11:51 p.m.
I am really sorry for the ridiculously late response. I will describe
briefly our 1st year and our current approach.
1st year approach.
During the first year, we used infinispan MR to implement our operators.
Most of our operators were Map-only (for example project,filter) and for
these we did not use the intermediate cache. For all the other operators
(join,group by) we used the collector interface. Our reducers always
returned null and the actual output was written to another cache, because
we had a workflow of operators.
Current approach
At the moment we do not use replaced MR, with two dist calls one for the
map and another for the reduce phase. The intermediate data are stored in a
cache (Cache<K,List<V>>). At some point we would like to change to a delta
aware cache. We changed from the MR to dist calls, because we want to run
MR tasks across multiple micro-clouds and the synchronization of Mappers
and reducers it would be more complicated than monitoring the execution of
independent dist calls ( 1 for each micro-cloud). The intermediate data are
written to a ensemble cache ( a LEADS cache), which spans multiple
micro-clouds.
In general, I find it quite useful to be able to "consistently" (without
missing data that are already inside) iterate over the values of a cache.
On Wed, Oct 15, 2014 at 7:41 PM, Emmanuel Bernard <emmanuel(a)hibernate.org>
wrote:
On 13 Oct 2014, at 10:45, Dan Berindei <dan.berindei(a)gmail.com> wrote:
On Fri, Oct 10, 2014 at 6:49 PM, Emmanuel Bernard <emmanuel(a)hibernate.org>
wrote:
> When wrestling with the subject, here is what I had in mind.
>
> The M/R coordinator node sends the M task per segment on the node where
> the segment is primary.
>
What's M? Is it just a shorthand for "map", or is it a new parameter that
controls the number of map/combine tasks sent at once?
M is short for Map. Sorry.
> Each "per-segment" M task is executed and is offered the way to push
> intermediary results in a temp cache.
>
Just to be clear, the user-provided mapper and combiner don't know
anything about the intermediary cache (which doesn't have to be temporary,
if it's shared by all M/R tasks). They only interact with the Collector
interface.
The map/combine task on the other hand is our code, and it deals with the
intermediary cache directly.
Interesting, Evangelos, do you actually use the collector interface or
actual explicit intermediary caches in your approach.
If that’s the collector interface, I guess that’s easier to hide that
sharding business.
We use explicit caches, but should that functionality become available, we
could possibly revert back to Infinspan MR.
> The intermediary results are stored with a composite key [imtermKey-i,
> seg-j].
> The M/R coordinator waits for all M tasks to return. If one does not
> (timeout, rehash), the following happens:
>
We can't allow time out map tasks, or they will keep writing to the
intermediate cache in parallel with the retried tasks. So the originator
has to wait for a response from each node to which it sent a map task.
OK. I guess the originator can see that a node is out of the cluster
though and act accordingly.
> - delete [intermKey-i, seg-i] (that operation could be handled by the
> new per-segment M before the map task is effectively started)
> - ship the M task for that segment-i to the new primary owner of
> segment-i
>
> When all M tasks are received the Reduce phase will read all
> [intermKey-i, *]
> keys and reduce them.
>
Note that if the reduction phase is itself distributed, we could apply
> the same key per segment and shipping split for these.
>
Sure, we have to retry reduce tasks when the primary owner changes, and it
makes sense to retry as little as possible.
>
> Again the tricky part is to expose the ability to write to intermediary
> caches per segment without exposing segments per se as well as let
> someone see a concatenated view if intermKey-i from all segments subkeys
> during reduction.
>
Writing to and reading from the intermediate cache is already abstracted
from user code (in the Mapper and Reducer interfaces). So we don't need to
worry about exposing extra details to the user.
>
> Thoughts?
>
> Dan, I did not quite get what alternative approach you wanted to
> propose. Care to respin it for a slow brain? :)
>
I think where we differ is that I don't think user code needs to know
about how we store the intermediate values and what we retry, as long as
their mappers/combiners/reducers don't have side effects.
Right but my understanding from the LEADS guys was that they had side
effects on their M/Rs. Waiting for Evangelos to speak up.
Should that be available for MapReduce, and the underlying ensemble cache
can
correctly handle one of the strategies described above, we might be
able to change back to Infinispan MR.
Otherwise I was thinking on the same lines: send 1 map/combine task for
each segment (maybe with a cap on the number of segments being processed at
the same time on each node), split the intermediate values per input
segment, cancel+retry each map task if the topology changes and the
executing node is no longer an owner. If the reduce phase is distributed,
run 1 reduce task per segment as well, and cancel+retry the reduce task if
the executing node is no longer an owner.
I had some ideas about assigning each map/combine phase a UUID and making
the intermediate keys [intermKey, seg, mctask] to allow the originator to
retry a map/combine task without waiting for the previous one to finish,
but I don't think I mentioned that before :)
Nice touch, that fixes the rogue node / timeout problem.
There are also some details that I'm worried about:
1) If the reduce phase is distributed, and the intermediate cache is
non-transactional, any topology change in the intermediate cache will
require us to retry all the map/combine tasks that were running at the time
on any node (even if some nodes did not detect the topology change yet). So
it would make sense to limit the number of map/combine tasks that are
processed at one time, in order to limit the amount of tasks we retry (OR
require the intermediate cache to be transactional).
I am not fully following that. What matters in the end it seems is for the
originator to detect a topology change and discard things accordingly, no?
If the other nodes are slaves of that originator for the purpose of that
M/R, we are good.
2) Running a separate map/combine task for each segment is not really an
option until we implement the the segment-aware data container and cache
stores. Without that change, it will make everything much slower, because
of all the extra iterations for each segment.
See my other email about physically merging down the per segment work into
a per node work when you ship that work.
3) And finally, all this will be overkill when the input cache is small,
and the time needed to process the data is comparable to the time needed to
send all those extra RPCs.
So I'm thinking it might be better to adopt Vladimir's suggestion to retry
everything if we detect a topology change in the input and/or intermediate
cache at the end of the M/R task, at least in the first phase.
It would also be an overkill to restart everything MR task if the volume
of data is
large.
I would propose a solution using the distributed iterator and that it would
not miss data whenever a topology change happens.
You half lost but I think that with my proposal to physically merge
the
RPC calls per node instead of per segment, that problem would be alleviated.
Emmanuel
_______________________________________________
infinispan-dev mailing list
infinispan-dev(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/infinispan-dev
Cheers,
Evangelos