Category: Devops

Using Aptly to Manage Highly-Available Debian Repositories

Every team managing more than a few Debian packages at some point faces issues with serving, or, if not that, how effective it’s being served. In our case, we used to use a different tool, Reprepro, which does a pretty good job on the package management, however it does not support having more than one version per package. This kind of constraint directly affects when we need to meet dependencies or even rollback to previous versions and was always problematic in achieving the required results when facing this kind of scenario.

Read more

Every event counts: a reliable Flume-NG fallback solution

This article quickly describes how we handled the transit of external-generated customers events toward our internal Kafka cluster and how we built a reliable failover system using Flume-NG.

We started to work with Apache Flume on its 0.9 version since the beginning, because it was fitting well our need to make internet events landing into a first dumb backet, before being processed. 

Read more

Anatomy of apps reachability in a Mesos cluster

Introduction

In the beginning of this story we used to cluster our Spark applications using  Apache Yarn as our main Resource Manager. At that time we considered a RM more like a “Spark extension”, basically used to optimize Spark processes and nothing more. Our usage of Yarn then, never went beyond deploying those applications, monitoring them via web browser typing something like http://yarn.url:4040/<spark_app>.

Read more