10:11
kanakb
ASFBot: meeting start
10:11
ASFBot
Meeting started. Anything beginning with [off] will be regarded as
off-the-record and not logged.
10:11
ASFBot
Use #topic to change the topic, #action to add an action item and #info to add
an information item.
10:11
kanakb
#topic administrator api
10:11
kanakb
osgigeek: changes are starting to look good, is there anything specific you
want to call out?
10:13
osgigeek
thanks, the one thing I wanted to ask was state models
10:13
osgigeek
where do we attach them off?
10:14
kanakb
i still like the notion of having state models at a cluster level and
referenced at a resource level
10:14
kanakb
they
10:14
kanakb
're inherently reusable
10:15
osgigeek
I see, ok cool I will take a stab at doing that but will need review when I
check in
10:15
kanakb
no problem
10:15
kanakb
is github commenting effective?
10:15
osgigeek
yes very
10:15
osgigeek
I liked that approach
10:15
kanakb
yeah me too
10:15
osgigeek
It gives me that place to go to for comments
10:15
osgigeek
I dont have to track it through mails
10:15
osgigeek
so please continue to do that
10:15
kanakb
sure
10:16
osgigeek
for now I dont have much
10:16
osgigeek
I am still to include all your and kishores feedback
10:16
kanakb
sure
10:16
osgigeek
and then I will do the state model
10:16
osgigeek
I will also try to do a sample
10:16
kanakb
i know this is more of a side thing for you, so what timeline are you planning?
10:16
osgigeek
as Kishore had asked
10:16
osgigeek
I am hoping to have the checkin before end of the week
10:17
osgigeek
I want to get the next review again say Sat or Sun or latest Mon
10:17
kanakb
sure
10:17
osgigeek
I will try to do it earlier if I can
10:17
osgigeek
but I dont want it to slip beyond end of week
10:17
kanakb
sure, whatever works for your schedule
10:18
osgigeek
what I would like from you is a list of next things we need to roll into the API
10:18
kanakb
ok, we can compile that
10:18
kanakb
i think kishore mentioned figuring out where rebalancer config would fit in
10:18
kanakb
and he leaned towards making it reusable
10:19
osgigeek
right totally skipped my mind .. the rebalancer
10:19
kanakb
there are some things that we need to move out of that, if we take that approach
10:19
kanakb
for instance, partitions needs to be a resource property rather than part of
rebalancer config
10:19
kanakb
i think you've already accounted for that
10:19
osgigeek
yes
10:20
osgigeek
yeah so I think this one I will need a bit more help
10:20
osgigeek
for where to place the resource balancer
10:20
osgigeek
or rather where to introduce it at
10:21
kanakb
so i think the challenge is that reusability is nice, but i feel like every
resource is just a little different
10:21
kanakb
like rebalancing modes can be common across resources
10:21
kanakb
but specific configurations may vary
10:21
kanakb
if kishore comes back, maybe he can provide his thoughts
10:22
osgigeek
sure I think I will ping you guys anyway if I am out of things to do
10:22
kanakb
ok
10:23
osgigeek
but so far I am good given that I want to get things on my plate checked in
before I embark on the next set
10:23
kanakb
alright
10:23
kanakb
that's all i have for admin then
10:24
kanakb
#topic tasks
10:24
kanakb
because of apachecon prep, i've only made a little progress on this
10:24
kanakb
basically the first step is separating out how the controller decides to assign
tasks
10:24
kanakb
so that we can plug in custom strategies
10:24
kanakb
that part is out for review
10:25
kanakb
then there's the matter of allowing partition-specific task behavior
10:25
kanakb
right now every partition does the same thing for a task group
10:25
kanakb
the step after that is decoupling workflows from task groups
10:25
osgigeek
so would that be a part of defining the partition that someone will attach
behavior?
10:26
kanakb
yeah potentially
10:26
kanakb
i'm not really thinking in terms of api yet, but rather required controller
support for use case
10:26
osgigeek
ok
10:27
kanakb
the use cases are basically: submit arbitrary placement strategy, and allow for
long running task groups whose tasks can come in one a time
10:27
kanakb
and then support a scheduling layer on that
10:36
osgigeek
by scheduling you dont mean "when" the task is executed
10:36
osgigeek
you mean "where" the task is executed?
10:36
kanakb
scheduling is "when", yes
10:37
kanakb
right now i'm focusing on where
10:37
kanakb
and also to an extent "how"
10:37
osgigeek
ok
10:37
kanakb
the "when" is mostly orthogonal, but i want to make sure that the base task
execution functionality is solid
10:38
osgigeek
correct
10:38
ASFBot has left IRC (*.net *.split)
10:38
kanakb
ugh ASFBot left
10:40
kanakb
but anyway I don't have any more topics for now
10:40
osgigeek
yeah me neither
10:40
osgigeek
we can call it a meeting
10:40
kanakb
zzhang: anything else you want to cover?
10:41
zzhang
not really
10:41
kanakb
ok then we'll do this again next week
10:41
kanakb
thanks guys
10:41
osgigeek
thx
10:41
osgigeek has left ()
10:42
zzhang
thanks!
