Deep.BI/Apache Druid

Production Apache Druid builds. Patched by us, supported 24/7.

We keep patched Druid builds for five release lines at once, and one of our engineers created the Kubernetes operator that later moved into Apache Druid. The same engineers answer P0/P1 incidents in under 1 hour, any hour of the day.

  • 5Druid release lines patched
  • <1 hP0/P1 response, 24/7
  • 300+production clusters
Now on shift

100+ enterprises served, including

The scale we support

300+production clusters supported across Druid, StarRocks and Flink
20countries on 5 continents where our clients run production
10 yearsin business, founded in 2016

Atos, security and observability platform

"It has truly been a pleasure working with Deep.BI over the years. Their support for our security and observability platform, along with the underlying pipeline of Druid, Flink, Kafka, and NiFi, has been invaluable."

Ravi Raman, Global Head, AIsaac Product Engineering

  • Apache Druid
  • Apache Flink
  • Apache Kafka
  • Apache NiFi

Druid software we wrote

  • Druid operatorOne of our engineers created the Kubernetes operator for Druid and maintained it for years. It moved into Apache Druid in October 2025.
  • 7 extensionsOpen-source Druid extensions we wrote, among them an exact distinct count aggregator, an IP range filter and OpenTelemetry metrics. Released for Druid 27.0.0 to 38.0.0.
  • Exact percentilesThe Druid exact percentile extension was built by Deep.BI and is now maintained by NTT DATA.

Our Druid builds run in production

When a CVE lands or a fix can't wait for upstream, we patch the version you run, test it on your workload and hand you the build. Then we send the fix upstream. Everything stays Apache 2.0.

Patched builds we maintain

  • 31, 34-37Five Druid release lines, patched at the same time
  • Sep 2026Log4j, Jackson, Netty and ZooKeeper CVE fixes shipped to all five lines
  • 27.0.0-38.0.0Our extensions, released for each Druid version

Merged into Apache Druid, for example

  • MSQCap on MSQ controller tasks
  • BrokerFix for a Broker thread leak
  • CWE-1333RE2/J regex engine, closing a ReDoS hole
  • ArraysNPE fix in array_contains and array_overlap

What happens when you raise a P0

On 24/7 plans, every hour of the day, with no cap on incidents.

  1. 1You raise a P0 or P1On the channels set up at onboarding
  2. 2An engineer is on itUnder 1 hour, any hour of the day
  3. 3Service restoredMitigation first, then the deeper fix
  4. 4Root cause foundIn whichever layer it sits
  5. 5RCA and reviewPatched build if needed

Whatever layer breaks, we follow it down

A slow Druid query can be a bad segment layout, an overloaded Historical, a Kubernetes limit or a dropped packet. We stay with it until we find the root cause.

Layer
Vendor support
Deep.BI
Application and queriesDruid SQL, native queries, ingestion specs
Not covered
Covered
PipelinesKafka ingestion, batch ingestion, NiFi, OpenTelemetry
Not covered
Covered
DatabaseDruid services, deep storage, metadata store
Its own product only
Covered
Kubernetes or bare metalDruid operator, storage, Terraform, Ansible
Not covered
Covered
NetworkCalico, Cilium, TLS, mTLS
Not covered
Covered

Where Druid sits in a real-time pipeline

Druid reads streams straight from Kafka and loads files in batch. We also support the systems that feed it and the tools that query it.

Swipe to see the full pipeline

The Druid problems we fix most often

Symptoms we see in production, and how we fix them.

SymptomWhat's happeningWhat we do
kafka ingestionLag from KafkaSupervisor tasks fall behind the topic under peak traffic, and handoff slows down.Task count, partition alignment and handoff settings matched to the topic.
segmentsToo many small segmentsStreaming ingestion leaves many small segments, so queries fan out and the metadata store grows.Compaction, segment granularity and partitioning set for your query patterns.
historicalsSegments not loadingHistoricals run short of disk or the load queue backs up, so recent data goes missing from queries.Load rules, tiering and Historical capacity planned together.
queriesTimeouts and slow dashboardsQueries scan too many segments, or the Broker merges more data than it should.Rollup, partitioning, caching and Broker sizing, before anyone adds hardware.
metadataCoordinator and Overlord slowdownsOld task logs, audit entries and unused segment records pile up in the metadata store.Cleanup with the druid.coordinator.kill and druid.indexer.logs.kill settings, then a check of ZooKeeper or Kubernetes discovery.
upgradesUpgrade backlogClusters stay on old Druid versions because nobody owns testing the next one.We do: tested on your workload, with a rollback path.

Vendor-level support for open-source Apache Druid

One team from the first design to the next upgrade. Pick a stage.

Architecture

We design ingestion, segment layout, rollup and tiering for your query patterns, and size Brokers, Historicals and MiddleManagers or Indexers to match.

  • Sizing for your data and concurrency
  • Ingestion specs, rollup and segment granularity
  • Hot and cold tiers with load rules

Best forNew clusters and redesigns

Deployment and security

We deploy Druid with the Kubernetes operator, on bare metal or in your cloud, and set up TLS, authentication and authorization before it carries production traffic.

  • Druid operator or bare metal
  • TLS, LDAP and Ranger authorization
  • Terraform and Ansible

Best forTeams going to production

Migration

Phased, with both clusters running side by side until the numbers match. Your 24/7 SLA starts before cutover. Deep Segment Reader reads existing Druid segments in Spark when data has to move or be backfilled.

  • Inventory of every feature you use
  • Ingestion, query and pipeline migration
  • Side-by-side run until results and latency match

Best forTeams leaving a vendor distribution

Performance and cost

We tune segments, compaction, caching and query patterns on your workload before anyone adds hardware. Deep Predictive Cache adds an external cache layer for the dashboards that run the same queries all day.

  • Segment sizing and compaction
  • Query, Broker and cache tuning
  • Historical tiering for cost

Best forClusters that got slow or expensive

Upgrades

We plan and run Druid upgrades, test them against your workload first, and keep a rollback path open until the new version is stable. Our extensions are released for each Druid version, so they don't hold you back.

  • Upgrade plan and workload testing
  • Rollback path until stable
  • Every Druid version, old ones included

Best forTeams stuck on an old release

Druid operator, Predictive Cache and Segment Reader

Three ways to run Druid in production

Compared on what matters after the contract is signed.

Swipe to compare

Deep.BIRun it in-houseVendor distribution
EngineUpstream open source, plus custom fixes and featuresUpstream open sourceVendor build on the vendor's release cycle
Cost as usage growsFlat: set by node count, or by the hour poolMore headcountGrows with your footprint
Fixes and CVEsPatched builds under the SLA, fixes contributed upstreamCommunity timeline, no commitmentThe vendor's release cycle
Problems across the pipelineOne team stays on it until the root cause is foundSeparate expertise for every technologyProblems bounce between vendors
LeavingEnd the support contract. The cluster keeps running.No contract to exitMigrate off the vendor build

Plans

All four come with senior Druid engineers. They differ in who owns upgrades and who runs the cluster.

Swipe to compare plans

24/7 Enterprise SLA24/7 SLA + MaintenanceManaged ServiceSupport hours
P0/P1 responseUnder 1 h, 24/7Under 1 h, 24/7Under 1 h, 24/7Next business day, or under 1 h on the 24/7 option
Production incidentsNo capNo capNo capFrom your hours
Upgrades and CVE fixesFrom your hoursIncludedIncludedFrom your hours
Who runs the clusterYour teamYour team, maintained by usWe doYour team

Or start with a fixed-price assessment

50 hours, $19,000. Move to a 24/7 plan within 90 days and we waive the onboarding fee.

Book an assessment

Full plans and pricing

Questions teams ask first

How fast do you respond to a production incident?

On the 24/7 Enterprise SLA, the 24/7 SLA + Maintenance plan and the Managed Service, P0 and P1 incidents get a first response in under 1 hour, at any time of day, with no cap on production incidents. A senior Druid engineer handles the ticket from first response to resolution, and every incident ends with a root cause analysis. Support hours plans come with next business day response, or P0/P1 under 1 hour on the 24/7 option. Response targets are written into your agreement.

Do you support open-source Apache Druid or a vendor distribution?

We support open-source Apache Druid, the Apache Software Foundation project under the Apache 2.0 license, on your cloud, on premises or on Kubernetes. If you run a vendor distribution today, we can support the move to open source and start your 24/7 SLA before cutover.

Can you patch a CVE without forcing us to upgrade?

Yes. On the 24/7 SLA + Maintenance plan and the Managed Service, we monitor CVEs that affect Druid and its dependencies, assess the impact on your deployment and patch the version you run. We keep patched builds for five Druid release lines at once, so a security fix doesn't have to wait for an upgrade project. On the 24/7 Enterprise SLA, this work is billed from your pool of support hours.

Which Druid versions do you support?

All of them, including releases that are no longer current upstream. If you're on an older version, we keep it stable and patched while we plan the upgrade with you.

What happens if we decide to leave?

You end the support contract and your cluster keeps running. You run standard open-source Apache Druid on your own infrastructure, so there is no proprietary build or platform license to unwind. Notice periods and renewal terms are set out in your agreement.

Talk to a Druid engineer

Send your Druid version, cluster size and the problem using the form below. A Druid engineer reads it and replies.

  1. 1Technical call A Druid engineer reviews your architecture and recent incidents.
  2. 2Scope and proposal Node count, environments, plan and response targets in writing.
  3. 3Contract NDA, MSA and SOW, plus your security questionnaires.
  4. 4Onboarding Access, runbooks and escalation paths tested before coverage starts.
Talk to a Druid engineer

Let's help you

run Druid cheaper and quieter

Contact us:
Email:
druid@deep.bi

Get in touch and we'll reach back out to you soon!

Deep.BI needs the contact information you provide to us to contact you about our products and services. You may unsubscribe from these communications at any time. For information on how to unsubscribe, as well as our privacy practices and commitment to protecting your privacy, please review our Privacy Policy.
Thank you! Your submission has been received!
We’ll get back to you shortly!
Oops! Something went wrong while submitting the form.
Please try again.
Contact information:
Email:
druid@deep.bi