SkySQL is the first and only database-as-a-service (DBaaS) engineered for MariaDB by MariaDB, to use a state-of-the-art multi-cloud architecture built on Kubernetes and ServiceNow, and to deploy databases and data warehouses for transactional, analytical and hybrid transactional/analytical workloads.
In this session, we’ll lay out the vision for SkySQL, provide an overview of its capabilities, take a tour of its architecture, and discuss the long-term roadmap. We’ll wrap things up with a live demo of SkySQL, including a preview of its deep learning–based workload analysis and visualization interface.
4. THE PROMISE
4
● Cloud ease of use
● MariaDB Community
● Development/testing
● Developer adoption
● Out of the box
● Self-service
● Automation
5. THE REALITY
5
Watered down
Out of date
No clustering
No sharding
Missing bug fixes
Missing security patches
Missing features
Antiquated
No columnar storage
No real-time analytics
Pseudo support
No technical support
No workarounds
No hotfixes
We’ll make do We’ll be patient We’ll find a way We’re in trouble
6. LIMITATIONS
6
● Unsupported/removed features
○ Multi-master clustering
○ Columnar format (analytics)
○ Transparent sharding
○ Delayed replication
○ SSD-optimized storage
○ GSSAPI auth (Active Directory)
● Missing features
○ Instant alter table (drop/reorder
columns)
○ Optimizer trace
○ Application-time period tables
Antiquated
● MariaDB Community Server
○ MariaDB Enterprise Server
○ MariaDB MaxScale
○ MariaDB ColumnStore
● Transactions only
○ No columnar, no object storage
○ No real-time analytics
○ No replication (row to columnar)
○ Requires separate services
(e.g., Amazon RDS + Redshift)
Watered down
7. PROBLEMS
7
● Lack of technical database expertise
and experience
○ 2018 Amazon Prime Day (Postgres)
○ Cannot resolve technical issues
■ No bug fixes
■ No security patches
■ No workarounds
○ Cannot provide recommendations
■ No best practices
■ No long-term strategic advice
Out of date
● Amazon RDS – MariaDB Server
10.4
● Azure Database – MariaDB Server
10.3
● Amazon Aurora – MySQL 5.7
● Google Cloud SQL – MySQL 5.7
MariaDB Server 10.3 released two (2) years ago
MySQL 5.7 released four (4) years ago
Pseudo support
8. WHAT THEN?
8
● Cloud ease of use
● MariaDB Community
● Development/testing
● Developer adoption
● Out of the box
● Self-service
● Automation
DBaaS
+
MariaDB
subscription
IaaS
+
MariaDB
subscription
Best
available
option…
until now
But…
10. VISION
10
● Path to multi, inter and hybrid cloud
○ Reduce risk, deploy to the same region of multiple clouds (e.g., US West – AWS, Azure and GCP)
○ Reduce cost, use a mix of clouds based on workload/price, adjust as needed (e.g., prices rise)
○ Improve productivity, let different departments use different clouds based staff expertise
● Databases for transactions, analytics and both (transactions + analytics)
○ Reduce costs and improve efficiency, no need for separate services (e.g., Amazon RDS + Redshift)
○ Innovate, add real-time analytics to transactional, customer-facing databases
● Access to the safest, most reliable and most current version of MariaDB
○ Reduce risk, deploy databases with the latest bug fixes and security patches
○ Innovate, take advantage of the latest and greatest features (e.g., application-time period tables)
● Deep-learning based workflow analysis and visualization
○ Reduce risk, prevent issues by detecting anomalies or changes in database usage
11. ENTERPRISE DBAAS
11
● Cloud ease of use
● MariaDB Community
● Development/testing
● Developer adoption
● Out of the box
● Self service
● Automation
SkySQL
➕ On-premises reliability
➕ MariaDB Enterprise
➕ Staging and production
➕ Enterprise governance
➕ Customizable
➕ Full service
➕ Expertise and experience
3rd party, first-generation DBaaS Native, second-generation DBaaS
12. SkySQL FIRSTS
12
Smart Transactions
The first DBaaS to support
transactional (OLTP),
analytical (OLAP) and hybrid
(HTAP) workloads for Smart
Transactions.
From the source
A DBaaS architected,
engineered, and supported
by database architects,
engineers, and extraordinary
support.
Mission-critical for all
Expansive Topologies for all
Applications Tiers.
SkySQL
Always current
The cloud offering the latest
bug fixes and security
updates, always.
13. WHAT’S DIFFERENT
13
● Kubernetes
● ServiceNow
● Prometheus and Grafana
● Deep learning
Database
● MariaDB Platform
● MariaDB Enterprise Server
● MariaDB MaxScale
● MariaDB ColumnStore
Support
● Database bugfixes and
security patches
● Database administration
with SkyDBA
● Databases for transactions
● Data warehouses for analytics (columnar)
● Databases for smart transactions (row + columnar)
● Transparent query routing
Features
Platform
15. ARCHITECTURE
15
● State of the art
○ Kubernetes for container orchestration and high availability
○ ServiceNow for configuration and workflow management
○ Prometheus and Grafana for real-time monitoring and visualization
● Modular
○ SkySQL portal in ServiceNow (six 9s high availability)
○ SkySQL databases in CSP (single tenant K8s clusters)
○ SkySQL control plane in CSP (sits between the portal and the databases)
● Secure
○ Compartmentalization, isolation and indirect access
○ Log all access and commands performed by MariaDB employees
16. SkySQL portal
Inventory
management
Configuration
management
Workflow
management
Organization
management
Coming soon Coming soon
SkySQL control plane
Google Kubernetes Engine
Cluster
Google Kubernetes Engine
Cluster
Google Kubernetes Engine
Cluster
Google Kubernetes Engine
Cluster
Database
Kubernetes API
SkySQL control plane SkySQL control plane
Google Kubernetes Engine
Cluster
Google Kubernetes Engine
Cluster
Google Kubernetes Engine
Cluster
Elastic Kubernetes Service
Cluster
Database
Google Kubernetes Engine
Cluster
Google Kubernetes Engine
Cluster
Google Kubernetes Engine
Cluster
Azure Kubernetes Service
Cluster
Database
17.
18. HIGH AVAILABILITY
18
● Kubernetes with MariaDB MaxScale
○ Best of both worlds: cloud native + MariaDB native
● Kubernetes provides self healing
○ Data stored on persistent disks (e.g., EBS/GPD)
○ Regional Kubernetes clusters span multiple zones
○ Failed instances automatically recovered
● MariaDB provides automatic failover
○ If the primary fails, the most up-to-date replica is automatically promoted
○ The failed primary is automatically recovered (by K8s), and becomes a replica
20. Regional Kubernetes cluster
HA: PERSISTENT DISKS
20
Zone a
MaxScale
Zone b Zone c
MariaDB
(replica)
MariaDB
(replica)
GPD
MariaDB
(primary)
21. Regional Kubernetes cluster
HA: AUTOMATIC FAILOVER
21
Zone a
MaxScale
Zone b Zone c
MariaDB
(replica)
MariaDB
(primary)
GPD
MariaDB
(primary)
22. Regional Kubernetes cluster
HA: SELF REPAIR
22
Zone a
MaxScale
Zone b Zone c
MariaDB
(replica)
MariaDB
(primary)
GPD
MariaDB
(primary)
MariaDB
(replica)
23. SMART TRANSACTIONS
23
● SkySQL is the reference cloud architecture for smart OLTP
○ OLTP – Data stored as rows on block storage (e.g., EBS/GPD)
■ Fast and persistent, same as others (e.g., RDS)
○ OLAP – Data stored as columns on object storage (e.g., S3/GCS)
■ Low-cost and unlimited capacity, same innovators like Snowflake
○ Smart OLTP – Data replicated from row/block storage to columnar/object storage
■ Transaction queries against row/block storage
■ Analytical queries against columnar/object storage
■ Any database instance can transactional queries, analytical queries or both
■ It’s all transparent to the application
27. OPTIONS
● Basic topologies (e.g., replicated)
● Basic HA (e.g., MaxScale self repair)
● Basic DR (e.g., restore instance)
● Basic workload analysis
● Standard menu options
● Standard instance types
SkySQL Power (customized)
● Custom topologies (e.g., tiered)
● Advanced HA (e.g., MaxScale failover)
● Enterprise DR (e.g., restore cluster)
● Advanced workload analysis
● Custom menu options
● Custom instance types
SkySQL Foundation (out of the box)
27
28. SkySQL Power
● Making the cloud work for your business, not the other way around
● Customize our DBaaS to meet your specific needs, not the other way around
● Powered by
○ People: the database experts at MariaDB
○ Technology: the configurability/extensibility of ServiceNow and Kubernetes
“Database deployments can be unique,
shouldn’t that be true in the cloud as well?”
28
29. Power customization
● Out of the box
○ Custom menu options
■ Restrict to approved options only (e.g., remove $$$ instance types)
■ Standardize on custom/in-house options (e.g., use custom instance types only)
● On demand
○ Custom HA/DR implementations and strategies
■ Region failover, deployed replication
○ Integration with cloud ecosystem
■ Kafka (e.g., Amazon MKS), Redis (e.g., Amazon ElastiCache)
○ Synchronization between databases and data warehouses
29
30. Custom instance types
● Custom instance types upon request
○ Any combination of vCPU and memory
○ Ideal for large working sets
■ Require more memory, but…
■ Not necessarily more vCPU
○ Can add memory without adding vCPU
■ Save 10% or more of the deployment cost
● Pricing
○ Compute – Up to 96 vCPU @$0.0558 / hour of vCPU
○ Memory – Up to 624GB @$0.0182 / hour of GB
30
32. Business Benefits
● Reduce costs with custom instance types
○ Add memory without paying for unnecessary vCPU
○ Operate at maximum resource utilization
● Reduce costs and accelerate migration to the cloud
○ Take advantage of lift and shift (easier and faster)
○ Remove the time and cost of architectural changes
● Avoid unnecessary trade-offs and compromises
○ Deploy to the cloud exactly what you deployed on premises
○ Continue to meet established enterprise HA/DR requirements
32
33. SkyDBA
33
● Certified cloud architects
● Expertise and best practices
● Consultative support
● Automation with oversight and intervention
● Database management and monitoring
● Proactive care
○ Health checks
○ Performance analysis
○ Security audits