ScyllaDB database services for production teams.
JusDB provides ScyllaDB database services for scoped projects, ongoing remote DBA coverage, production support, performance tuning, upgrades, and migrations from Cassandra or DynamoDB. We work on your own servers and cloud VMs, on Kubernetes with the ScyllaDB Operator, and with ScyllaDB Cloud. 24x7 on-call is available under your service agreement.
Or request a free audit of one database instance. Already in an incident? See how incidents are handled.
- NDA before access
- Read-only by default
- Changes measured against your baseline
ScyllaDB version status
Pick the version you run
ScyllaDB 2026.3
Newest releaseA feature release, generally available since September 2026, and supported until 2027.1 ships.
From 2026.2, vnode keyspaces can move to tablets online, within limits such as a single datacenter and Incremental Compaction Strategy tables.
Plan the move to tabletsScyllaDB 2026.1 LTS
Newest LTSThe newest long-term support (LTS) release, generally available since March 2026, and supported until 2028.1 ships.
ScyllaDB supports the last two LTS releases, and an upgrade cannot skip one, so a cluster on 2025.1 goes to 2026.1 next.
Plan an upgrade to 2026.1ScyllaDB 2025.1 LTS
Supported LTSThe first source-available release, generally available since April 2025, and supported until 2027.1 ships.
Releases after 2025.1 require consistent topology updates through Raft, so we confirm they are enabled before the upgrade.
Plan the upgrade off 2025.1ScyllaDB Open Source 6.2
Last AGPL releaseReleased in October 2024. It is the last open-source (AGPL) release and stays under the AGPL.
The upgrade path from 6.2 is 2025.1, free only within 50 vCPUs and 10 TB of storage per organization.
Keep 6.2 running while you planVersion status from the ScyllaDB version support page, the upgrade policy, and the source-available license FAQ, checked 28 October 2026.
- Platforms
- Your own servers, cloud VMs, Kubernetes with the ScyllaDB Operator, and ScyllaDB Cloud
- High availability
- Racks and datacenters with NetworkTopologyStrategy, LOCAL_QUORUM, Raft for schema and topology, and scheduled repair
- Incidents
- Initial-response targets are set in your service agreement. See the targets
Symptoms
Eight ScyllaDB problems we are asked to fix
Each one has a cause we can measure. The scoping call starts from the symptom you see and the version and platform you run.
One CPU core runs at its limit while the rest sit idle because one partition key takes most of the requests, and each partition is served by one shard on each replica.
Deleted rows come back after a node outage because the table uses tombstone_gc timeout mode, and no repair covered it within gc_grace_seconds (10 days by default).
One partition grows without bound because its key has no time bucket, and by default ScyllaDB warns only once a partition passes 1,000 MB or 100,000 rows.
A datacenter goes down and node replacement stops because a cluster with two equal datacenters loses its Raft majority, which schema and topology changes need.
Disks run short during compaction because size-tiered compaction (STCS) can need up to half the disk free, where Incremental Compaction Strategy (ICS) needs far less.
A change data capture (CDC) consumer misses changes after falling behind, because rows in the CDC log table expire after 24 hours by default.
The upgrade off Open Source 6.2 keeps slipping because 6.2 was the last AGPL release, and later releases are free only within 50 vCPUs and 10 TB of storage per organization.
DynamoDB calls fail after a move to Alternator because Alternator does not implement multi-item transactions such as TransactWriteItems, PartiQL, or the DynamoDB backup APIs.
If one of these is costing your team time, a scoping call is the place to start.
Book a 30-min scoping call (opens Google Calendar in a new tab)Hot shard
How one hot partition stalls a ScyllaDB shard
ScyllaDB splits each node into shards, one per CPU core, and each partition is served by one shard on each replica. When one key takes most requests, its shard saturates while the others idle, and more nodes do not split it.
One node, three of its shards, three outcomes
Choose a state to see where the requests land.
Application and routing
A shard-aware driver sends each request straight to the shard that owns its partition.
Shard 0
Many partitions, none dominant
Busy, not saturated
Shard 1
Many partitions, none dominant
Busy, not saturated
Shard 2
Many partitions, none dominant
Busy, not saturated
Requests spread over many partitions, so each core carries a similar share of the work.
Application and routing
Most requests carry the same partition key, so they all queue on the shard that owns it.
Shard 0
Owns the hot partition
At full load
Shard 1
Owns other partitions, little traffic
Mostly idle
Shard 2
Owns other partitions, little traffic
Mostly idle
Requests for the hot key queue on one shard, so every partition on that shard slows down. The same happens on each replica, and more nodes do not help.
Application and routing
The application splits the hot key into 8 bucket partitions and reads all 8 to answer a query.
Shard 0
Owns some of the 8 buckets
Busy, not saturated
Shard 1
Owns some of the 8 buckets
Busy, not saturated
Shard 2
Owns some of the 8 buckets
Busy, not saturated
The buckets hash to different token ranges, so the load usually lands on several shards. Each read touches every bucket, and one bucket can still run hot.
We alert when one shard runs far busier than the others, then find the key behind it with nodetool toppartitions. We redesign that key with your developers, adding a bucket where one key can take a large share of traffic.
For a key that can still spike, we set per_partition_rate_limit (max_reads_per_second and max_writes_per_second), so ScyllaDB rejects part of the excess to reduce the overload. It is protection, not an exact cap: the accepted rate can be up to the replication factor times the limit. We rehearse each change on a copy.
What we cover
Full-spectrum ScyllaDB engineering
From a saturated core to an upgrade off 6.2: nodetool, ScyllaDB Monitoring Stack, and repair history drive every change.
Cluster architecture
Clusters laid out for your data, failure domains, and growth.
- Tablets or vnodes chosen when each keyspace is created
- Replication factor and racks set per datacenter
- A third datacenter, which can be one zero-token node, to keep a Raft majority
Performance tuning
Latency traced to the shard, partition, or compaction behind it, and measured against your baseline.
- Hot partitions found with nodetool toppartitions
- A compaction strategy matched to each table's writes
- Workload prioritization, with more shares for live traffic than analytics
Migration from Cassandra
Moves from Apache Cassandra over CQL, with a tested way back (a rollback plan).
- CQL schema and driver compatibility checked first
- Dual writes, then a bulk load with ScyllaDB Migrator
- Row counts and sampled reads compared before cutover
Alternator setup
The DynamoDB-compatible API (Alternator) set up for the calls your application makes.
- A write isolation policy per table, from always_use_lwt to forbid_rmw
- Load balancing in front of the nodes
- Unimplemented calls, such as TransactWriteItems, found before cutover
High availability
Replication and repair designed around the data loss and downtime you can accept (RPO and RTO), then drilled.
- Racks and datacenters laid out so one failure leaves a quorum
- Repair scheduled with ScyllaDB Manager, or automatic repair on tablets
- tombstone_gc repair mode, which drops a delete marker only after repair
Managed operations
Backups, monitoring, and upgrades under an agreed scope, with restores tested on a copy.
- Backups to Amazon S3 or Google Cloud Storage with ScyllaDB Manager
- ScyllaDB Monitoring Stack dashboards and alerts
- Rolling upgrades, one node at a time
Incident support
A critical (P1) page, answered within your agreement's target
JusDB keeps your recovery procedures (runbooks) written down, so an incident starts from a known plan instead of a blank page.
Initial-response targets
2 to 4 hours
Business hours
During covered business hours
15 minutes
24x7 on-call
P1 incidents
Production database down, primary unreachable, data corruption risk, or active failover event.
Severe latency spike, replication lag climbing rapidly, connection exhaustion, or degraded HA node.
Non-critical slow query regression, staging issue, backup warning, or replication check alert.
Configuration question, index advisory, planned schema review, or general operational request.
Our engineering team is in India; business-hours coverage is Mon–Fri 09:00–18:00 IST (UTC+5:30), and 24x7 on-call is available under your service agreement.
Initial-response targets, not resolution guarantees. The target for your engagement is set in your service agreement.
What happens on a P1 page
Page
You page the on-call engineer
Through the escalation path your service agreement names.
Triage
Root cause from per-shard metrics, logs, and tracing
Saturated shards, compaction backlogs, repair gaps, and timeouts traced to the table and partition behind them.
Mitigate
Stabilize first, from the runbook
Shed load on a hot partition with a per-partition rate limit, throttle compaction, pause scheduled repairs, or stop CQL traffic to a struggling node with nodetool disablebinary, then fix the cause.
After
A review after every P1
What happened, why, and the change that stops it happening again.
Before and between incidents
- Your setup, written down firstDuring onboarding we record your versions, datacenters, racks, replication factors, and repair and backup schedules, so an incident starts with context.
- One escalation path, in writingYour service agreement names how to reach the on-call engineer and in which hours.
- Reviews between incidentsWe review shard balance, large partitions, compaction, repair coverage, free disk, and backups, and report open risks with next steps.
- Version dates trackedWe track ScyllaDB support dates and license limits for the releases you run, and plan upgrades before support ends.
Process
How a ScyllaDB engagement runs, step by step
Written findings come before any change. Your team approves production changes, except routine actions you have pre-approved.
Scoping call
Your clusters, versions, datacenters, platforms, and what you need covered.
Read-only review under NDA
Our engineers use named accounts with only the permissions the review needs, then read your schema, scylla.yaml, nodetool output, ScyllaDB Monitoring Stack dashboards, logs, and ScyllaDB Manager history.
A written report before any change
Each finding comes with the change we propose, how we will measure it, and a way to undo it (a rollback plan).
Changes, then handover
We rehearse risky changes on a copy, apply them in windows you approve, measure them against the baseline, and update the operating instructions (runbooks).
Step one is a 30-minute scoping call.
Book a 30-min scoping call (opens Google Calendar in a new tab)Ways to work
Choose the ScyllaDB service your team needs
Start with the responsibility you need covered. Each engagement has its own scope, deliverables, access requirements, and acceptance criteria.
Projects
Scoped ScyllaDB projects
A defined piece of work with written findings, hands-on changes, and validation against your baseline.
- ScyllaDB consultingShard placement, compaction, repair, and hot partitions, reviewed as one scoped project.
- ScyllaDB on KubernetesScyllaDB Operator deployments, storage, racks, backups, and upgrades.
- Migration from Cassandra or DynamoDBOver CQL or Alternator, with validation and a rollback plan.
- Performance tuningHot partitions, compaction, and slow queries, measured before and after.
- High availabilityRacks, datacenters, quorum, and repair, drilled on a copy.
- Backup and disaster recoveryScyllaDB Manager backups, restore drills, and recovery steps written down.
Ongoing
Ongoing coverage and support
JusDB monitors the databases, carries out the agreed maintenance, and responds to incidents during the coverage the proposal names. The team keeps a change record, recovery procedures, and a review after every P1.
Where your ScyllaDB runs, and the tools around it
Your team and ours
What a managed ScyllaDB service still leaves to your team
Managed cloud databases automate much of the infrastructure layer, but your team still owns query performance, capacity, cost, and incident decisions.
| Question | Managed cloud service | Your own servers or Kubernetes |
|---|---|---|
| Servers, storage, and patches | The provider runs them. | Your team, or JusDB under your agreement. |
| Backups and restores | Built in on ScyllaDB Cloud: a full backup daily, kept for 14 days by default. You still test a restore. | Your team runs ScyllaDB Manager backups to object storage and tests the restores. |
| Data model, partitions, and queries | Yours. The provider runs the platform, not your tables or queries. | Yours. |
| Where the data runs | In the provider's cloud account, or in your own AWS or Google Cloud account with the bring-your-own-account option. | On servers, VMs, or Kubernetes nodes you run. |
| Moving off later | Over CQL or Alternator, read by a migration tool you run. | Backups, snapshots, or a read over CQL move the data elsewhere. |
Your team keeps control
- Set business priorities, application requirements, and release plans
- Approve production changes, except routine actions you have pre-approved
- Own your cloud accounts, access policies, and vendor relationships
- Decide when maintenance can happen and how quickly service must recover
- Move the work in-house or to another provider after your agreement's notice period, with your data returned or deleted and the transition help your agreement sets out (Terms, section 8)
JusDB handles the agreed database work
- Monitor database health, investigate alerts, and involve the right people
- Review query performance, replication, capacity, and backup health
- Carry out approved changes and update the operating instructions they affect
- Report completed work, open risks, and recommended next steps
Weighing ScyllaDB Cloud against running ScyllaDB yourself? We compare both against your workload on a scoping call.
Book a 30-min scoping call (opens Google Calendar in a new tab)Further reading
Two modeling guides and a comparison
Guides to designing CQL and DynamoDB tables from your queries, and a comparison with Apache Cassandra.
Comparison
Cassandra vs ScyllaDB
When each one fits, their failure modes, and how to move between them.
Guide
Cassandra data modeling: partition keys and query patterns
Tables designed from queries, with bounded time buckets. The same CQL modeling applies on ScyllaDB.
Guide
DynamoDB single-table design and indexes
Keys and indexes designed from access patterns, as an Alternator table needs.
FAQ
Common ScyllaDB questions
What buyers ask about Cassandra, DynamoDB, performance, security, and monitoring.
Still have a question?
Ask it on a scoping call. No production access needed.
Book a 30-min scoping call (opens Google Calendar in a new tab)How does ScyllaDB compare to Apache Cassandra?
JusDB treats ScyllaDB as a Cassandra-compatible engine. It speaks CQL and works with Cassandra drivers. It is written in C++ and runs one shard per CPU core, with no Java virtual machine (JVM) and no garbage-collection pauses. Tuning moves to shard balance, compaction, and repair.
What is the migration path from Cassandra to ScyllaDB?
JusDB first checks your schema and drivers against ScyllaDB's CQL support. Then we start dual writes from the application, bulk-load existing data with ScyllaDB Migrator, and compare both clusters before cutover. We rehearse the cutover on a copy, with a written rollback plan. JusDB does not guarantee one cutover duration.
Can ScyllaDB replace DynamoDB using Alternator?
JusDB answers per application: Alternator, ScyllaDB's DynamoDB-compatible API, covers most of the API, not all of it. It lacks multi-item transactions (TransactWriteItems and TransactGetItems), PartiQL, and the backup and point-in-time restore APIs, and it ignores provisioned throughput. We check each call your application makes before cutover.
What is ScyllaDB's performance advantage?
JusDB explains it as a design, not a promised number. ScyllaDB is written in C++ on the Seastar framework, and each CPU core gets its own shard, memory, and scheduler. That avoids JVM garbage-collection pauses and most locking between cores. We measure the gain on a copy of your workload.
How do you handle ScyllaDB security?
JusDB first turns on authentication and authorization, because the defaults, AllowAllAuthenticator and AllowAllAuthorizer, perform no checks. We then set least-privilege roles, encryption in transit (TLS) and at rest, and audit logging. Alternator does not enforce authorization by default either, so we turn on alternator_enforce_authorization. These controls can support your compliance programs, but do not certify compliance.
What monitoring do you provide for ScyllaDB?
JusDB sets up ScyllaDB Monitoring Stack (Prometheus, Grafana, and Alertmanager), with alerts on latency, shard balance, compaction backlog, repair age, free disk, and failed backups. Each critical (P1) incident ends with a written review. Initial-response targets are 2 to 4 hours during covered business hours, Mon–Fri 09:00–18:00 IST (UTC+5:30). Where your service agreement includes 24x7 on-call, the initial-response target for P1 incidents is 15 minutes. Targets are not resolution guarantees.
Next step
Talk to a ScyllaDB engineer about your databases
Tell us what is slow, fragile, or due to change. The first 30-minute call covers your databases, current ownership, and required coverage. No production access is needed. Or request a free audit of one database instance, subject to availability: mutual NDA before read-only access, a prioritized report within one week after access is ready, and no obligation.
NDA before access, read-only by default, and written scope first
Compare with other databases
If you are weighing alternatives, these are the databases most often compared with this one.
Cassandra
The database ScyllaDB is built to be compatible with. Compare CQL coverage, repair and compaction, and the JVM tuning ScyllaDB does not need.
DynamoDB
ScyllaDB's Alternator API is compatible with DynamoDB, so teams compare the two on cost and on where the data runs.
Aerospike
A key-value database with its index in memory and data on flash, on the same shortlist for low-latency work at high throughput.
MongoDB
A document model rather than wide-column, compared when queries are less predictable than a partition-key design allows.