On this page
Short answer
Move a Cassandra 3.11 cluster to 5.0 in two rolling upgrades, via 4.0 or 4.1, with a stop to verify between them. Upgrades are rolling restarts, one node at a time, with no downtime when done correctly. The work is in the checks: Java, drivers, tables that use removed features, and the 5.0 compatibility-mode steps. This guide is for teams planning the move themselves or with Cassandra support alongside.
Move a Cassandra 3.11 cluster to 5.0 in two rolling upgrades, via 4.0 or 4.1, with a stop to verify between them. Upgrades are rolling restarts, one node at a time, with no downtime when done correctly. The work is in the checks: Java, drivers, tables that use removed features, and the 5.0 compatibility-mode steps. This guide is for teams planning the move themselves or with Cassandra support alongside.
What is the supported route from 3.11 to 5.0?
Go from 3.11 to 4.0 or 4.1, then to 5.0. The project's release notes say upgrades from 3.0.x and 3.x to 4.0 are supported from 3.0.13 or 3.11.0 onward, and that 4.0 or later cannot read pre-3.0 SSTables. So the first step is to be on a recent 3.11 patch. The Apache download page lists 5.0 (latest 5.0.9), 4.1 (4.1.12) and 4.0 (4.0.21), all released 7 August 2026, with 5.0 maintained until the 8.0 release and 4.1 until 7.0. Landing on 4.1 first gives you the longer-maintained intermediate stop.
| Step | From | To | Why separate |
|---|---|---|---|
| 0 | Older 3.x | Recent 3.11 patch | Earlier patches cause rolling-upgrade issues |
| 1 | 3.11 | 4.1 (or 4.0) | Protocol, Thrift and config changes |
| 2 | 4.x | 5.0 | Java 11 minimum, removed features, storage format |
Do not combine the steps in one maintenance window. A version change that fails is easier to diagnose when only one thing changed.
What should you check before touching a node?
- Snapshot every node. The release notes advise a snapshot before any upgrade; it takes almost no extra disk until compaction rewrites live files. Note that snapshots from the previous major cannot be restored with
sstableloaderuntil converted withsstableupgrade. - Java. Java 8 was removed in 5.0; the lowest supported version is Java 11, and Java 17 support is no longer experimental as of 5.0.5.
- Thrift and drivers. Thrift was removed in 4.0. Check that every application uses CQL and a current driver, and test protocol versions. The 4.x notes warn that protocol v3 had a bug that can appear in rolling upgrades from 4.1.0 or from 4.0.0 through 4.0.7; use protocol v4 or higher.
- Removed table features. DateTieredCompactionStrategy was removed in 5.0: move those tables to TimeWindowCompactionStrategy before upgrading. The
sstable_compressionandchunk_length_kboptions were also removed. - Credentials in config. Keystore and truststore passwords no longer default to "cassandra" in 5.0; set them explicitly in
cassandra.yaml. - COMPACT STORAGE. Still supported in 5.0, but the project says it will be removed in a later major, and
ALTER ... DROP COMPACT STORAGEis described as experimental and not recommended for production. Inventory these tables now. - Schema agreement.
nodetool describeclustershould show one schema version before you start. - Cluster health. No node down, no repair running, no pending streaming.
How do you run the rolling upgrade?
For each node in turn: stop traffic with nodetool drain, stop the service, install the new binaries and merge config, start it, and wait until nodetool status shows UN and logs are quiet before the next node. Repeat per datacenter. During the mixed-version period, our rule is no repair and no schema changes; keep ring operations such as bootstrap and decommission out of the window too. After all nodes move, run nodetool upgradesstables so data files are rewritten in the new format (the 5.0 release notes record a new SSTable format "oa" with an optional trie-indexed format "bti").
If you used incremental repair before an upgrade to 4.0, the project's 4.0 notes recommend a full repair afterwards to resolve inconsistencies.
What changes at the 5.0 step?
5.0 introduces storage_compatibility_mode, whose default is CASSANDRA_4, so a node stays rollback-compatible with 4.x. The release notes give three stages:
- Rolling upgrade to 5.0 with the default mode. The TTL limit stays at 2038 and you can still roll back.
- Rolling restart with
storage_compatibility_mode=UPGRADING. Once all nodes are on storage version 5, the TTL limit moves to 2106. - Rolling restart with
storage_compatibility_mode=NONE. Mixed 2038 and 2106 nodes are no longer possible, and rollback is not available.
Keep the cluster on stage 1 for a soak period, with the same checks as before. The property has to be set for all executables and tools. If you use very long TTLs (10 to 20 years), read the note on CASSANDRA-14092 first: inserts with TTLs past the maximum are rejected by default.
What can you adopt after the upgrade?
Only after the cluster is stable on 5.0, not on upgrade day: the Unified Compaction Strategy, storage-attached indexes and trie-based SSTable format. Cassandra 5.0.8 also added an optional Auto Repair scheduler that needs a JVM property set on every node. Each is a separate change with its own rollback story.
When to bring in help
Bring in help for rehearsal, not just rescue: a staging ring cloned from production, driver tests and a written rollback per stage. AceMQ provides 24/7 Cassandra support across the Americas, EMEA and APAC with response targets from 15 minutes, by plan, and replies to quote requests within 1 hour. Support typically runs $24,000-72,000 a year (USD), by SLA, scope and deployment. The operations checks that run alongside an upgrade are in the repair, compaction and tombstones guide, and the version overview is in Cassandra support options.
Sources
- Apache Cassandra download page (cassandra.apache.org/_/download.html), read 17 Nov 2026.
- Cassandra NEWS.txt, branch cassandra-5.0 (github.com/apache/cassandra/blob/cassandra-5.0/NEWS.txt), read 17 Nov 2026: Java 11 minimum, DTCS removal, Thrift removal in 4.0, storage_compatibility_mode steps, supported upgrade from 3.0.13/3.11.0, SSTable format, COMPACT STORAGE, full repair after incremental pre-upgrade, Auto Repair in 5.0.8.
Frequently Asked Questions
Can I upgrade Cassandra 3.11 directly to 5.0?
The project's notes document 3.x to 4.0 and 4.x to 5.0 as the tested routes. Plan 3.11 to 4.x, verify, then 5.0.
Do I need downtime?
No. Upgrades are rolling, one node at a time, if clients tolerate a node restarting and the cluster is healthy.
What is the minimum Java version for Cassandra 5.0?
Java 11. Java 8 was removed.
Is Thrift still available?
No. Thrift was removed in 4.0.
What is storage_compatibility_mode?
A 5.0 setting that keeps nodes compatible with 4.x until you move through UPGRADING to NONE.
Talk to an engineer
Need a hand with Cassandra?
Senior engineers who run it in production. Get a quote or a free consultation, no sales pitch.
