metrics, logs, transaction traces, anomaly detection at scale

41
Processing Metrics, Logs & Traces … at Scale Otis Gospodnetić

Upload: sematext-group-inc

Post on 28-Jul-2015

2.070 views

Category:

Data & Analytics


0 download

TRANSCRIPT

Processing Metrics, Logs & Traces

… at ScaleOtis Gospodnetić

WHO WHY

WHAT HOW

WHO

HQ: BrooklynPeople: Everywhere

WHO Otis Gospodnetić

Sematext founderApache memberBook authorex-Lucene/Solr dev

WHO Services

Solr Elasticsearch* Kafka Spark HBase Cassandra...

* We’ve got serious Solr & Elasticsearch ninjas on the team!

WHY

WHO Clients want

Performance Bottlenecks Tuning Scaling

WHY

WHO

Before you can fix things need to know what to fix

WHY

WHY We need….INSIGHT

Performance Metrics! Anomalies! Logs!

WHY i.e. Need Tools!

Metrics monitoring

Log searching Anomaly alerting

OSS

Use the (Open) Source, Luke

OSS

OpenTSDB InfluxDB Ganglia Graphite Nagios ELK ...

OSS

“I have an ELK stack that has been suffering as of late. The logstash service will continually crash, the elasticsearch cluster is hardly in the green, and it is taking a constant amount of maintenance.”

WHAT

WHAT

SPM → monitoring

Logsene → logging

On PremisesCloud

http://sematext.com/spm http://sematext.com/logsene

WHAT SPM

Logsene

HOW

WHAT Agent

Java Node.jsWant Traces? Embed it!Collectd ⇒ SIGAR for OSFlume SpilloverChannelES API

WHAT Interesting finds

Variable Collectd supportCollectd ⇒ SIGARApache Flume Elasticsearch Stats APIMetrics 2nd class citizen

WHAT Transaction Tracing

Java Bytecode Instrumentation

Bottleneck finderAppMap maker

WHAT Custom Pointcuts

<method signature="java.lang.String com.company.example.Service#getUserName(com.company.model.Company company)"/>

Write-agg vs. Read-agg

Anomalies > Thresholds

WHAT Alerts

HeartbeatsThresholdsAnomalies

WHAT Anomaly Detection

ExponentialSTDFromMAKNN ...

boolean result = anomalyCount / (notAnomalyCount + anomalyCount) >= 3d / 4d;

WHAT Anomaly issues

Warn early / create noiseNormal abnormalitiesSlow change

Scalable Data Stores

http://blog.sematext.com/2015/06/09/docker-monitoring-support/

Logging

Hot vs. Cold

HOT COLD

Drop, don’t Delete

HOT COLDdrop

Pull, don’t Push

GET QUEUEpull

ES

Beware of Aggregations

Circuit Breakers

http://blog.sematext.com/2014/10/06/top-5-most-popular-log-shippers/

Thank you!

@[email protected]

@sematexthttp://sematext.com