Hi everyone.
Would the Resource Aware Scheduler be the way to control/lower resource 
consumption on supervisor machines given the following symptoms ? :

1) Machines running supervisors have 100% busy CPU forever,  even when there is 
nothing coming into the spouts and nothing being done by the Bolts and 
topologies are not working at all.

Why is that and how to achieve a more reasonable resource consumption ?

2)Nimbus shows the same amount (800 mb) memory allocated to each topology
Why is that and how to lower it ?

-Details:
2 supervisors running each one on a m3.large(2 vCPU 7.5 GiB Mem , 32 GB SSD 
storage) instance.
1 nimbus running on an dedicated m3.large
20 slots available on each supervisor.
18 topologies so far.
Storm 1.1.0 running on Red Hat Enterprise Linux 7.
Each topology is composed by 1 Kafka Spout and 1 Bolt that writes information 
to Cassandra. Nothing more.
The volume of messages coming into Kafka topics  at this moment is near to 
zero, because less than 5 people uses the application at this moment.
every topology is set for 1 worker 1 executor.
parallelism is set to 1 on storm.yaml.

- other questions
a) Do I need to enable  the number of Slots equal to the total number of 
topologies or can the number of slots be lower than the total number of 
topologies?
b) is there a best practice of  a ratio between the number of slots and the 
cpus/cores available ?

Any help will be appreciated.

PS: I am newbie on Storm and getting my MVP live soon.

thanks in advance .

IPVP


















Reply via email to