Tuning Storm Topologies – Again

Unified Click

I spent the majority of the day tuning and scaling the primary data-pipe storm topology as we have been adding then removing features, and we had to get this to a size that handled the load, but didn't burden the system by using workers that weren't needed. Every little change doesn't necessarily require a re-tune of the topology, but when I'm adding or removing bursts of 100k msgs/sec, it's really important to make sure that the new system responds and is as efficient as reasonable.

The problem is that it takes a considerable amount of time to change something, and watch it develop. This isn't like a success/failure, it's a matter of degree, and it takes between 5 and 30 mins to know if the change has helped or hurt the overall cluster.

So that was my day.