Find out how Cozy Cloud uses the OVH Metrics Data Platform to monitor and optimise its SaaS service for the general public. From performance data aggregation to customer usage metrics, the Cozy Cloud teams will share their data-centric collaboration experience with you.
6. Services
• Metrics: time series at scale
• Loops: cron for queries
• Omni: Shinken for time series
• Studio: Tableau for time series
• Insights: OVH products’ metrics
8. Use cases
• DC temperature/electricity/cooling maps
• Pay-as-you-go billing (PCI/IPLB)
• GSCAN
• Monitoring
• ML model scoring (anti-fraud)
• Pattern detection for medical applications
9. Digital services at home
in the era of GAFA and GDPR
A PERSONAL DATA PLATFORM - CHANGE THE PARADIGM TO CHANGE THE RULES
12. Companies need to
develop closer digital relationships
with their customers…
… but the expectation set by GAFA through their way of
simplifying their customers’ daily lives is incredibly
high
13. … but without using Facebook
Messenger…
or WeChat
How?
… companies won’t be able to legitimately attract the
variety of data needed to offer a competitive experience
with their capacity alone
14. On the other hand, customers became dependent on
online services that make their daily lives easier.
They give up more and more of their digital autonomy,
since it’s the condition for accuracy and “convenience”.
2
… if Uber revolutionised the cabs market, it’s because of few details that
change everything: they have access to my credit card numbers, my
localisation, and drivers’ one to streamline the service.
15. …
GAFA… Energy… Telecom… Assurance… Government… IoT…
But this convenience is limited by numerous
silos
… while GAFA’s services are “seamless”, our digital lives are nightmares of
passwords, isolated services, unnumbered clicks to get your documents,
duplicates, and out-of-sync devices…
16. ?
Who is legitimately able to
provide you with a seamless
experience with all your data?
…
GAFA… Energy… Telecom… Insurance… Government… IoT…
17. You, and only you!
…
GAFA… Energy… Telecom… Insurance… Government… IoT…
19. 3 integrated blocks for a personal data platform
Automated data
collections
1
20. 3 integrated blocks for a personal data platform
2
A universal drive,
connected to devices
and applications
21. 3 integrated blocks for a personal data platform
Open a new, ethical data-centric economy where
Provider’s interests are "by design" aligned with users’ones
3
Digital services
at home
22. Cozy: trust is at the heart…
“You will stay because you can leave.”
1. Security: a necessary pre-requisite … but insufficient
2. Decentralised: break the profit/cost ratio for the newcomer
3. No blackbox effect: open source, as audits are possible
4. No lock-in: service portability and self-hosting
5. Prescribers: statements only engage those who believe them…
6. Clear positions: CGU and CGU synthesis!
7. Show what’s done with my data: more transparency…
8. A business aligned with users’ interests: make them subscribers…
25. OVH Metrics REX
A PERSONAL DATA PLATFORM - CHANGE THE PARADIGM TO CHANGE THE RULES
26. The goals
• First requirement was observability, driven by the “infra”
team:
– OS (CPU load, memory, disks, network, etc.)
– Middleware (haproxy stats, couchDB, etc.)
• Secondary use cases with other teams:
– Application performance monitoring
– Business KPIs
27. Making the choice
• Internal management
– InfluxDB
– Complex and expensive to manage
• Testing and using OVH Metrics
– Cost control
– Extensible
– Direct access to lower primitives
– More time to focus on more valuable tasks
28. OVH Metrics solution
• Data Storage
– Warp10
– Multi-protocol, supporting: graphite, influxdb, opentsdb, prometheus
• Restitution/Dataviz
– Grafana
– Each user has an account
• Data production
– Pushing into warp10
– Beamium, a scraping/pushing daemon, compatible with prometheus
exporters
30. Server monitoring
• Use of the Noderig exporter to produce the data
– Open source: https://github.com/ovh/noderig
– CPU, memory, disks, load, network
• Use of Beamium to push data
– Open source: https://github.com/ovh/beamium
– Scraping prometheus exporters
– Buffering and pushing into Warp10
– Handling errors and retries
• Beware of dynamic environments (containers)
34. Observability middleware
• Use of available open-source prometheus exporters
• Beamium to push data into Warp10
• Haproxy, Openstack swift, Ceph, CouchDB, Redis…
36. Application observability
• Code instrumentation with http endpoint for metrics
exposition in the prometheus format
• Use of beamium to scrap and push data to the platform
• Observability based on business KPIs and application
internals
38. Creating a new dashboard
• Needs assessment (sponsor)
• Produce the data (dev team)
• Exporter/Beamium setup to push data into Warp10 (infra
team)
• Dashboard setup in grafana (infra team)
• Use of the dashboard (sponsor)
39. Use case: Business observability
• Based on centralised access logs
• In-house development to generate data from the logs and
push them into Warp10
• Fallback via Beamium to handle eventual unavailability
40. A (very) complex need for (very) simple usability…
• Define metrics catalog in an application ecosystem (not
just one, but many verticals to monitor)
• Non-technical translation/vulgarisation
• Daily use to define trends and measure product attraction
• 1-click visualisation – no need to set up anything for the
marketing team
41. And a quick analysis of key performance indicators…
Marketing campaign impact
Monitoring usage
43. Volumes
• 30,000 active metrics (time series)
• 14M+ daily data points
• 1-year retention
• Plan L: 179.99€/month TTC
– Up to 100k active metrics & 288M daily data points
45. Conclusion
• A turnkey solution
– To store all observability data
– To visualise (dataviz) and query (warp10/warpscript)
• A readily available team
• Planned evolution of Cozy’s metrics
– Adapt observability to reflect infrastructure evolution
– Add more business KPIs (churn rate, etc.)
– Monitoring/alerting
• So far, monitoring and observability are managed by two different
services
• OVH Metrics Alerting to come