#Are there any issues with load balancing 2 instances of Harvest?

1 messages · Page 1 of 1 (latest)

uncut nymph
#

I'm thinking about running a pair of harvest instances and deploying a load balancer in active/passive mode. Any thoughts or considerations on this?

tardy coral
#

hi @uncut nymph can you say more about the problem(s) that you're trying to solve? Setting up multiple instances of Harvest works fine if done correctly. If you poll a cluster from multiple pollers, it will place more load on ONTAP. You probably don't want to export to the same instance of Prometheus or VM since that will either duplicate or drop metrics, but perhaps you meant multiple pollers exporting to different instances of Prometheus.

uncut nymph
#

My thought is to have two Harvest instances running the same config 1:1. They only have polling (18) and a Prometheus exporter (prometheus1). The Prometheus instance is separate in another environment. The load against the clusters seems light. Longest avg poller time is 1.2s and collector longest avg time is 800ms. My reasoning is that having a single doubled or dropped metric is better than being offline for an extended period. I guess a question I should ask, is there a HA deployment solution for Harvest?

tardy coral
#

what you're describing sounds reasonable, especially with those response times. There is no HA deployment of Harvest at the moment.