Monitoring solution
DalmatinerDB andCockroachDB
DalmatinerDB Solution Overview
• Multiple Metrics formats available
• Redundancy
• Solution totally scalable
• High availability
• The fastest endpoint API in the market
• SQL friendly
• Deduplication/compression ZFS
• SmartOS/Solaris Zones based
• Zabbix,checkMK,sensu,prometheusintegration
Why not?
• Prometheus – federation still too complicated for metrics HA (no cluster)
• OpenTSDB – depends on HDFS (nameservice sucks)
• InfluxDB– change product architecture every week
• KairosDB – depends on Cassandra which is not Time Series Database and have some limitations.
SmartOS
• In-memory operating system and OpenSolarisbased
• Solaris zones (lxc-like)
• Pkgsrc package management from NetBSD
• Easy to upgrade
• ZFS (deduplication,compression,no data loss/corruption)
Creating Zones
The platform is mainly composed of three services:
• DalmatinerDB – Time Series Database based on Riak Core
• PostgresSQL/CockrouachDB - metric metadata
• DalmatinerFE – Service which provides API endpoint to userquery metrics.
• Dalmatiner-Proxy – Layer responsible for collecting data sent from servers by different formats (InfluxDB, Graphite, OpenTSDB, Prometheus, Metrics 2.0 e etc)
• Grafana (optional) – Dashboard for customized graphics
Components
Architecture
DalmatinerDB Concepts
• Each node in the cluster has a copy of this ring. (Cassandra like)
• Each virtual node (or vnode) is a process responsible for a given partition in the ring. A node will host any number of vnodes, depending on the size of the ring and the number of nodes in the cluster.
• Buckets are nothing but namespaces with configuration properties.
• Changes to the ring need to be propagated throughout the cluster. This is what the gossip protocol is for.
DalmatinerDB-proxy
Graphite
Listeners.dp_grafite.bucket = grafite
Listeners.dp_grafite.port = 2003
InfluxDB
Listeners.dp_influx.bucket = influx
Listeners.dp_influx.bucket = http
Listeners.dp_influx.port = 8086
Prometheus
Listeners.dp_prom_writer.bucket = promwriter
Listeners.dp_prom_writer.port = 1234
Listeners.dp_prom_writer.protocol = http
Metrics 2.0
Listeners.dp_metrics2. bucket = grafite
Listeners.dp_metrics2.port = 2004
Listeners.dp_metrics2.protocol = tcp
PostgreSQL/cockroachDB
• PostgreSQL for metadata indexes
• DalmatinerFE e Dalmatiner-proxy services checksmetadata
• CockroachDB support ( hopefully, issue already oppenedon github)
• HA DRBD (if you need to use PostgreSQL for any case)
CockRoachDB
• CockroachDB is a distributed, scale-out SQL database
• Single go binary
• Async replication
• PostgreSQL like
• horizontal scalability
• Raft consensus algorithm (no gossip)
• Basics setup (3 nodes recommended)
• straightforward configuration
CockRoachDB UI
Sending Data
• Collectd/statsd
• Sensu Handler available for every protocols (grafite, influxDB etc)
• Using Graphios to import perfdata files
• Netcat/curl for testing purposes
The platform is mainly composed of three services:
• Zpool create –f –o ashift=12 data dispositivos
• Zfs create /data/dalmatinerdb –o compression=lz4 –o atime=off –o logbias-throughput
Extra - ZFS Tunning
Ddb-adm cluster cli
Getting metrics – SQL friendly
Thank you!
Questions?More information:
https://www.linkedin.com/in/leandro-totino-pereira-06726227
Facebook:
https://www.facebook.com/leandro.totinopereira
