ScyllaDB 2025.1 LTS Expert

Creating a keyspace with tablets (default behavior):

-- Tablets are used automatically in 2025.1+
CREATE KEYSPACE my_ks WITH replication = {
    'class': 'NetworkTopologyStrategy',
    'dc1': 3
};

-- Explicitly set initial tablet count
CREATE KEYSPACE my_ks WITH replication = {
    'class': 'NetworkTopologyStrategy',
    'dc1': 3
} AND tablets = {'enabled': true, 'initial': 128};

-- Disable tablets for a specific keyspace (use vnodes)
CREATE KEYSPACE legacy_ks WITH replication = {
    'class': 'NetworkTopologyStrategy',
    'dc1': 3
} AND tablets = {'enabled': false};

Tablet benefits realized in 2025.1:

• Add multiple nodes simultaneously (not sequentially)
• Faster scaling -- new nodes serve reads/writes as soon as the first tablet migrates
• Mixed instance types in a cluster (tablets auto-balance across heterogeneous hardware)
• Per-table replication groups (different tables can have different replica placements)

Limitations in 2025.1:

• Counters are NOT supported with tablets (use vnodes for counter tables) -- fixed in 2026.1
• CDC with tablets is experimental
• Migration from vnodes to tablets requires recreating the keyspace

Tablet Merge

New in 2025.1. Complements the existing tablet split feature:

• Split: When a tablet grows beyond its threshold, it splits into two tablets
• Merge: When a table shrinks (data deleted, TTL expired), tablets merge to reduce overhead
• Merge reduces the number of tablets for a shrinking table, reclaiming metadata and scheduling resources
• Automatic -- no manual intervention required

Strongly Consistent Topology Updates (Raft)

All topology operations are now sequenced through Raft consensus:

What changed:

• Node join, decommission, removenode, and other topology operations go through Raft Group 0
• Operations are serialized -- no concurrent conflicting topology changes
• This is the default for new clusters in 2025.1
• Existing clusters must explicitly enable Raft topology after upgrading

Enabling Raft topology on existing clusters (after upgrade to 2025.1):

# Verify all nodes are on 2025.1
nodetool version   # run on each node

# Enable Raft topology (run once, from any node)
# This is a one-time, irreversible operation
curl -s -X POST http://localhost:10000/storage_service/raft_topology/upgrade

Monitoring Raft status:

SELECT * FROM system.raft;
SELECT * FROM system.topology;

curl -s http://localhost:10000/storage_service/raft_topology/status

Benefits:

• No split-brain during topology changes
• Safe to add/remove multiple nodes concurrently (with tablets)
• Topology operations are idempotent and recoverable

Strongly Consistent Auth Updates (Raft)

RBAC operations are now strongly consistent:

What changed:

• CREATE ROLE, GRANT, REVOKE, and other auth operations go through Raft
• No risk of auth metadata getting out of sync between nodes
• No need to repair system_auth after adding a datacenter
• Safe to run RBAC commands in parallel from any node

Before 2025.1:

# Was required after adding a DC:
nodetool repair system_auth

After 2025.1:

# No longer needed -- auth is strongly consistent via Raft

Strongly Consistent Service Levels (Raft)

Workload prioritization settings are now consistent via Raft:

-- Service levels are consistent across all nodes immediately
CREATE SERVICE LEVEL gold WITH timeout = '5s' AND workload_type = 'interactive';
CREATE SERVICE LEVEL silver WITH timeout = '30s' AND workload_type = 'batch';

-- Attach to roles
ATTACH SERVICE LEVEL gold TO 'app_user';
ATTACH SERVICE LEVEL silver TO 'analytics_user';

-- List service levels
LIST ALL SERVICE LEVELS;

-- Detach
DETACH SERVICE LEVEL FROM 'analytics_user';

File-Based Streaming

Tablet migration uses file-based streaming for dramatically faster topology changes:

What changed:

• During tablet migration, entire SSTable files are streamed directly to the target shard
• No deserialization/reserialization of data -- raw file transfer
• Significantly reduces CPU load and time for topology changes
• Only applies to tablet-based keyspaces

Performance impact:

• Node addition: 2-5x faster than vnode-based streaming (depending on data size)
• Repair streaming: Also benefits from file-based streaming for tablet-based tables

Unified Codebase

2025.1 merges Open Source and Enterprise into a single product:

What changed:

• All Enterprise features are now available in the unified codebase
• Features previously Enterprise-only (now available to all):
  • Workload prioritization (service levels)
  • Incremental Compaction Strategy (ICS)
  • Audit logging
  • LDAP authentication
  • Encryption at rest support
• Free tier: up to 10TB data and 50 vCPUs
• Commercial license required above free tier limits

Migration from Open Source 6.x:

ScyllaDB Open Source 6.x --> ScyllaDB 2025.1 (source-available)
- All OSS features preserved
- Additional Enterprise features available
- License change: AGPL --> Source-Available

Migration from Enterprise 2024.x:

ScyllaDB Enterprise 2024.x --> ScyllaDB 2025.1 (source-available)
- All Enterprise features preserved
- Unified versioning
- License change: Proprietary --> Source-Available

Upgrade Path to 2025.1

From ScyllaDB Open Source 6.2

# Step 1: Verify compatibility
nodetool version   # must be 6.2.x

# Step 2: Rolling upgrade (one node at a time)
# On each node:
nodetool drain
sudo systemctl stop scylla-server

# Install 2025.1 packages (replaces 6.x)
# (follow ScyllaDB docs for your package manager)

sudo systemctl start scylla-server
nodetool version   # verify 2025.1

# Step 3: After all nodes upgraded, enable Raft topology
curl -s -X POST http://localhost:10000/storage_service/raft_topology/upgrade

# Step 4: Verify
nodetool describecluster   # single schema version
nodetool status            # all nodes UN

From ScyllaDB Enterprise 2024.2

# Same rolling upgrade process
# Enterprise 2024.2 --> 2025.1 is a supported upgrade path
# All Enterprise features carry over

Configuration Changes in 2025.1

New/Changed Parameters

# scylla.yaml changes in 2025.1

# Tablets mode (new default)
tablets_mode_for_new_keyspaces: enabled   # was: disabled in 6.x

# Raft topology (new default for new clusters)
# For existing clusters, enable via REST API after upgrade

# Audit logging (formerly Enterprise-only)