[huyplb]

What this adds

Redis in the SQL editor reads like the rest of the app.

• The schema explorer lists the keyspace. Before, it showed "No provider registered for dialect: redis" and listed nothing. Now there is a schema provider: a prefix (users for users:*) is a table, and a list, set, sorted set or stream key is a table of its own. Each comes with a type badge. The listing is from a bounded SCAN sample and says when it is partial.
• Every key type reads as rows.
  • A list, set, sorted set or stream key reads as its elements. Before, it read as just { id, type }.
  • A JSON document (the JSON type, or JSON text in a string) reads as its top-level fields.
  • JSON elements in a list or stream widen the table.
  • Reads are bounded: a page reads only what it shows, and a WHERE stops after 50,000 elements.
• Paging works. The editor wrapped every SELECT in SELECT * FROM (…) LIMIT …, which the Redis translator refused with "Unsupported clause". The page now goes onto the statement itself (pageSubsetSelect). The same fix applies to MongoDB.
• A passwordless Redis or MongoDB is usable. It is asked for a password only when the connection names a user. Every gate reads one rule, needsSessionPassword.
• Refusals. Writes to a collection or a JSON document are refused with a reason. Force-migrate refuses a target with no SQL DDL: for a Redis target it used to fall back to Db2 DDL. The explorer offers no table to create and no blueprint for Redis.

Simplify and review pass (2026-10-25)

Four reviewers looked at reuse, simplification, efficiency and altitude.

• Keyspace reader:
  • One table describes the four collection types.
  • Reads are pipelined: a 40-prefix sample went from about 4,000 sequential round trips to a few.
  • There is one TYPE per key, and UPDATE asks TYPE once.
• The subset parser reuses splitTopLevel / splitTop. isSubsetDialect is derived from the registry instead of a hand-kept list.
• The password rule is now one helper where there were 12 copies. A test that mirrored a stale copy now calls the real one.
• e2e:
  • helpers/redis.ts seeds through one redis-cli --pipe per database.
  • redis-browse is folded into redis-editor.
  • openDataPeek can wait for rows on a read-only peek.

Tests

• Unit: the keyspace reader against an in-memory node-redis double, plus the adapter, provider, subset parser, password rule and force-migrate guard. The new efficiency and guard tests fail on the old code and pass on the new.
• Live: redis.it.test.ts passes 19/19 against the dev container (node-redis 6.3.0).
• e2e: redis-editor.test.ts passes 21/21. It covers basic SQL, every built-in sample, Compare data with Data migrate, the schema explorer, and Data peek with the row editor.
• Gates: typecheck (web, e2e, cli), eslint . and the security lint are clean. Unit suite: 6517 pass. The 2 failures are the known timing flakes (code-cell-execute, migration-file linear time); neither file is touched here.

Not in this PR

• A Redis with only requirepass, saved without its password, is not prompted. The rule keys on the username, so that password stays saved on the connection. An explicit "No password" sign-in choice is on the board.
• The workflow SQL step's password warning ignores this rule. Its connection summary has no username. Filed separately.

🤖 Generated with Claude Code

---

Note

Medium Risk
Touches SQL execution, connection auth gating, and a large Redis adapter/keyspace surface; behavior changes for Mongo paging and password prompts, but guarded by extensive unit/integration/e2e tests.

Overview
Redis (and MongoDB subset paging) in the SQL editor — keyspaces appear in the schema explorer with store-type badges; lists, sets, sorted sets, streams, and JSON documents render as queryable rows, not stub { id, type } rows. A new Redis schema provider and redis.keyspace layer power browsing and SQL execution; hash-only writes stay enforced with explicit errors for collections and JSON.

Paging and errors — subset dialects no longer use a nested SELECT * FROM (…) AS fox_page; the server appends LIMIT/OFFSET via pageSubsetSelect, and Redis/Mongo statement failures surface without a misleading “Paging failed” wrapper. MongoDB OFFSET paging and empty limit(0) pages are fixed similarly.

Auth UX — needsSessionPassword centralizes when the UI must prompt (passwordless Redis/Mongo when no username; Redis with ACL user still needs a secret). Compare retest gates and editor destination selection use this helper.

Explorer / DDL guardrails — New table / blueprint actions hide for engines without SQL DDL (Redis, MongoDB). Subset explorers insert bare table names and column refs without aliases.

History & migrate — Schema capture, force-migrate, and related routes refuse Redis/Mongo before loading a sampled keyspace as a version or emitting wrong DDL.

E2E — redis-editor.test.ts (docker redis-cli helper) exercises SQL, samples, compare/migrate, explorer, and data peek; test IDs on grid/run/result errors support assertions.

Parser hardening — SQL subset gains quoted dotted table names, parenthesized WHERE (Data peek), stricter predicate/column rules, and semicolon-safe autoAliasSelectColumns.

^{Reviewed by Cursor Bugbot for commit 90416f7. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: b675247b-b927-43be-b009-7a1384114700)

[github-actions]

Test IDs

Added (4)

• sql-explorer-store-type-{…}
• sql-grid-error
• sql-result-error
• sql-run-error

{…} is a part filled in at run time. The full tree: docs/testing/TEST_IDS.md.