Hear the tale of how we revamped an open source API management platform (API Umbrella) to meet the growing needs of the U.S. federal government. Over the past year, we've transitioned from a stack running on Node.js and Elasticsearch to a more efficient stack utilizing Lua/OpenResty and Hadoop/Apache Kylin. We'll dive into why we underwent this migration, and why you may or may not want to consider these technologies in other cases. We'll detail how we rolled out these significant architecture changes without anybody noticing, and also touch on the US government's increasing adoption of open source.
34. Data Cubes
• Pick the dimensions (columns) you
want to aggregate on.
• Define measurements (counts,
sums, etc).
• For every possible combination of
dimensions, measurements are pre-
aggregated.
40. ANSI SQL is a lovely thing
• If Kylin can’t answer a query from its
pre-aggregations, we can fallback
to slower Hadoop queries against
the raw dataset.
• Presto gives an ANSI SQL interface
against the raw dataset.