turbine-orm 0.55.0 → 0.56.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/cli/index.js CHANGED
@@ -14,7 +14,7 @@
14
14
  * turbine migrate status , Show migration status
15
15
  * turbine seed , Run seed file
16
16
  * turbine status , Show schema summary
17
- * turbine doctor - Cost-aware missing-FK-index triage (--fix, --json, --no-concurrently, --unused, --audit)
17
+ * turbine doctor - Index + cached-plan triage (--fix, --json, --no-concurrently, --unused, --audit, --no-plan-divergence)
18
18
  * turbine studio : Launch local read-only web UI (--demo for a seeded sample DB)
19
19
  * turbine mcp , Start read-only MCP server over JSON-RPC stdio
20
20
  * turbine observe , Launch metrics dashboard (requires TURBINE_OBSERVE_URL)
@@ -32,6 +32,7 @@ import { generate, generatePrismaMap } from '../generate.js';
32
32
  import { buildCreateIndexSql, buildDropIndexSql, collectDoctorProbeIndexNames, collectRelationProbeColumns, findMissingRelationIndexes, } from '../index-advisor.js';
33
33
  import { auditDoctorIndexes, collectStatsSnapshot, collectTableHeat, findInvalidIndexes, findRedundantIndexes, findUnusedIndexes, formatBytes, isSnapshotUsable, STATS_THRESHOLDS, scoreMissingIndex, } from '../index-stats.js';
34
34
  import { introspect } from '../introspect.js';
35
+ import { collectDivergenceCandidateColumns, findPlanDivergence, PLAN_DIVERGENCE_THRESHOLDS, } from '../plan-divergence.js';
35
36
  import { DestructivePushRefusal, schemaDiff, schemaPush } from '../schema-sql.js';
36
37
  import { configTemplate, DEFAULT_INIT_SEED_FILE, findConfigFile, loadConfigResult, looksLikeSchemaFilePath, resolveConfig, resolveSeedFile, unwrapModuleDefault, } from './config.js';
37
38
  import { DESTRUCTIVE_KIND_LABEL } from './destructive.js';
@@ -142,6 +143,9 @@ export function parseArgs(argv = process.argv.slice(2)) {
142
143
  result.metricsUrl = next;
143
144
  i++;
144
145
  break;
146
+ case '--no-plan-divergence':
147
+ result.noPlanDivergence = true;
148
+ break;
145
149
  case '--zod':
146
150
  result.zod = true;
147
151
  break;
@@ -2259,20 +2263,27 @@ async function cmdDoctor(args, config) {
2259
2263
  const missing = findMissingRelationIndexes(schema);
2260
2264
  // Collect live statistics. The collector reads whole-schema indexes (for
2261
2265
  // invalid-index detection) plus per-table stats + probed-column null_frac.
2262
- const probedTables = [...new Set(missing.map((m) => m.table))];
2263
2266
  const probedColumns = [];
2264
2267
  for (const m of missing) {
2265
2268
  if (m.columns.length === 1 && m.columns[0] !== undefined) {
2266
2269
  probedColumns.push({ table: m.table, column: m.columns[0] });
2267
2270
  }
2268
2271
  }
2272
+ // Plan-divergence candidates are the columns that ARE indexed, so their tables
2273
+ // are usually disjoint from the missing-index set: both lists feed the same
2274
+ // one-connection snapshot rather than opening a second read.
2275
+ const divergenceOn = args.noPlanDivergence !== true;
2276
+ const divergenceColumns = divergenceOn ? collectDivergenceCandidateColumns(schema) : [];
2277
+ const probedTables = [...new Set(missing.map((m) => m.table))];
2278
+ const statsTables = [...new Set([...probedTables, ...divergenceColumns.map((c) => c.table)])];
2269
2279
  let snapshot;
2270
2280
  try {
2271
2281
  snapshot = await collectStatsSnapshot({
2272
2282
  connectionString: url,
2273
2283
  schema: config.schema,
2274
- tables: probedTables,
2284
+ tables: statsTables,
2275
2285
  columns: probedColumns,
2286
+ distributionColumns: divergenceColumns,
2276
2287
  });
2277
2288
  }
2278
2289
  catch (err) {
@@ -2311,9 +2322,16 @@ async function cmdDoctor(args, config) {
2311
2322
  ? auditDoctorIndexes(snapshot, collectDoctorProbeIndexNames(schema), { minScans, relationProbes })
2312
2323
  : [];
2313
2324
  const subtract = { unusedRan, auditRan, minScans, unused, redundant, audit };
2325
+ // Plan divergence has its OWN freshness gate. The cost tiers require a
2326
+ // trustworthy stats_reset age because they normalize write counters by it;
2327
+ // this check reads no counter, only pg_stats, whose freshness is ANALYZE. A
2328
+ // cluster with a NULL stats_reset (the default) must still get the check.
2329
+ const divergence = divergenceOn && snapshot.available
2330
+ ? findPlanDivergence(schema, snapshot)
2331
+ : { findings: [], notices: [], candidatesConsidered: 0 };
2314
2332
  if (jsonMode) {
2315
2333
  spinner?.stop();
2316
- console.log(JSON.stringify(buildDoctorJson({ schema, findings, invalid, snapshot, usable, heat, subtract, args }), null, 2));
2334
+ console.log(JSON.stringify(buildDoctorJson({ schema, findings, invalid, snapshot, usable, heat, subtract, divergence, args }), null, 2));
2317
2335
  return;
2318
2336
  }
2319
2337
  await renderDoctorHuman({
@@ -2325,6 +2343,7 @@ async function cmdDoctor(args, config) {
2325
2343
  usable,
2326
2344
  heat,
2327
2345
  subtract,
2346
+ divergence,
2328
2347
  args,
2329
2348
  config,
2330
2349
  });
@@ -2392,13 +2411,17 @@ function buildDoctorJson(ctx) {
2392
2411
  out.redundant = ctx.subtract.unusedRan ? ctx.subtract.redundant : [];
2393
2412
  out.audit = ctx.subtract.auditRan ? ctx.subtract.audit : [];
2394
2413
  out.invalid = ctx.invalid;
2414
+ // Always an array, never absent: a consumer must not have to write `?? []`
2415
+ // just because the section was skipped or found nothing.
2416
+ out.planDivergence = ctx.divergence.findings;
2417
+ out.planDivergenceNotices = ctx.divergence.notices;
2395
2418
  return out;
2396
2419
  }
2397
2420
  async function renderDoctorHuman(ctx) {
2398
- const { spinner, schema, findings, invalid, snapshot, usable, heat, subtract, args, config } = ctx;
2421
+ const { spinner, schema, findings, invalid, snapshot, usable, heat, subtract, divergence, args, config } = ctx;
2399
2422
  spinner.succeed(`Scanned ${bold(String(Object.keys(schema.tables).length))} tables`);
2400
2423
  const subtractRan = subtract.unusedRan || subtract.auditRan;
2401
- const nothingToAdd = findings.length === 0 && invalid.length === 0;
2424
+ const nothingToAdd = findings.length === 0 && invalid.length === 0 && divergence.findings.length === 0;
2402
2425
  const nothingToSubtract = subtract.unused.length === 0 && subtract.redundant.length === 0 && subtract.audit.length === 0;
2403
2426
  if (nothingToAdd && (!subtractRan || nothingToSubtract)) {
2404
2427
  if (subtractRan) {
@@ -2429,6 +2452,7 @@ async function renderDoctorHuman(ctx) {
2429
2452
  }
2430
2453
  }
2431
2454
  renderInvalidIndexes(invalid);
2455
+ renderPlanDivergence(divergence);
2432
2456
  if (subtract.unusedRan) {
2433
2457
  renderUnusedIndexes(subtract.unused, subtract.minScans, snapshot);
2434
2458
  renderRedundantIndexes(subtract.redundant);
@@ -2594,6 +2618,92 @@ function renderInvalidIndexes(invalid) {
2594
2618
  newline();
2595
2619
  }
2596
2620
  }
2621
+ /** Round to a whole number and group it, for the divergence report's estimates. */
2622
+ function divInt(n) {
2623
+ if (!Number.isFinite(n))
2624
+ return 'unbounded';
2625
+ return Math.round(n).toLocaleString('en-US');
2626
+ }
2627
+ /**
2628
+ * The plan-divergence section: columns whose value distribution can flip a
2629
+ * cached plan. Finding-only by design, there is no `--fix` for it: the fix is
2630
+ * application code (scope the plan-cache mode to the affected reads), and the
2631
+ * index that looks like a fix is measured NOT to be one.
2632
+ */
2633
+ function renderPlanDivergence(divergence) {
2634
+ const { findings, notices } = divergence;
2635
+ if (findings.length === 0 && notices.length === 0)
2636
+ return;
2637
+ if (findings.length > 0) {
2638
+ warn(`${bold(String(findings.length))} column(s) whose value distribution can flip a cached plan`);
2639
+ newline();
2640
+ console.log(` ${dim('Postgres may promote a named prepared statement to a GENERIC plan from its sixth execution,')}`);
2641
+ console.log(` ${dim('but only when the generic plan is not ESTIMATED to cost more than the average custom')}`);
2642
+ console.log(` ${dim('plan. A generic plan cannot see your values: it estimates "col = $1" as rows /')}`);
2643
+ console.log(` ${dim('n_distinct and an unknown LIMIT as 10% of the child estimate. When those defaults')}`);
2644
+ console.log(` ${dim('land on the other side of a plan boundary from the real value, the plan flips.')}`);
2645
+ newline();
2646
+ }
2647
+ for (const f of findings) {
2648
+ console.log(` ${yellow(symbols.warning)} ${bold(cyan(`${f.table}.${f.column}`))} ${gray('SPARSE-VALUE FLIP')}`);
2649
+ console.log(` ${dim(symbols.tee)} generic estimate ${bold(divInt(f.genericEstimate))} rows ${dim(`(${divInt(f.rows)} rows / ${divInt(f.distinctValues)} distinct values)`)}`);
2650
+ console.log(` ${dim(symbols.tee)} rarest value bucket ${bold(divInt(f.rarestBucket))} rows ${dim('(pg_stats most_common_freqs / residual bucket)')}`);
2651
+ console.log(` ${dim(symbols.tee)} crossover ${bold(divInt(f.crossoverRows))} rows ${dim(`(sqrt(limit ${f.assumedLimit} x ${divInt(f.pages)} pages); ${divInt(f.crossoverRowsWide)} at limit ${f.thresholds.wideLimit})`)}`);
2652
+ const analyzed = f.lastAnalyze === null
2653
+ ? 'last ANALYZE unknown'
2654
+ : `last analyzed ${Math.max(0, Math.round((Date.now() - f.lastAnalyze.getTime()) / 86_400_000))}d ago`;
2655
+ console.log(` ${dim(symbols.tee)} ${dim(`values below the crossover: ${divInt(f.valuesBelowCrossover)} of ${divInt(f.distinctValues)}, correlation ${f.correlation.toFixed(2)}, ${analyzed}`)}`);
2656
+ console.log(` ${dim(symbols.tee)} for such a value the generic plan walks ~${bold(divInt(f.walkPages))} of ${divInt(f.pages)} pages ${dim(`(${Math.round(f.walkFraction * 100)}% of the table)`)}`);
2657
+ console.log(` ${dim(`for reads shaped WHERE ${f.column} = $1 ORDER BY ${f.orderColumn} LIMIT $n,`)}`);
2658
+ console.log(` ${dim("where the custom plan reads only that value's own rows.")}`);
2659
+ console.log(` ${dim(symbols.tee)} ${dim('No amplification figure is printed, deliberately. This models how many rows a')}`);
2660
+ console.log(` ${dim('value has, not WHERE they sit in the heap, and the second half can move the')}`);
2661
+ console.log(` ${dim('real cost by an order of magnitude. Measure it instead:')}`);
2662
+ console.log(` ${dim(symbols.teeEnd)} ${dim('confirm with YOUR values before changing anything:')}`);
2663
+ for (const line of f.diagnosticSql.split('\n')) {
2664
+ console.log(` ${green(line)}`);
2665
+ }
2666
+ newline();
2667
+ }
2668
+ if (findings.length > 0) {
2669
+ console.log(` ${bold('What to do, in order:')}`);
2670
+ console.log(` 1. Check that this shape is promoted AT ALL. Step 1 of the block above: while`);
2671
+ console.log(` ${dim('generic_plans is 0, Postgres is planning with your real values and there is nothing')}`);
2672
+ console.log(` ${dim('to fix. A finding is exposure, not an incident, and many shapes never promote.')}`);
2673
+ console.log(` 2. If it does promote, compare the two plans. Both SETs matter: without them a`);
2674
+ console.log(` ${dim('repeated seq scan resumes where the last one stopped and a catastrophic case reads')}`);
2675
+ console.log(` ${dim('as harmless.')}`);
2676
+ console.log(` 3. If the flip is real, scope the fix to those reads:`);
2677
+ const first = findings[0];
2678
+ console.log(` ${cyan(`db.${first.table}.findMany({ where: { ${first.columnField}: value }, orderBy: { ${first.orderColumnField}: 'asc' },`)}`);
2679
+ console.log(` ${cyan(` limit: 20, forceCustomPlan: true })`)}`);
2680
+ console.log(` ${dim('That withholds the prepared-statement NAME for that one query, so the driver')}`);
2681
+ console.log(` ${dim('re-parses it every execution and it is always planned with the real values. No')}`);
2682
+ console.log(` ${dim('GUC, no SET LOCAL, no transaction, no extra round trip.')}`);
2683
+ console.log(` 4. Do NOT set planCacheMode on the client to fix this. There are measured shapes`);
2684
+ console.log(` ${dim('where a generic plan is dramatically better (an unordered LIMIT over a value whose')}`);
2685
+ console.log(` ${dim('rows are packed at the end of the heap: 4,262 buffers custom vs 71 generic on a')}`);
2686
+ console.log(` ${dim('reproducible fixture). A custom plan is not automatically the better plan.')}`);
2687
+ console.log(` 5. A composite index on (${first.column}, ${first.orderColumn}) makes the GOOD plan better. It`);
2688
+ console.log(` ${dim('does NOT stop the generic plan from choosing the other one, and it can widen the gap.')}`);
2689
+ console.log(` ${dim('Add it for the custom-plan win, not as a fix for this finding.')}`);
2690
+ newline();
2691
+ console.log(` ${dim('This finding is derived from statistics, not from your traffic: it says the DISTRIBUTION')}`);
2692
+ console.log(` ${dim('admits a damaging flip, not that a query is running one today. It models ONE shape,')}`);
2693
+ console.log(` ${dim('the rare value that loses its bitmap plan. It cannot see where a value physically')}`);
2694
+ console.log(` ${dim('sits in the heap, so a clean report is not evidence of immunity.')}`);
2695
+ console.log(` ${dim(`Gates: the wrong plan must walk >= ${PLAN_DIVERGENCE_THRESHOLDS.minWalkPages} pages and >= ${Math.round(PLAN_DIVERGENCE_THRESHOLDS.minWalkFraction * 100)}% of the table, at an assumed`)}`);
2696
+ console.log(` ${dim(`LIMIT ${PLAN_DIVERGENCE_THRESHOLDS.assumedLimit}. ${divergence.candidatesConsidered} column(s) were scored. Skip this section with --no-plan-divergence.`)}`);
2697
+ newline();
2698
+ }
2699
+ if (notices.length > 0) {
2700
+ console.log(` ${dim('Not scored for cached-plan divergence (statistics missing):')}`);
2701
+ for (const n of notices) {
2702
+ console.log(` ${dim(`- ${n.table}.${n.column}: ${n.reason}`)}`);
2703
+ }
2704
+ newline();
2705
+ }
2706
+ }
2597
2707
  /** Write the --fix migration (CONCURRENTLY + directive by default; plain with --no-concurrently). */
2598
2708
  function renderFixMigration(findings, config, args) {
2599
2709
  const concurrently = args.noConcurrently !== true;
@@ -3067,7 +3177,7 @@ function showHelp() {
3067
3177
  console.log(` ${dim('status')} Show applied/pending migrations`);
3068
3178
  console.log(` ${cyan('seed')} Run seed file`);
3069
3179
  console.log(` ${cyan('status')} ${dim('| info')} Show schema summary`);
3070
- console.log(` ${cyan('doctor')} Cost-aware missing-FK-index triage ${dim('(--fix, --json, --unused, --audit)')}`);
3180
+ console.log(` ${cyan('doctor')} Index + cached-plan triage ${dim('(--fix, --json, --unused, --audit)')}`);
3071
3181
  console.log(` ${cyan('studio')} Launch local read-only web UI ${dim('(--write for writes, --demo for a sample DB)')}`);
3072
3182
  console.log(` ${cyan('mcp')} Start read-only MCP server over stdio`);
3073
3183
  console.log(` ${cyan('observe')} Launch metrics dashboard ${dim('(requires TURBINE_OBSERVE_URL)')}`);
package/dist/client.d.ts CHANGED
@@ -295,7 +295,22 @@ export interface TurbineConfig {
295
295
  * Default `undefined`: Turbine issues NOTHING and the backend keeps its own
296
296
  * default (`auto`), byte-identical to not setting the option.
297
297
  *
298
- * SESSION-LEVEL, NOT PER QUERY. It is applied as a connection parameter
298
+ * SESSION-LEVEL. THE PER-QUERY LEVER IS A DIFFERENT, NARROWER ONE. This
299
+ * option is a connection parameter and cannot be unset for a single query, so
300
+ * a client-wide value cannot express "custom here, `auto` there". The read
301
+ * option `forceCustomPlan: true` covers that case, and only that case: it
302
+ * withholds the prepared-statement NAME for one query, and because the driver
303
+ * re-parses an unnamed statement on every execution, the counter that
304
+ * promotion depends on is reset before it is ever reached. That
305
+ * mechanism can only ever mean custom, so there is deliberately no per-query
306
+ * `force_generic_plan`; that direction stays here, at the connection. The two
307
+ * do NOT compose in the other order either: `'force_generic_plan'` set here
308
+ * governs unnamed statements as well as named ones (measured), so a
309
+ * per-query `forceCustomPlan: true` cannot escape it and is REFUSED with
310
+ * `ValidationError` (E003) instead of silently doing nothing. Leave this
311
+ * option unset (or `'auto'`) if any query needs the per-query lever.
312
+ *
313
+ * IT IS APPLIED as a connection parameter
299
314
  * (`options=-c plan_cache_mode=...`) when the pool opens a connection, so it
300
315
  * is in force for that connection's very first statement and persists for its
301
316
  * whole life: every pooled checkout, `$transaction`, stream and pipeline on
package/dist/client.js CHANGED
@@ -650,6 +650,10 @@ export class TurbineClient {
650
650
  jsonEncoding: config.jsonEncoding,
651
651
  globalFilters: config.globalFilters,
652
652
  preparedStatements: envDisablePrepared ? false : (config.preparedStatements ?? !config.pool),
653
+ // Forwarded so a per-query `forceCustomPlan` can refuse the combination
654
+ // this client's own connections would silently defeat, see the
655
+ // `planCacheMode` note on QueryInterfaceOptions.
656
+ planCacheMode,
653
657
  sqlCache: config.sqlCache ?? true,
654
658
  sqlCacheSize: config.sqlCacheSize,
655
659
  dialect: config.dialect,
package/dist/dialect.d.ts CHANGED
@@ -375,6 +375,15 @@ export interface Dialect {
375
375
  * setting (`SET plan_cache_mode = auto | force_custom_plan |
376
376
  * force_generic_plan`). Gates the opt-in `planCacheMode` client option.
377
377
  *
378
+ * It ALSO gates the per-query `forceCustomPlan` read option, which uses a
379
+ * different mechanism (it withholds the prepared-statement name, so the
380
+ * driver re-parses the statement on every execution and the counter that
381
+ * generic-plan promotion depends on never reaches its threshold) but
382
+ * asks the identical capability question: does this engine have a PostgreSQL
383
+ * plan cache whose generic-plan promotion is worth pinning? An engine that
384
+ * answers no cannot honour either option, so both refuse on the same flag
385
+ * rather than on two flags that could never disagree.
386
+ *
378
387
  * A capability flag rather than a `dialect.name === 'postgresql'` test, for
379
388
  * the same reason every other refusal here is one: the setting is a property
380
389
  * of the PostgreSQL PLAN CACHE, not of the SQL string, so a
@@ -86,6 +86,12 @@ export interface TableStats {
86
86
  table: string;
87
87
  /** pg_class.reltuples. 0 or -1 means never-analyzed → treated as UNKNOWN (null rows). */
88
88
  reltuples: number;
89
+ /**
90
+ * pg_class.relpages. The SIZE input the plan-divergence crossover is computed
91
+ * from (a plan boundary tracks pages relative to the LIMIT, not rows).
92
+ * Absent when the pg_class read degraded; 0 means never analyzed.
93
+ */
94
+ relpages?: number;
89
95
  /** pg_stat_user_tables.n_live_tup, a cross-check for reltuples. */
90
96
  nLiveTup?: number;
91
97
  nTupIns?: number;
@@ -100,6 +106,13 @@ export interface TableStats {
100
106
  tableSizeBytes?: number;
101
107
  /** Count of indexes already on the table (pg_index). */
102
108
  existingIndexCount?: number;
109
+ /**
110
+ * The later of pg_stat_user_tables.last_analyze / last_autoanalyze: when the
111
+ * planner's column statistics were last refreshed. This, NOT stats_reset, is
112
+ * the freshness that matters for anything read out of pg_stats. Null when
113
+ * never analyzed; absent when the pg_stat read degraded.
114
+ */
115
+ lastAnalyze?: Date | null;
103
116
  }
104
117
  /**
105
118
  * Placeholder for an index column that is an EXPRESSION, not a plain column
@@ -133,6 +146,27 @@ export interface IndexStat {
133
146
  /** pg_relation_size of the index heap, bytes. Size reclaimed by a drop. */
134
147
  sizeBytes?: number;
135
148
  }
149
+ /**
150
+ * The value distribution of one column, read from pg_stats with
151
+ * `inherited = false`. Consumed by the plan-divergence advisor, which needs to
152
+ * reproduce the planner's OWN estimates rather than approximate them.
153
+ */
154
+ export interface ColumnDistribution {
155
+ table: string;
156
+ column: string;
157
+ /**
158
+ * pg_stats.n_distinct, RAW: a positive value is a count, a negative value is a
159
+ * fraction of the row count. Kept undecoded so the consumer decodes it exactly
160
+ * the way the planner does. 0 means the column was never analyzed.
161
+ */
162
+ nDistinct: number;
163
+ /** pg_stats.correlation, signed. NULL for types with no ordering. */
164
+ correlation: number | null;
165
+ /** pg_stats.most_common_freqs, or null when the column has no MCV list. */
166
+ mostCommonFreqs: number[] | null;
167
+ /** cardinality(most_common_vals): how many values the MCV list actually covers. */
168
+ mcvCount: number;
169
+ }
136
170
  /**
137
171
  * A point-in-time read of the statistics the triage needs. Every part is
138
172
  * optional at the field level so the pure scorer degrades honestly.
@@ -150,6 +184,13 @@ export interface StatsSnapshot {
150
184
  indexes: IndexStat[];
151
185
  /** null_frac per probed column, keyed `table.column`. */
152
186
  nullFrac: Record<string, number>;
187
+ /**
188
+ * Value distribution per candidate column, keyed `table.column`. Optional at
189
+ * the snapshot level: a caller that never asked for distribution statistics
190
+ * (or whose pg_stats read degraded) leaves it absent, and the plan-divergence
191
+ * advisor reports that as a suppressed candidate rather than scoring a guess.
192
+ */
193
+ columnStats?: Record<string, ColumnDistribution>;
153
194
  /** Per-signal degradation notices (privileges, catalog gaps, timeouts). */
154
195
  notices: string[];
155
196
  }
@@ -347,6 +388,11 @@ export interface CollectSnapshotOptions {
347
388
  tables: string[];
348
389
  /** Columns to read null_frac for (single-column probes). */
349
390
  columns: ProbedColumn[];
391
+ /**
392
+ * Columns to read the full value distribution for (n_distinct, correlation,
393
+ * MCV frequencies). One extra pg_stats query; empty skips it entirely.
394
+ */
395
+ distributionColumns?: ProbedColumn[];
350
396
  /** statement_timeout for each catalog read. Default 5000ms. */
351
397
  statementTimeoutMs?: number;
352
398
  }
@@ -102,6 +102,7 @@ export function emptyStatsSnapshot(notices = []) {
102
102
  tables: {},
103
103
  indexes: [],
104
104
  nullFrac: {},
105
+ columnStats: {},
105
106
  notices,
106
107
  };
107
108
  }
@@ -544,12 +545,14 @@ export async function collectStatsSnapshot(options) {
544
545
  }
545
546
  }
546
547
  // --- table stats (pg_stat_user_tables) ---------------------------------
547
- const statRows = await run('pg_stat_user_tables', `SELECT relname, n_tup_ins, n_tup_upd, n_tup_del, n_tup_hot_upd, seq_scan, seq_tup_read, n_live_tup
548
+ const statRows = await run('pg_stat_user_tables', `SELECT relname, n_tup_ins, n_tup_upd, n_tup_del, n_tup_hot_upd, seq_scan, seq_tup_read, n_live_tup,
549
+ greatest(last_analyze, last_autoanalyze) AS last_analyze
548
550
  FROM pg_stat_user_tables
549
551
  WHERE schemaname = $1 AND relname = ANY($2)`, [options.schema, options.tables]);
550
552
  // --- class size + existing index count (pg_class) ----------------------
551
553
  const classRows = await run('pg_class size', `SELECT c.relname,
552
554
  c.reltuples::bigint::text AS reltuples,
555
+ c.relpages::bigint::text AS relpages,
553
556
  pg_total_relation_size(c.oid)::text AS total_size,
554
557
  pg_relation_size(c.oid)::text AS table_size,
555
558
  (SELECT count(*) FROM pg_index i WHERE i.indrelid = c.oid)::text AS index_count
@@ -572,6 +575,7 @@ export async function collectStatsSnapshot(options) {
572
575
  for (const row of classRows) {
573
576
  const s = ensure(row.relname);
574
577
  s.reltuples = Number(row.reltuples);
578
+ s.relpages = Number(row.relpages);
575
579
  s.totalSizeBytes = Number(row.total_size);
576
580
  s.tableSizeBytes = Number(row.table_size);
577
581
  s.existingIndexCount = Number(row.index_count);
@@ -588,6 +592,7 @@ export async function collectStatsSnapshot(options) {
588
592
  s.seqScan = Number(row.seq_scan);
589
593
  s.seqTupRead = Number(row.seq_tup_read);
590
594
  s.nLiveTup = Number(row.n_live_tup);
595
+ s.lastAnalyze = row.last_analyze == null ? null : new Date(row.last_analyze);
591
596
  }
592
597
  }
593
598
  // --- invalid + all indexes (whole schema, for invalid detection) -------
@@ -651,6 +656,42 @@ export async function collectStatsSnapshot(options) {
651
656
  }
652
657
  }
653
658
  }
659
+ // --- value distribution for divergence candidates (pg_stats) -----------
660
+ const distCols = options.distributionColumns ?? [];
661
+ if (distCols.length > 0) {
662
+ // `inherited = false` is required: on a partitioned parent the inherited
663
+ // row describes the whole tree, and a per-partition plan is not chosen
664
+ // from it. most_common_vals is an anyarray, so its cardinality is read
665
+ // through a text[] cast (array_length on anyarray cannot resolve a type).
666
+ const distRows = await run('pg_stats.distribution', `SELECT s.tablename, s.attname,
667
+ s.n_distinct::text AS n_distinct,
668
+ s.correlation::text AS correlation,
669
+ s.most_common_freqs,
670
+ coalesce(array_length(s.most_common_vals::text::text[], 1), 0)::text AS mcv_count
671
+ FROM pg_stats s
672
+ JOIN unnest($2::text[], $3::text[]) AS probe(t, c)
673
+ ON probe.t = s.tablename AND probe.c = s.attname
674
+ WHERE s.schemaname = $1 AND s.inherited = false`, [options.schema, distCols.map((c) => c.table), distCols.map((c) => c.column)]);
675
+ if (distRows) {
676
+ const byKey = snapshot.columnStats ?? {};
677
+ for (const row of distRows) {
678
+ byKey[`${row.tablename}.${row.attname}`] = {
679
+ table: row.tablename,
680
+ column: row.attname,
681
+ nDistinct: Number(row.n_distinct),
682
+ correlation: row.correlation == null ? null : Number(row.correlation),
683
+ mostCommonFreqs: row.most_common_freqs == null ? null : row.most_common_freqs.map(Number),
684
+ mcvCount: row.mcv_count == null ? 0 : Number(row.mcv_count),
685
+ };
686
+ }
687
+ snapshot.columnStats = byKey;
688
+ }
689
+ else {
690
+ // The read failed (privileges, catalog gap). Leave columnStats absent so
691
+ // the advisor suppresses every candidate instead of scoring zeros.
692
+ snapshot.columnStats = undefined;
693
+ }
694
+ }
654
695
  }
655
696
  finally {
656
697
  try {