Phase D: AI-safe views — capability-surface projections over base data

Implements the llms3.com "AI-safe views" pattern: a named projection
that exposes only whitelisted columns, with optional row filter and
per-column redactions. AI agents (or Phase 13 roles) bind to the view;
they can never accidentally see PII even if they write raw SQL.

Schema (shared::types):
- AiView { name, base_dataset, columns: Vec<String>, row_filter,
           column_redactions: HashMap<String, Redaction>, ... }
- Redaction enum: Null | Hash | Mask { keep_prefix, keep_suffix }

Catalog (catalogd::registry):
- put_view validates base dataset exists + columns non-empty
- Persists JSON at _catalog/views/{name}.json (sanitized name)
- rebuild() loads views alongside dataset manifests on startup

Query layer (queryd::context):
- build_context registers every AiView as a DataFusion view object
- Constructed SELECT applies whitelist projection, WHERE filter, and
  redaction expressions per column
  - Mask: substr(prefix) + repeat('*', mid_len) + substr(suffix)
  - Hash: digest(value, 'sha256')
  - Null: CAST(NULL AS VARCHAR) AS col
- DataFusion handles JOINs/aggregates over the view natively — it's a
  real view, not a query rewrite

HTTP (catalogd::service):
- POST /catalog/views (create)
- GET  /catalog/views (list)
- GET  /catalog/views/{name} (full def)
- DELETE /catalog/views/{name}

End-to-end test on candidates (100K rows, 15 columns):

  candidates_safe view:
    columns: candidate_id, first_name, city, state, vertical,
             skills, years_experience, status
    row_filter: status != 'blocked'
    redaction: candidate_id mask(prefix=3, suffix=2)

  SELECT * FROM candidates_safe LIMIT 5
    -> 8 columns only, candidate_id shown as "CAN******01"
       (PII fields email/phone/last_name absent from result)

  SELECT email FROM candidates_safe
    -> fails (column not in projection)

  SELECT email FROM candidates
    -> succeeds (raw table still accessible by name —
       Phase 13 access control is the gate, not the view itself)

Survives restart — view definitions reload from object storage.

Limits / not in MVP:
- View CANNOT shadow base table by name (DataFusion treats them as
  separate identifiers; access control must restrict raw-table access)
- row_filter is treated as trusted SQL — operators must validate
  before persisting; only authenticated admin path should call put_view
- Redaction expressions assume column is castable to VARCHAR; numeric
  redactions could be misleading (a Hash on Int64 returns a hex string
  that won't equi-join with another hash on the same value type)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
root
2026-04-16 09:16:44 -05:00
co-authored by Claude Opus 4.6
parent 24f1249a62
commit 09fd446c8d
5 changed files with 279 additions and 3 deletions
+64
View File
@@ -201,6 +201,70 @@ impl QueryEngine {
}
}
// Phase D: register every AiView as a DataFusion view that wraps
// the base table with the safe projection + filter + redactions.
for view in self.registry.list_views().await {
// Build the SELECT clause with column whitelist and any
// redaction expressions. Each column is either the raw column
// name or an expression aliased back to the column name so
// downstream queries see the same shape.
let select_cols: Vec<String> = view.columns.iter().map(|col| {
match view.column_redactions.get(col) {
None => format!("\"{}\"", col),
Some(shared::types::Redaction::Null) => {
format!("CAST(NULL AS VARCHAR) AS \"{}\"", col)
}
Some(shared::types::Redaction::Hash) => {
// DataFusion has digest('sha256', value) — but column
// type might not be string. Cast first to be safe.
format!("digest(CAST(\"{}\" AS VARCHAR), 'sha256') AS \"{}\"", col, col)
}
Some(shared::types::Redaction::Mask { keep_prefix, keep_suffix }) => {
// Keep first N + last M chars, replace middle with stars.
// CASE NULL through with COALESCE for null-safe behavior.
format!(
"CASE WHEN \"{c}\" IS NULL THEN NULL ELSE \
concat(\
substr(CAST(\"{c}\" AS VARCHAR), 1, {p}), \
repeat('*', greatest(0, length(CAST(\"{c}\" AS VARCHAR)) - {p} - {s})), \
substr(CAST(\"{c}\" AS VARCHAR), greatest(1, length(CAST(\"{c}\" AS VARCHAR)) - {s} + 1))\
) END AS \"{c}\"",
c = col, p = keep_prefix, s = keep_suffix,
)
}
}
}).collect();
let where_clause = view.row_filter
.as_deref()
.map(|f| format!(" WHERE {}", f))
.unwrap_or_default();
let view_sql = format!(
"SELECT {} FROM \"{}\"{}",
select_cols.join(", "),
view.base_dataset,
where_clause,
);
tracing::debug!("registering AiView '{}': {}", view.name, view_sql);
let df = match ctx.sql(&view_sql).await {
Ok(df) => df,
Err(e) => {
tracing::warn!("view '{}' SQL invalid, skipping: {e}", view.name);
continue;
}
};
if let Err(e) = ctx.register_table(view.name.as_str(), df.into_view()) {
let msg = e.to_string();
if msg.contains("already exists") {
tracing::debug!("skip duplicate view registration: {}", view.name);
} else {
tracing::warn!("view '{}' registration failed: {e}", view.name);
}
}
}
Ok(ctx)
}
}