postgresai 0.16.0-dev.0 → 0.16.0-dev.10

package/dist/sql/06.helpers.sql

@@ -2,126 +2,6 @@
 -- These functions use SECURITY DEFINER to allow the monitoring user to perform
 -- operations they don't have direct permissions for.
 
-/*
- * explain_generic
- *
- * Function to get generic explain plans with optional HypoPG index testing.
- * Requires: PostgreSQL 16+ (for generic_plan option), HypoPG extension (optional).
- *
- * Security notes:
- * - EXPLAIN without ANALYZE is read-only (plans but doesn't execute the query)
- * - PostgreSQL's EXPLAIN only accepts a single statement (primary protection)
- * - Input validation uses a simple heuristic to detect multiple statements
- *   (Note: may reject valid queries containing semicolons in string literals)
- *
- * Usage examples:
- *   -- Basic generic plan
- *   select postgres_ai.explain_generic('select * from users where id = $1');
- *
- *   -- JSON format
- *   select postgres_ai.explain_generic('select * from users where id = $1', 'json');
- *
- *   -- Test a hypothetical index
- *   select postgres_ai.explain_generic(
- *     'select * from users where email = $1',
- *     'text',
- *     'create index on users (email)'
- *   );
- */
-create or replace function postgres_ai.explain_generic(
-  in query text,
-  in format text default 'text',
-  in hypopg_index text default null,
-  out result text
-)
-language plpgsql
-security definer
-set search_path = pg_catalog, public
-as $$
-declare
-  v_line record;
-  v_lines text[] := '{}';
-  v_explain_query text;
-  v_hypo_result record;
-  v_version int;
-  v_hypopg_available boolean;
-  v_clean_query text;
-begin
-  -- Check PostgreSQL version (generic_plan requires 16+)
-  select current_setting('server_version_num')::int into v_version;
-
-  if v_version < 160000 then
-    raise exception 'generic_plan requires PostgreSQL 16+, current version: %',
-      current_setting('server_version');
-  end if;
-
-  -- Input validation: reject empty queries
-  if query is null or trim(query) = '' then
-    raise exception 'query cannot be empty';
-  end if;
-
-  -- Input validation: detect multiple statements (defense-in-depth)
-  -- Note: This is a simple heuristic - EXPLAIN itself only accepts single statements
-  -- Limitation: Queries with semicolons inside string literals will be rejected
-  v_clean_query := trim(query);
-  if v_clean_query like '%;%' then
-    -- Strip trailing semicolon if present (common user convenience)
-    v_clean_query := regexp_replace(v_clean_query, ';\s*$', '');
-    -- If there's still a semicolon, reject (likely multiple statements or semicolon in string)
-    if v_clean_query like '%;%' then
-      raise exception 'query contains semicolon (multiple statements not allowed; note: semicolons in string literals are also not supported)';
-    end if;
-  end if;
-
-  -- Check if HypoPG extension is available
-  if hypopg_index is not null then
-    select exists(
-      select 1 from pg_extension where extname = 'hypopg'
-    ) into v_hypopg_available;
-
-    if not v_hypopg_available then
-      raise exception 'HypoPG extension is required for hypothetical index testing but is not installed';
-    end if;
-
-    -- Create hypothetical index
-    select * into v_hypo_result from hypopg_create_index(hypopg_index);
-    raise notice 'Created hypothetical index: % (oid: %)',
-      v_hypo_result.indexname, v_hypo_result.indexrelid;
-  end if;
-
-  -- Build and execute EXPLAIN query
-  -- Note: EXPLAIN is read-only (plans but doesn't execute), making this safe
-  begin
-    if lower(format) = 'json' then
-      execute 'explain (verbose, settings, generic_plan, format json) ' || v_clean_query
-        into result;
-    else
-      for v_line in execute 'explain (verbose, settings, generic_plan) ' || v_clean_query loop
-        v_lines := array_append(v_lines, v_line."QUERY PLAN");
-      end loop;
-      result := array_to_string(v_lines, e'\n');
-    end if;
-  exception when others then
-    -- Clean up hypothetical index before re-raising
-    if hypopg_index is not null then
-      perform hypopg_reset();
-    end if;
-    raise;
-  end;
-
-  -- Clean up hypothetical index
-  if hypopg_index is not null then
-    perform hypopg_reset();
-  end if;
-end;
-$$;
-
-comment on function postgres_ai.explain_generic(text, text, text) is
-  'Returns generic EXPLAIN plan with optional HypoPG index testing (requires PG16+)';
-
--- Grant execute to the monitoring user
-grant execute on function postgres_ai.explain_generic(text, text, text) to {{ROLE_IDENT}};
-
 /*
  * table_describe
  *
@@ -435,5 +315,3 @@ comment on function postgres_ai.table_describe(text) is
   'Returns comprehensive table information in compact text format for LLM analysis';
 
 grant execute on function postgres_ai.table_describe(text) to {{ROLE_IDENT}};
-
-

package/dist/sql/sql/06.helpers.sql

@@ -2,126 +2,6 @@
 -- These functions use SECURITY DEFINER to allow the monitoring user to perform
 -- operations they don't have direct permissions for.
 
-/*
- * explain_generic
- *
- * Function to get generic explain plans with optional HypoPG index testing.
- * Requires: PostgreSQL 16+ (for generic_plan option), HypoPG extension (optional).
- *
- * Security notes:
- * - EXPLAIN without ANALYZE is read-only (plans but doesn't execute the query)
- * - PostgreSQL's EXPLAIN only accepts a single statement (primary protection)
- * - Input validation uses a simple heuristic to detect multiple statements
- *   (Note: may reject valid queries containing semicolons in string literals)
- *
- * Usage examples:
- *   -- Basic generic plan
- *   select postgres_ai.explain_generic('select * from users where id = $1');
- *
- *   -- JSON format
- *   select postgres_ai.explain_generic('select * from users where id = $1', 'json');
- *
- *   -- Test a hypothetical index
- *   select postgres_ai.explain_generic(
- *     'select * from users where email = $1',
- *     'text',
- *     'create index on users (email)'
- *   );
- */
-create or replace function postgres_ai.explain_generic(
-  in query text,
-  in format text default 'text',
-  in hypopg_index text default null,
-  out result text
-)
-language plpgsql
-security definer
-set search_path = pg_catalog, public
-as $$
-declare
-  v_line record;
-  v_lines text[] := '{}';
-  v_explain_query text;
-  v_hypo_result record;
-  v_version int;
-  v_hypopg_available boolean;
-  v_clean_query text;
-begin
-  -- Check PostgreSQL version (generic_plan requires 16+)
-  select current_setting('server_version_num')::int into v_version;
-
-  if v_version < 160000 then
-    raise exception 'generic_plan requires PostgreSQL 16+, current version: %',
-      current_setting('server_version');
-  end if;
-
-  -- Input validation: reject empty queries
-  if query is null or trim(query) = '' then
-    raise exception 'query cannot be empty';
-  end if;
-
-  -- Input validation: detect multiple statements (defense-in-depth)
-  -- Note: This is a simple heuristic - EXPLAIN itself only accepts single statements
-  -- Limitation: Queries with semicolons inside string literals will be rejected
-  v_clean_query := trim(query);
-  if v_clean_query like '%;%' then
-    -- Strip trailing semicolon if present (common user convenience)
-    v_clean_query := regexp_replace(v_clean_query, ';\s*$', '');
-    -- If there's still a semicolon, reject (likely multiple statements or semicolon in string)
-    if v_clean_query like '%;%' then
-      raise exception 'query contains semicolon (multiple statements not allowed; note: semicolons in string literals are also not supported)';
-    end if;
-  end if;
-
-  -- Check if HypoPG extension is available
-  if hypopg_index is not null then
-    select exists(
-      select 1 from pg_extension where extname = 'hypopg'
-    ) into v_hypopg_available;
-
-    if not v_hypopg_available then
-      raise exception 'HypoPG extension is required for hypothetical index testing but is not installed';
-    end if;
-
-    -- Create hypothetical index
-    select * into v_hypo_result from hypopg_create_index(hypopg_index);
-    raise notice 'Created hypothetical index: % (oid: %)',
-      v_hypo_result.indexname, v_hypo_result.indexrelid;
-  end if;
-
-  -- Build and execute EXPLAIN query
-  -- Note: EXPLAIN is read-only (plans but doesn't execute), making this safe
-  begin
-    if lower(format) = 'json' then
-      execute 'explain (verbose, settings, generic_plan, format json) ' || v_clean_query
-        into result;
-    else
-      for v_line in execute 'explain (verbose, settings, generic_plan) ' || v_clean_query loop
-        v_lines := array_append(v_lines, v_line."QUERY PLAN");
-      end loop;
-      result := array_to_string(v_lines, e'\n');
-    end if;
-  exception when others then
-    -- Clean up hypothetical index before re-raising
-    if hypopg_index is not null then
-      perform hypopg_reset();
-    end if;
-    raise;
-  end;
-
-  -- Clean up hypothetical index
-  if hypopg_index is not null then
-    perform hypopg_reset();
-  end if;
-end;
-$$;
-
-comment on function postgres_ai.explain_generic(text, text, text) is
-  'Returns generic EXPLAIN plan with optional HypoPG index testing (requires PG16+)';
-
--- Grant execute to the monitoring user
-grant execute on function postgres_ai.explain_generic(text, text, text) to {{ROLE_IDENT}};
-
 /*
  * table_describe
  *
@@ -435,5 +315,3 @@ comment on function postgres_ai.table_describe(text) is
   'Returns comprehensive table information in compact text format for LLM analysis';
 
 grant execute on function postgres_ai.table_describe(text) to {{ROLE_IDENT}};
-
-

package/lib/aas-onboard.ts

@@ -28,7 +28,11 @@ function grafanaBaseUrl(): string {
 }
 
 function grafanaAdminUser(): string {
-  return process.env.GF_SECURITY_ADMIN_USER || "admin";
+  // The monitoring stack's compose hardcodes the Grafana admin user to
+  // "monitor" (GF_SECURITY_ADMIN_USER: monitor), so default to that rather than
+  // Grafana's stock "admin" — otherwise AAS arming logs in as the wrong user
+  // and every datasource lookup 401s. An explicit env override still wins.
+  return process.env.GF_SECURITY_ADMIN_USER || "monitor";
 }
 
 /** Parse a vcpus input (flag/env) to a non-negative integer; 0 = "unknown" fallback. */
@@ -130,17 +134,26 @@ export async function mintAasServiceAccountToken(
  * Resolve the single Prometheus-typed datasource's numeric id on the local
  * Grafana. The monitoring stack's VictoriaMetrics datasource is type
  * "prometheus" (VM speaks PromQL), and the stack registers exactly one such
- * datasource — the same one the collector queries. >1 or 0 → null (skip),
- * matching v1.aas_onboard's discovery contract.
+ * datasource — the same one the collector queries. 0 / API-not-ready → null
+ * (a provisioning transient — the readiness loop retries); >1 → "ambiguous"
+ * (a permanent misconfiguration — the loop stops at once), matching
+ * v1.aas_onboard's >1 skip.
  */
-export async function resolveDatasourceId(adminPassword: string, debug = false): Promise<number | null> {
+export async function resolveDatasourceId(adminPassword: string, debug = false): Promise<number | "ambiguous" | null> {
   try {
     const res = await grafanaApi("GET", "/api/datasources", adminPassword);
     if (!res.ok) return null;
     const list = (await res.json().catch(() => [])) as Array<{ id?: unknown; type?: unknown }>;
     const prom = list.filter((d) => d.type === "prometheus");
-    if (prom.length !== 1) {
-      if (debug) console.error(`Debug: AAS: expected 1 prometheus datasource, found ${prom.length}`);
+    if (prom.length > 1) {
+      // >1 is a permanent misconfiguration, not a provisioning transient: the
+      // datasource count only grows as Grafana provisions, so retrying can never
+      // resolve it. Signal a definitive skip so the readiness loop bails at once.
+      if (debug) console.error(`Debug: AAS: ${prom.length} prometheus datasources (ambiguous); not retrying`);
+      return "ambiguous";
+    }
+    if (prom.length === 0) {
+      if (debug) console.error(`Debug: AAS: no prometheus datasource resolvable yet`);
       return null;
     }
     return typeof prom[0].id === "number" ? prom[0].id : null;
@@ -169,6 +182,10 @@ export async function registerAasCollection(
     apiBaseUrl?: string;
     debug?: boolean;
     fetchImpl?: typeof fetch;
+    // Grafana-readiness polling for the datasource lookup (Grafana has just
+    // been started by `compose up`). Defaults: 20 attempts × 3s.
+    datasourceMaxAttempts?: number;
+    datasourceRetryDelayMs?: number;
   }
 ): Promise<AasRegisterResult> {
   const debug = !!opts.debug;
@@ -178,7 +195,24 @@ export async function registerAasCollection(
     const labels = resolveAasLabels(opts.instancesPath);
     if (!labels) return { ok: false, reason: "could not determine a single (cluster, node_name) target" };
 
-    const datasourceId = await resolveDatasourceId(opts.grafanaPassword, debug);
+    // Grafana was just started by `compose up`; it needs time to create its
+    // admin user, provision datasources, and serve its API. Querying too early
+    // makes the datasource lookup fail transiently, so poll until it resolves
+    // (best-effort, capped — the install never blocks on this).
+    const maxAttempts = opts.datasourceMaxAttempts ?? 20;
+    const retryDelayMs = opts.datasourceRetryDelayMs ?? 3000;
+    let datasourceId: number | null = null;
+    for (let attempt = 1; attempt <= maxAttempts; attempt++) {
+      const resolved = await resolveDatasourceId(opts.grafanaPassword, debug);
+      if (typeof resolved === "number") { datasourceId = resolved; break; }
+      // "ambiguous" (>1 prometheus datasource) is permanent — retrying can't fix
+      // it, so stop polling immediately instead of waiting out the whole budget.
+      if (resolved === "ambiguous") break;
+      if (attempt < maxAttempts) {
+        if (debug) console.error(`Debug: AAS: datasource not resolvable yet (attempt ${attempt}/${maxAttempts}); waiting for Grafana…`);
+        await new Promise((resolve) => setTimeout(resolve, retryDelayMs));
+      }
+    }
     if (datasourceId == null) return { ok: false, reason: "could not resolve the Prometheus datasource id" };
 
     const saToken = await mintAasServiceAccountToken(opts.grafanaPassword, debug);