Add OCI NoSQL Database: tables, rows, indexes

[arunesh-j]

Summary

• Implements OCI NoSQL Database Cloud Service against the existing portable database driver.
• Includes a real DDL parser, since OCI's CreateTable takes a SQL-ish statement rather than an attribute list.
• Nothing added to services/database/driver; OCI-only behaviour is a consumer-side Extras interface, per Move OCI-only capabilities out of shared driver packages #393.

Closes #412. Part of #376.

Changes

• providers/oci/nosql/ — Mock over memstore implementing driver.Database, guarded by a sync.RWMutex, plus the DDL parser and a query evaluator.
• server/oci/nosql/ — the /20190828/ surface.
• Wiring is one line each in providers/oci/oci.go and server/oci/oci.go.

Operations: tables (Create/List/Get/Update/Delete/ChangeCompartment), indexes (Create/List/Get/Delete), rows (GetRow/UpdateRow/DeleteRow), and Query. All 24 portable Database methods implemented.

DDL — what is parsed, what is refused

Parsed: CREATE TABLE and ALTER TABLE — scalar and JSON types, NOT NULL, DEFAULT, PRIMARY KEY with optional SHARD, USING TTL <n> DAYS, ADD/DROP on non-key columns, IF NOT EXISTS.

Refused by name, never silently accepted:

Refused	Why
Primary keys wider than two columns, composite shard keys	The portable partition/sort pair cannot identify such a row. This is a correctness rejection, not a shortcut
ARRAY / MAP / RECORD / ENUM	Not modelled
generated / MR_COUNTER / UUID modifiers	Not modelled
USING TTL … HOURS	OCI's own Schema reports TTL only in days
MODIFY, schema freezing, JSON-path index keys	Not modelled
Query projections, aggregates, joins, ORDER BY, range conditions	Not modelled

Judgement calls

• No change streams. OCI publishes no DynamoDB-Streams equivalent, so UpdateStreamConfig/GetStreamRecords return Unimplemented rather than an empty iterator. Noted in docs/services.md.
• Multi-delete is DELETE FROM … WHERE over /query — OCI's REST API has no MultiDelete operation; this is the real mechanism.
• ListIndexes requires compartmentId although real OCI marks it optional, so no list is ever unscoped. Documented.
• 501, named, for /tables/{id}/usage and /query/prepare, /query/summarize. usage is omitted from row and query responses for the same reason — zeros would read as real telemetry.

Provider Coverage

• AWS
• Azure
• GCP
• OCI

Checklist

• All tests pass (go test ./...) — exit 0, 272 packages
• Linter passes (golangci-lint run --timeout=9m) — 0 issues
• Every provider the change applies to implements the same behavior — OCI-only, additive
• Integration tests added to cloudemu_test.go — driver + handler tests instead
• Unit tests added to provider test files

Test Plan

go build ./...                                              clean
go test ./...                                               exit 0, 272 packages
go test -race ./providers/oci/... ./server/oci/...          11/11 ok
golangci-lint run --timeout=9m ./providers/oci/... ./server/oci/...   0 issues
go generate ./...                                           docs/coverage committed

Coverage leak check clean: no OCI operation in docs/coverage/{aws,azure,gcp}/*.md; git diff development -- services/ empty.

End-to-end on a running server (port 4613):

CreateTable                        -> 202 + Opc-Work-Request-Id
GetTable                           -> schema, primaryKey [id,email], shardKey [id], ttl 0
UpdateRow / GetRow                 -> 200 / value round-trips
ListTables                         -> 1 table; other compartment -> []; no compartmentId -> 400
CreateIndex                        -> 202 + work request; index ACTIVE
SELECT * FROM users                -> 3 rows
DELETE FROM users WHERE id = 2     -> NumRowsDeleted 2
DeleteRow twice                    -> 200, then isSuccess false
TRUNCATE TABLE                     -> 400 naming "TRUNCATE TABLE"
3-column PRIMARY KEY               -> 400 naming the limit
ARRAY(STRING)                      -> 400 naming "ARRAY"
USING TTL 6 HOURS                  -> 400 naming "HOURS"
ON_DEMAND + maxReadUnits           -> 400
undeclared column                  -> 400
SELECT name FROM users             -> 400 naming projections
GET /tables/users/usage            -> 501
DeleteTable                        -> 202; GetTable -> 404; ListTables -> []
USING TTL 7 DAYS                   -> schema.ttl 7; row carries timeOfExpiration

A note on parallel worktrees

An earlier run of this branch failed cmd/cloudemu TestServeOutOfProcess. It is not this change — cmd/cloudemu is untouched by the diff. Six Wave 2 worktrees were running their suites concurrently and that test contends on the shared ~/.cloudemu daemon lock. Verified: with nothing else running it passes, and the full suite is exit 0. Worth knowing as shared-state fragility whenever suites run in parallel.

[NitinKumar004]

Review notes

Real data-plane engine (per #427): N/A.

Findings

Medium · real-engine — Table names are globally unique, not per-compartment
providers/oci/nosql/nosql.go:417
If a user creates table "orders" in compartment A and then "orders" in compartment B -> the second CreateTable returns 409 AlreadyExists, because table names are a global key rather than compartment-scoped; real OCI scopes NoSQL table names per compartment and both creates succeed, so a multi-compartment SDK/CLI test that reuses a table name across compartments breaks against the emulator only.

m.tables is keyed by table name globally (nosql.go:175); newTable rejects with AlreadyExists on m.tables.Has(name) (nosql.go:417), and CreateOCITable checks m.tables.Get(d.Table) irrespective of spec.CompartmentID (table_extras.go:38-44). resolve()/lookup() also address tables by name globally.

Low · coverage — Portable Query/Scan operators and typed coercion below the 90% coverage pillar
providers/oci/nosql/rows.go:324
If a user issues a portable Query/Scan with a relational sort condition (>, BETWEEN) or stores a LONG/FLOAT/BOOLEAN-typed row -> the ordering (compareOrdering/compareStrings) and coercion (parseTyped/convertTyped) execute for the first time in production with no test guarding them, so a comparison- or coercion-logic regression would ship undetected.

go test -cover: provider 85.8%, server 83.4% (both < 90%). compareOrdering (rows.go:324) 0%, compareStrings (rows.go:341) 0%, compareOp (rows.go:306) 28.6%, parseTyped (row_extras.go:211) 35.7%, convertTyped (row_extras.go:242) 36.8%, SetMonitoring (nosql.go:194) 0%. Query tests exercise only SortOp "="; <, >, <=, >=, BETWEEN and CONTAINS/BEGINS_WITH ordering are never run.

Low · wire-fidelity — No oci-go-sdk SDK-compat test for the handler
server/oci/nosql/handler_test.go:94
If the real oci-go-sdk nosql client shapes a request/response field differently than the hand-modeled types.go structs (e.g. a body vs query-parameter for the Query limit, or a header the SDK requires) -> the divergence is not caught by the current tests, so a genuine client could fail at a step the httptest cases pass; adding one SDK create/get/query round-trip would close this.

handler_test.go drives the handler via raw httptest requests only; the only reference to github.com/oracle/oci-go-sdk is the package doc comment in handler.go. oci-conventions.md calls an SDK round-trip against httptest.NewServer "the strongest evidence the handler is right" (recommended, not required).