Imported from couchbaselabs/sequoia (tests/integration/7.2/AGENTS.md). Install upstream with npx skills add couchbaselabs/sequoia --skill 7.2. Copyright stays with the author.

Integration Test — 7.2 (Magma)

Baseline multi-service integration test for Couchbase 7.2 using Magma storage. Covers KV, GSI, FTS, N1QL, Analytics, Eventing, XDCR, Views, Sync Gateway, and Backup across a two-cluster topology. This is the earliest version in the versioned integration test series; it lacks shard affinity, bucket history retention, bucket ranking, and encryption at rest — all added in later releases.

What Changed Since Neo

Neo vs 7.2 relationship: The scope files are structurally near-identical (same 31/29 local node count with full service set, same 10+4 Magma bucket layout, same bucket names). The differences are in test configuration and cluster tuning — 7.2 formalises the versioned test series with a pinned CLI, fixes an OSO bug workaround, and adjusts cluster sizing. The milestone decomposition pattern from Neo is dropped in favour of a single complete test file.

Summary: 7.2 is a clean, versioned, production-ready codification of the Neo/Magma baseline. The main practical differences are the versioned CLI, OSO re-enablement, 3 FTS nodes, smaller remote cluster, and the removal of the milestone decomposition pattern. No new Couchbase features are introduced relative to Neo.

Files

Scope: scope_7.2_magma.yml

Clusters

Local Cluster — Service Distribution

Memory: 50% data RAM, 80% index RAM, 80% FTS RAM, 90% eventing/analytics RAM. No encryption at rest.

Local Buckets (10 total)

No history retention, no bucket rank, no encryption — all storage is Magma + fullEviction.

Remote Buckets (4 total)

Key Differences vs Later Versions

• No historyretentionbytes / historyretentionseconds on any bucket
• No bucket rank field
• No enable_encryption_at_rest or enableEncryptionAtRest
• No shard affinity (indexer.settings.enable_shard_affinity not set)
• CLI image: sequoiatools/couchbase-cli:7.2
• Smaller catapult loads (scale × 500 per bucket vs scale × 2000 in 8.x)
• Smaller pillowfight rate: scale × 500 docs, scale × 100 batch, 512 B docs

Test: test_7.2.yml

Phase 1 — Cluster Configuration (one-time)

Note: No shard affinity step — added in 7.6.

Phase 2 — Scope & Collection + XDCR Setup (one-time)

Scopes/collections created on local buckets 4–9 and remote buckets 4, 8, 9 (scale × 2 scopes, scale × 10 collections each, uniform). 300 s sleep for manifest sync. XDCR replications (compression only, no conflict logging) for default→remote0, bucket4→remote1, bucket8→remote2, bucket9→remote3.

Phase 3 — Data Loading & Background Workloads

Phase 4 — Service Setup

GSI indexes on buckets 4–7, build deferred (batch size 2), wait for completion. FTS indexes on buckets 4–7 (300 s sleep between each). Eventing (create_and_deploy). Analytics (analytics_setup). N1QL UDFs (10 per scope, buckets 4–7). Sync Gateway (async). N1QL query workloads (REQUEST_PLUS) on buckets 4–7. FTS query + flex query workloads.

Phase 5 — Topology Changes

Rebalance out → swap rebalance → eventing topology_change → analytics topology_change → view attacks → FTS child-field + nested-type indexes → XDCR advanced filtering → eventing pause → pillowfight_htp → 2i topology change → eventing resume → swap failover → XDCR pause/resume → swap hard failover → autofailover → FTS rebalance in/failover/addback/rebalance out → add 2 inactive nodes.

Phase 6 — Teardown & Validation

Stop background containers → sleep 1200 s → XDCR item count check (buckets 4, 8, 9) → GSI item count check (scale × 5 sample, buckets 4–7) → drop GSI indexes (buckets 4–9) → FTS item count check → drop FTS indexes (buckets 4–7) → drop UDFs → sleep.

Cross-Referenced Tests

Running

./sequoia -provider file:hosts.json -skip_setup \
  -scope tests/integration/7.2/scope_7.2_magma.yml \
  -test  tests/integration/7.2/test_7.2.yml