Skip to content

rackspace-blueflood-multi-tenated-timeseries-datastore.md

Latest commit

 

History

History
63 lines (38 loc) · 1.24 KB

rackspace-blueflood-multi-tenated-timeseries-datastore.md

File metadata and controls

63 lines (38 loc) · 1.24 KB
Raw

Blueflood - the multi-tenated Time-Series Datastore

RackSpace

Compared to Others

  • KairosDB

Put everything into one big con family

  • InfluxDB

Doesn't scale-up to requirements, some use Vertica(Paid Solution)

  • OpenTSDB

Like metric-tagging; don't pre-calculate roll-ups; Uses HBase for name-nodes so single point failure

What's New from last talk on Blueflood, now that's it in Production

  • REST API

  • ElasticSearch Indexer, to query metric

  • Exporters (to Kafka and others)

  • Batch Operations

  • More StatD Support (as a new datat-type support more than Gauges)

  • Ingestion Integrations (getting metrics via REST, StatsD, Logstash, CollectD-WIP)

  • Visual Integrations (Graphite-Web, Grafana)

Cloud Metrics (Team that uses Blueflood)

Our Prod Cluster:

  • 32 node Cassandra Cluster

  • 2 Ingestion Nodes

  • 2 Roll-up Nodes

  • 3 ES nodes for indexing

  • augment space on 32-node cluster with CloudBox storage

  • Replication Factor 3

  • Consistency Level 1

  • 6 billion data-points a day

  • Upgraded to Cassandra 2.0

  • Moving to OnMetal (not VM but Elastic and SSD)

Architectural Changes

  • Gotta move to native DataStax driver
  • Stats go through Graphite API, so all Graphite features are available.