Apache Storm

What do you like best?

It is one of the early processing engines exists for data processing. It provides lot of capability to add layers in terms of bolts, each of these bolts can get executed in different worker process gives the capability of distribution of the tasks.

Now storm is integrated with Machine learning frameworks makes it competent with any other streaming solutions

What do you dislike?

Its performance I feel is its show stopper, initially we used storm, as its nature is to distribute the single tasks across nodes eventhough we dont want it tohappen leads to log of performance impact.

thread tuning is itself a one lot job to do, deciding how many threads you need for bolts, sink and source consumes lot of time. Very poor design with respect to back pressure handling.

Recommendations to others considering the product:

Storm has its own use case, we need to match our use case if it macthes otherwise leads to lot of tome being spent to correct something which we dont expect that these problems can happen, for example performance optimization, thread tuning etc.

In older versions it is the single point of failure, later they imporove nimbus to be more resilinet. I recomend with caution of words because i have negative opinion on storm so use it wise only if it fits your usecase

What problems are you solving with the product? What benefits have you realized?

data streaming, enriching and storing it in elasticsearch

we also used it as logic layer to do some computations

we streaming cloud watch metrics from aws also and enriched using storm processors