MySQL HeatWave (Oracle)

What MySQL HeatWave actually is

MySQL HeatWave is Oracle's managed MySQL service with an analytical engine bolted into the same database. The transactional side is standard MySQL (built on MySQL Enterprise Edition, full InnoDB and ACID support), and the analytical side is the HeatWave cluster: an in-memory, massively parallel, hybrid columnar engine that sits next to the MySQL node and receives transaction updates in real time. The pitch is that you run OLTP and OLAP in one place and skip the ETL pipeline into a separate warehouse. Oracle announced HeatWave in late 2020 and has since added a lakehouse layer over object storage, in-database ML (HeatWave AutoML), a vector store, and GenAI features on top.

The practical difference from bolting a warehouse onto MySQL yourself is where the data lives. Updates from transactions replicate into the HeatWave cluster in seconds, so analytical queries see current rows, not yesterday's export. Queries can also reach object storage through HeatWave Lakehouse, so cold or external data joins the same query surface. For reporting stacks that currently wait on nightly ETL jobs into BigQuery or Redshift, that architecture is the whole argument.

One detail we appreciate: it is still plain MySQL at the wire. Existing applications, drivers, BI tools such as Tableau and Looker, and DBA habits transfer without rewrites. You are choosing a hosting and acceleration model, not a new query language.

Pricing: what you really pay

Start with the free tier, because it is unusually generous. Oracle's Always Free tier includes a standalone HeatWave DB system with a single-node HeatWave cluster in your home region, 50 GB of storage, and 50 GB of backup storage, for an unlimited time — plus a US$300 cloud credit valid for 30 days on other OCI services. We have not seen another managed database hand out an in-memory analytical engine for free, and it is enough to run a small production app or a serious evaluation. Always Free instances do not get a public IP address; you connect through a compute instance or private networking.

Past that, pricing is consumption-based and layered, and the layering is where bills surprise people. The base MySQL DB System bills per OCPU-hour (Oracle is moving new users to an ECPU model where the unit blends CPU hours with work done — a meter you cannot fully cap by picking a shape) plus provisioned storage at about $0.04 per GB per month. Backup storage adds no charge up to 100% of total database storage, then costs about $0.04 per GB per month. The HeatWave analytics cluster is a separate, optional add-on billed per node; third-party licensing analysts put list pricing at roughly $3.20 per hour per node, and nodes are billed on allocated memory-hours, so a cluster sized for month-end reporting costs the same on a quiet Tuesday. High availability multiplies the base cost (around 2x for active-active per one pricing guide, and up to 3x once replica storage is counted), and HeatWave ML and GenAI carry their own per-OCPU meters.

The arithmetic that matters: a deployment that replaces both a transactional database and a separate analytics platform can land at 3x to 5x the base MySQL DB System bill once the cluster and ML layers are switched on. Reserved capacity cuts DB system rates meaningfully, and you can stop a HeatWave cluster outside analytics windows — but data is in memory, so restarting means reloading, and guides put that at 10 to 60 minutes. Our view: model the full three-layer bill before committing, and treat per-hour figures published by third parties as approximations to verify on the Oracle pricing page the day you sign.

Performance & infrastructure

Oracle's own benchmark claims are aggressive — the HeatWave Lakehouse marketing page cites 15x faster query performance than Amazon Redshift, 18x faster than Databricks and Snowflake, 35x faster than Google BigQuery. We treat vendor-published TPC-H and CH-benCHmark numbers as marketing until independently reproduced, but the architectural case is sound: an in-memory hybrid columnar engine with a scheduler that overlaps compute and network traffic will beat disk-based warehouses on hot analytical queries, and the Autopilot component handles indexing and thread-pool decisions that would otherwise need a DBA.

On availability, the service is fully managed — Oracle handles provisioning, patching, backups, recovery, and scaling. The published availability targets run from 99.99% with a three-node HA cluster across availability domains, 99.95% in a single AD, down to a 99.9% SLO for standalone configurations, and Oracle's features page markets 99.99% uptime via group replication and automatic failover. Failover is only as good as the topology you pay for; the standalone tier is not HA, whatever the marketing page implies.

Footprint is a strength. HeatWave runs natively in all OCI regions, on AWS (currently mapped across eight AWS regions — Mumbai, Seoul, Tokyo, Frankfurt, London, Queretaro, N. Virginia, and Oregon — each paired to an OCI region), and on Azure via the Oracle Database Service for Azure interconnect. Data persists to OCI Object Storage or AWS S3 depending on the platform. European, US, and Asia-Pacific coverage is genuine. Note the trade-off on AWS and Azure: some features are OCI-only, and the Azure path is an interconnect arrangement rather than a first-class Azure service.

Where it falls short

Pricing opacity is the biggest one. The move from OCPU shapes to a blended ECPU meter means part of the bill tracks runtime behavior, and Oracle's own conversion guidance has been criticized for over-provisioning by 15% to 30% if taken unmodified. Licensing analysts who review these contracts repeatedly recommend pulling the ECPU rate, the HeatWave capacity-unit rate, and the HA multiplier from the pricing page in writing at order time — advice we would follow. A dynamically rendered price table is not a rate commitment.

Second, the cost structure punishes curiosity. Every layer worth trying — the cluster, ML, GenAI — is a separate meter, and the cluster bills on allocation rather than use. Teams that enable features during a pilot and forget to disable them carry that cost indefinitely.

Third, there is real platform gravity. You can run HeatWave on AWS, but the full feature set and the deepest integrations sit on OCI, and the operational surface (console, IAM, Database Management, Ops Insights) is the Oracle stack. None of that is fatal, but it is a long-term commitment, not a drop-in stop.

Finally, this is a database service, not a general cloud. There is no bare metal, no GPU compute attached to the database, and no S3-compatible storage product of its own — it reads from object storage rather than providing it. The surrounding questions (DDoS posture, IPv6 reachability of endpoints) are handled at the OCI platform level and are not documented per service, so we cannot rate them here.

Who should pick this

We would pick HeatWave when a team already lives on MySQL and is genuinely paying to move data into a warehouse for reporting. If the ETL pipeline is the pain — nightly exports, stale dashboards, a second system to secure and audit — the consolidated model earns its keep, and the Always Free tier makes the evaluation cheap and honest. Mixed workloads that need sub-second analytics over live operational rows are exactly the target.

We would not pick it for a simple WordPress database or a small app with no analytical queries; the base MySQL DB System alone is fine but unremarkable value, and the cluster is an expensive ornament without analytics demand. Teams allergic to Oracle contracting should price the alternatives carefully first — but for MySQL shops with real analytics load, the convergence argument is strong.

Key takeaways

• Always Free tier includes a HeatWave DB system with a single-node cluster, 50 GB storage, and 50 GB backup storage, unlimited time — best free managed-analytics entry we have seen.
• Pricing is layered: base DB system per OCPU/ECPU-hour plus storage at about $0.04/GB/month, HeatWave cluster billed separately at roughly $3.20/hr per node on allocated memory, with ML and GenAI as extra meters.
• A full replacement for a transactional database plus a warehouse can cost 3x to 5x the base database bill; get the ECPU, capacity-unit, and HA rates in writing at order time.
• Availability is managed and well documented — 99.99% with multi-AD HA, but standalone runs on a 99.9% SLO.
• Runs natively on OCI and AWS (eight regions) with Azure via interconnect; some features are OCI-only.
• Still wire-compatible MySQL: existing applications, drivers, and BI tools work unchanged, which keeps migration risk low.

Sources

• https://www.oracle.com/mysql/free/
• https://www.oracle.com/heatwave
• https://www.oracle.com/mysql/features/
• https://blogs.oracle.com/mysql/mysql-cloud-services-cost-comparison-who-provides-the-best-value-2025
• https://blogs.oracle.com/mysql/post/introducing-heatwave-always-free
• https://dev.mysql.com/doc/heatwave-aws/en/heatwave-aws-regions.html
• https://downloads.mysql.com/docs/heatwave-en.pdf
• https://oraclelicensingexperts.com/blog/oracle-oci-mysql-heatwave
• https://www.oracle.com/a/ocom/docs/mysql/oracle-unleashes-heatwave-in-mysql-market.pdf