Overview

One Database Manager, cluster-wide. The 🗄️ Databases page at the top of the sidebar is the single place in WolfStack to reach every configured database — MariaDB, MySQL, or PostgreSQL — regardless of which node in the cluster actually owns it. The old per-node MySQL editor has been retired: there is now one manager at the datacenter level that serves every connection.

Connections are stored as profiles: a label, engine type, host/port/user/password, and the cluster+node that can reach it. Because profiles replicate to every WolfStack peer, any node can proxy queries — if the main node goes down, the others pick up seamlessly.

Supported Databases

Database	Driver	Notes
MariaDB	mysql_async (pure Rust)	All MariaDB versions, TCP + SSL/TLS
MySQL	mysql_async (pure Rust)	MySQL 5.7+ and 8.0+
Percona Server	mysql_async (pure Rust)	Compatible with MySQL protocol
PostgreSQL	tokio-postgres + deadpool	Pooled connections, SSL prefer/require/disable

Cluster-Aware Routing

When you add a connection you pick the target cluster, then the owning node, then the IP reachable from that node — the dropdown includes host interfaces, Docker containers, LXC containers, and VMs with their names so you can find them at a glance. You can also paste a shorthand like WolfStack-Proxmox:wolfstack-1:10.10.10.105.

Queries execute on the node that owns the connection. If you open a connection from a different node, WolfStack proxies the query to the owning peer over the cluster-secret-authenticated inter-node channel — no exposing database ports on the public network, no copying credentials between nodes. Profile changes replicate automatically.

Database Manager — Features

The Manager takes the whole page when a connection is open. Six tabs across the top:

📊 Data

• Browse rows with pagination (50 / 100 / 250 / 500 / 1000 rows per page)
• Click column headers to sort ascending, click again for descending
• Quick text filter across all visible cells
• Inline row editing: click ✏️ for a full row-edit modal with one input per column, NULL toggles, PK fields protected, only-changed-columns UPDATE
• Edit / Delete buttons pinned to the left — always visible, even on wide rows
• ➕ Row — insert a new row via prompts for each column
• 📥 Import CSV — upload a CSV, header row maps to columns, rows inserted in a single batch
• 📤 Export — CSV / JSON / SQL INSERTs of the current page

🏗️ Structure

• Full DDL editing with Schema permission: Add column / Drop column, Add index / Drop index, Add foreign key / Drop foreign key
• Engine-aware syntax: DROP CONSTRAINT on Postgres, DROP FOREIGN KEY on MySQL/MariaDB

🗺️ Diagram (ERD)

• SVG schema diagram — tables rendered as draggable boxes with PK 🔑 markers
• Foreign-key relationships drawn as bezier arrows
• Mousewheel zoom, drag-empty-space pan, double-click a table to jump to its Structure tab
• Auto-layout places tables in a grid (heights variable so wide tables don't overlap)
• 💾 Save SVG — exports a standalone .svg file for docs and hand-off

🧱 Build (Visual Query Builder)

• Drag tables from the picker onto the canvas
• Tick columns to include in the SELECT
• Click-drag column-to-column to draw a JOIN (INNER by default)
• Add WHERE conditions in the bottom panel (table/column/operator/value, one row per condition)
• Configurable row limit
• Generated SQL updates live at the bottom — hit Send to Query to open it in the Query tab

⚡ Query

• Multi-tab editor (+ new tab, ✕ close, double-click to rename)
• Lightweight SQL syntax highlighting (keywords / strings / numbers / comments / operators)
• Live sqlparser-backed validation with tier badge (Read / Update / Delete / Schema)
• Ctrl+Space autocomplete with context: after FROM/JOIN suggests tables, after . suggests columns, otherwise keywords + tables. Each candidate shows its source ("table in schema", "column in schema.table", "SQL keyword")
• 💾 Saved queries — named, per-user, server-side (follow you across devices)
• Recent — auto ring-buffer of the last 20 executed queries per connection, per user
• 🔍 Explain / ✨ Format / 📂 Run SQL file (splits on ;, handles DELIMITER, runs via /query-multi)
• Per-tab schema context (Navicat-style): new tabs inherit the currently-selected schema, existing tabs keep theirs
• Destructive-query confirm on TRUNCATE, DROP TABLE, unqualified DELETE, unqualified UPDATE
• Unlimited-query warning when a SELECT/WITH has no LIMIT
• Draggable splitter between editor and result panel
• ⛶ Fullscreen toggle for the whole Databases page

🖥️ Server

• Server version, database, current user, host, port
• Key variables — click ✏️ to change (modal with session / global / ALTER SYSTEM scope, requires Schema permission, auto-runs pg_reload_conf() after Postgres ALTER SYSTEM)
• Active sessions (process list) — click ✕ to kill via KILL <id> (MySQL) or pg_terminate_backend(pid) (Postgres)

Schema tree — left panel

• Tables + Views grouped per schema with counts
• Procedures / Functions / Triggers listed read-only — click to view the CREATE definition in a modal
• Per-schema + New table, Dump schema, Dump + data actions
• ◀ / 📁 ▶ toggle to hide the tree and give the editor full width

Create Table & Dump / Restore

Create Table wizard — form with a column grid (name, type, NOT NULL, PK, default), live SQL preview, runs at Schema permission.

Dump to SQL — client-driven export of a whole schema. Choose schema-only (fast) or schema + data (paginated 500-row batches). Writes a standalone .sql file with DROP TABLE IF EXISTS / CREATE TABLE / INSERT statements ready for restore. Postgres uses information_schema reconstruction since it has no SHOW CREATE TABLE.

Restore from SQL — the Query tab's 📂 Run SQL file button accepts any .sql upload. The client splits on ; honouring comments, string literals, and DELIMITER directives; the new /query-multi endpoint executes each statement sequentially with stop-on-error and reports per-statement outcomes including the line number where any failure occurred.

Schema Compare

The 🧭 Compare schemas button on the Databases page header opens a side-by-side diff of two schemas (same connection or different). Reports:

• Tables only in source / only in target / in both-with-differences
• Per-table: column additions, removals, type / nullability / default changes
• Index and foreign-key additions / removals

Click 📜 Generate migration SQL to produce CREATE TABLE / DROP TABLE / ALTER TABLE statements that would bring the target in line with the source. The generator leaves complex type-changes as TODO comments for the operator to review — always treat the output as a starting point.

Enterprise ACL

Each connection profile can be restricted to a list of named WolfStack users. Operators without access don't see the profile at all — not in the Databases page, not in the Settings editor, not via the API. A 🔒 badge next to the connection in the list indicates ACL is active.

ACL is an Enterprise feature; on Community installs all authenticated operators share every profile. The restriction applies uniformly to the UI, the REST API, and the AI agent tool surface — there is no bypass path.

AI Agents & WolfFlow

AI agents can talk to the database. Agents configured with SQL permissions use the same pipeline as the UI — same profiles, same cluster routing, same audit log, same sqlparser classifier, same enterprise ACL. An agent asked "how many customers signed up yesterday?" can issue SELECT COUNT(*) FROM customers WHERE created_at > NOW() - INTERVAL 1 DAY and feed the result back into its reply.

Per-agent settings gate what the agent can touch via three independent checks:

1. Flags — each agent has separate sql_read, sql_update, sql_delete switches (all default off). DDL via sql_schema is never granted to agents — AI writing ALTER TABLE unsupervised is not a surface we want open.
2. Allowlist — an operator-set list of connection IDs the agent is allowed to query (empty list = none).
3. Parser — every query is classified by sqlparser before dispatch. An UPDATE pretending to be a SELECT is rejected before it hits the database. Multi-statement queries are refused outright.

WolfFlow workflows get a matching SQL Query action node with the same gates — operators pick a connection, pick a permission tier, write the query; the workflow runs it on schedule or trigger and can feed the result into subsequent steps. Every execution — agent, workflow, or UI — is appended to /var/log/wolfstack/sql-audit.log with caller tag, connection ID, query, outcome, row count, and elapsed time.

Security Model

• Encrypted at rest — passwords stored AES-256-GCM encrypted with the cluster secret (same scheme as OIDC client secrets); only decrypted in memory at connect time
• Cluster-secret auth on inter-node calls — session cookies can't reach the /query-proxy, /receive, or /audit endpoints; only trusted peers
• Hardcoded API denylist — agents using the generic wolfstack_api tool cannot POST to /api/sql-connections/* or the node-IP enumerator; the only way to run SQL is through the dedicated sql_query tool, which respects all three permission gates
• Filesystem denylist — /etc/wolfstack/sql-connections.json, /etc/wolfstack/sql-saved-queries.json, and /var/log/wolfstack/sql-audit.log are off-limits to agent filesystem tools
• Per-user saved queries — operators' pinned SQL is keyed to their username, not shared across the cluster
• Bounded execution — 5s connect timeout, 30s default query timeout (configurable per query), 10,000-row / 10 MB result cap; hung databases can't starve the actix workers