drill-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "weijie.tong (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (DRILL-5975) Resource utilization
Date Wed, 22 Nov 2017 00:44:00 GMT

    [ https://issues.apache.org/jira/browse/DRILL-5975?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16261772#comment-16261772
] 

weijie.tong commented on DRILL-5975:
------------------------------------

Well, I could not understand how to schedule MinorFragments as a Task of YARN. Anyway, we
can't let our scheduler to depend on Yarn, some  companies like Alibaba have their own scheduler
systems. Says, the application has a scheduler to schedule their application-level tasks,
the tasks themselves also need to be scheduled as Tasks of Linux threads.  (some digression,
we now depend on ZK too much to complete some works ,once ZK died, the system also died).

Maybe I have not describe the design clearly. Your first difference does not exist. The task
dependency problem already  solved by Drill now. I will not changed that logic. Every Foreman
will have its own FragmentScheduler. The FragmentScheduler holds the PlanFragments (different
from prior implementation, the PlanFragment will not have a explicit hosts initially, the
hosts will be assigned by the scheduler at runtime.). The `FragmentScheduler` do two things:
*  schedule the leaf fragments (i.e. the Scan nodes, to your examples, A,B)  to run actively.

*  accept the first RecordBatch ready event passively, schedule the next MajorFragment's MinorFragments
(to your example, the C )to run. The running process is the same as before, only one difference:
the `receiver` MinorFragments need to let the `sender` know its destination hosts (through
the graph's number 4 step).  Other things left will behave the same as before, the receiver
of `upstream` MinorFragment (i.e. C) will decide whether to begin the probe side after all
the build side's data have arrived.

Your second question about the system property. I can not answer well. The other changing
beside the scheduler is the RecordBatchManager. It just acts as a buffer stage between the
sender MinorFragments and receiver MinorFragments. This design stays in most of the system.
Flink is what you already known. Spark is the BlockManager, which is master-slave mode ,not
easy to applied to Drill. Both systems support the stream and batch models. While Drill now
will not let the data stay in disk at the exchange stage, it let the batch stay in memory
and flow through the network by a throttling strategy. This design seems to well at low response
time and low concurrent queries(i.e interactive query).The bad thing is resource wasteful
, a whole run time tree (someone called it memory-pipeline ) which is too fat to schedule
(this maybe one reason we have a infinite thread pool). Now the new design seems to lost some
response time performance, but I think it‘s a tradeoff. If the block nodes (sort ,join ,aggregate)
are slow and the scan nodes are quick,  the `memory-pipeline` model's advantage is not represented.

Without a scheduler, Drill is not sufficient to run long-running jobs, as it lacks fault-tolerance
. We can make up this leverage this design step by step.

I think constrained resource is complement with the scheduler each other.


> Resource utilization
> --------------------
>
>                 Key: DRILL-5975
>                 URL: https://issues.apache.org/jira/browse/DRILL-5975
>             Project: Apache Drill
>          Issue Type: New Feature
>    Affects Versions: 2.0.0
>            Reporter: weijie.tong
>            Assignee: weijie.tong
>
> h1. Motivation
> Now the resource utilization radio of Drill's cluster is not too good. Most of the cluster
resource is wasted. We can not afford too much concurrent queries. Once the system accepted
more queries with a not high cpu load, the query which originally is very quick will become
slower and slower.
> The reason is Drill does not supply a scheduler . It just assume all the nodes have enough
calculation resource. Once a query comes, it will schedule the related fragments to random
nodes not caring about the node's load. Some nodes will suffer more cpu context switch to
satisfy the coming query. The profound causes to this is that the runtime minor fragments
construct a runtime tree whose nodes spread different drillbits. The runtime tree is a memory
pipeline that is all the nodes will stay alone the whole lifecycle of a query by sending out
data to upper nodes successively, even though some node could run quickly and quit immediately.What's
more the runtime tree is constructed before actual running. The schedule target to Drill will
become the whole runtime tree nodes.
> h1. Design
> It will be hard to schedule the runtime tree nodes as a whole. So I try to solve this
by breaking the runtime cascade nodes. The graph below describes the initial design. !https://raw.githubusercontent.com/wiki/weijietong/drill/images/design.png!
   [graph link|https://raw.githubusercontent.com/wiki/weijietong/drill/images/design.png]
> Every Drillbit instance will have a RecordBatchManager which will accept all the RecordBatchs
written by the senders of local different MinorFragments. The RecordBatchManager will hold
the RecordBatchs in memory firstly then disk storage . Once the first RecordBatch of a MinorFragment
sender of one query occurs , it will notice the FragmentScheduler. The FragmentScheduler is
instanced by the Foreman.It holds the whole PlanFragment execution graph.It will allocate
a new corresponding FragmentExecutor to run the generated RecordBatch. The allocated FragmentExecutor
will then notify the corresponding FragmentManager to indicate that I am ready to receive
the data. Then the FragmentManger will send out the RecordBatch one by one to the corresponding
FragmentExecutor's receiver like what the current Sender does by throttling the data stream.
> What we can gain from this design is :
> a. The computation leaf node does not to wait for the consumer's speed to end its life
to release the resource.
> b. The sending data logic will be isolated from the computation nodes and shared by different
FragmentManagers.
> c. We can schedule the MajorFragments according to Drillbit's actual resource capacity
at runtime.
> d. Drill's pipeline data processing characteristic is also retained.
> h1. Plan
> This will be a large PR ,so I plan to divide it into some small ones.
> a. to implement the RecordManager.
> b. to implement a simple random FragmentScheduler and the whole event flow.
> c. to implement a primitive FragmentScheduler which may reference the Sparrow project.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Mime
View raw message