachman,
20:51:30)
- edwarnicke says load balancers would have to
understand all the way down to the guts of the OVSDB protocol
(tbachman,
20:51:47)
- colindixon says first thing first: worry about
internal cluster state; address failover second (tbachman,
20:52:16)
- shague says the main request was to get the
clustering support in; sounds like this addresses the needs of the
users (tbachman,
20:52:40)
- edwarnicke agrees, but wants to make sure that
controller node local and internal state is the only data cached
(e.g. in hash-maps) (tbachman,
20:53:03)
- shague says the schema is down in the lib; if
the node goes away, the schema goes away (tbachman,
20:53:37)
- shague says in order to drive OVSDB, you need
that schema; on a reconnect, you get the schema back (tbachman,
20:54:01)
- colindixon says do not store any state in the
cluster that you can easily rebuild (tbachman,
20:54:39)
- colindixon says it’s a lot easier to handle the
lazy loading of caches (tbachman,
20:55:06)
- flaviof says there are a bunch of gerrits that
edwarnicke needs merging — asks if shague or vishnoianil can look at
these (tbachman,
20:57:35)
- ACTION: shague and
vishnoianil to look at edwarnicke’s gerrits (tbachman,
20:57:45)
- colindixon says his suggestion is to use the
gerrit comments where possible, moving to live discussion if
needed (tbachman,
20:58:33)
Meeting ended at 21:04:11 UTC
(full logs).
Action items
- flaviof to come up with concrete examples of schema shifts
- shague and vishnoianil to look at edwarnicke’s gerrits
Action items, by person
- edwarnicke
- shague and vishnoianil to look at edwarnicke’s gerrits
- flaviof
- flaviof to come up with concrete examples of schema shifts
- shague
- shague and vishnoianil to look at edwarnicke’s gerrits
People present (lines said)
- tbachman (128)
- cdub (11)
- flaviof (7)
- odl_meetbot (6)
- dneary (2)
- edwarnicke (2)
- odp-gerritbot (1)
- colindixon (1)
- shague (0)
Generated by MeetBot 0.1.4.