feat: add AI catalog description generation

This commit is contained in:
Codex
2026-08-29 16:42:56 +02:00
parent b0afba81ca
commit 376dd5a09d
76 changed files with 14860 additions and 102 deletions
File diff suppressed because it is too large Load Diff
@@ -0,0 +1,163 @@
import type { CatalogPostgresAccess } from "./postgres-access.js";
import type { WorkspaceDatabase } from "./types.js";
const MAX_SOURCE_ROWS = 5;
const MAX_REPRESENTATIVE_VALUES = 5;
const MAX_SOURCE_COLUMNS_PER_TARGET = 8;
const MAX_SOURCE_VALUE_BYTES = 256;
export type DescriptionSourceSampleValue = string | number | boolean | null;
export interface DescriptionSourceSampleField {
name: string;
value: DescriptionSourceSampleValue;
}
export interface DescriptionSourceSampleRow {
fields: readonly DescriptionSourceSampleField[];
}
export interface DescriptionSourceRepresentativeValues {
column: string;
values: readonly Exclude<DescriptionSourceSampleValue, null>[];
}
export interface DescriptionTargetSourceSample {
targetId: string;
tableName: string;
rows: readonly DescriptionSourceSampleRow[];
representativeValues: readonly DescriptionSourceRepresentativeValues[];
}
export interface DescriptionSourceSamplingTarget {
targetId: string;
tableName: string;
columnNames: readonly string[];
}
/** Optional, transient source context for one model-completion batch. */
export interface DescriptionSourceSampler {
sample(
database: WorkspaceDatabase,
targets: readonly DescriptionSourceSamplingTarget[],
signal: AbortSignal,
): Promise<readonly DescriptionTargetSourceSample[]>;
}
function quoteIdentifier(identifier: string): string {
return `"${identifier.replaceAll('"', '""')}"`;
}
function boundedUtf8(value: string, maxBytes: number): string {
const normalized = value
.normalize("NFC")
.replace(/\r\n?/g, "\n")
.replace(/[\u0000-\u0008\u000b\u000c\u000e-\u001f\u007f]/g, " ");
if (Buffer.byteLength(normalized, "utf8") <= maxBytes) return normalized;
let result = "";
let bytes = 0;
for (const character of normalized) {
const characterBytes = Buffer.byteLength(character, "utf8");
if (bytes + characterBytes > maxBytes) break;
result += character;
bytes += characterBytes;
}
return result;
}
function normalizeValue(value: unknown): DescriptionSourceSampleValue | undefined {
if (value === null) return null;
if (typeof value === "string") return boundedUtf8(value, MAX_SOURCE_VALUE_BYTES);
if (typeof value === "boolean") return value;
if (typeof value === "number") return Number.isFinite(value) ? value : undefined;
if (typeof value === "bigint") return boundedUtf8(String(value), MAX_SOURCE_VALUE_BYTES);
if (value instanceof Date && !Number.isNaN(value.valueOf())) return value.toISOString();
return undefined;
}
function distinctKey(value: Exclude<DescriptionSourceSampleValue, null>): string {
return `${typeof value}:${String(value)}`;
}
/** PostgreSQL-wire sampler. REST bindings remain unsupported by CatalogPostgresAccess. */
export class PostgresDescriptionSourceSampler implements DescriptionSourceSampler {
constructor(private readonly access: CatalogPostgresAccess) {}
async sample(
database: WorkspaceDatabase,
targets: readonly DescriptionSourceSamplingTarget[],
signal: AbortSignal,
): Promise<readonly DescriptionTargetSourceSample[]> {
const client = await this.access.connect(database, signal);
let transactionOpen = false;
try {
await client.query("BEGIN TRANSACTION READ ONLY", []);
transactionOpen = true;
const samples: DescriptionTargetSourceSample[] = [];
for (const target of targets) {
const columnNames = [...new Set(target.columnNames)].slice(0, MAX_SOURCE_COLUMNS_PER_TARGET);
if (columnNames.length === 0) {
samples.push({
targetId: target.targetId,
tableName: target.tableName,
rows: [],
representativeValues: [],
});
continue;
}
const projections = columnNames.map((columnName) => {
const identifier = quoteIdentifier(columnName);
return `LEFT((${identifier})::text, $1) AS ${identifier}`;
});
const sql = [
`SELECT ${projections.join(", ")}`,
`FROM ${quoteIdentifier(database.schema)}.${quoteIdentifier(target.tableName)}`,
"LIMIT $2",
].join(" ");
const result = await client.query(sql, [MAX_SOURCE_VALUE_BYTES, MAX_SOURCE_ROWS]);
const rows = result.rows.slice(0, MAX_SOURCE_ROWS).map((row) => ({
fields: columnNames.flatMap((name) => {
const value = normalizeValue(row[name]);
return value === undefined ? [] : [{ name, value }];
}),
}));
const valuesByColumn = new Map<
string,
Exclude<DescriptionSourceSampleValue, null>[]
>();
const seenByColumn = new Map<string, Set<string>>();
let representativeValueCount = 0;
for (const row of rows) {
for (const field of row.fields) {
if (representativeValueCount === MAX_REPRESENTATIVE_VALUES) break;
if (field.value === null) continue;
const seen = seenByColumn.get(field.name) ?? new Set<string>();
const key = distinctKey(field.value);
if (seen.has(key)) continue;
seen.add(key);
seenByColumn.set(field.name, seen);
const values = valuesByColumn.get(field.name) ?? [];
values.push(field.value);
valuesByColumn.set(field.name, values);
representativeValueCount += 1;
}
if (representativeValueCount === MAX_REPRESENTATIVE_VALUES) break;
}
const representativeValues = columnNames.flatMap((column) => {
const values = valuesByColumn.get(column);
return values && values.length > 0 ? [{ column, values }] : [];
});
samples.push({
targetId: target.targetId,
tableName: target.tableName,
rows,
representativeValues,
});
}
return samples;
} finally {
if (transactionOpen) await client.query("ROLLBACK", []).catch(() => undefined);
await client.end().catch(() => undefined);
}
}
}
+174
View File
@@ -2,7 +2,10 @@ import { randomUUID } from "node:crypto";
import {
CatalogConflictError,
CatalogConnectorError,
DescriptionGenerationRunActiveError,
type CatalogColumn,
type CatalogDescriptionConsolidationCounts,
type CatalogDescriptionTarget,
type CatalogDatabaseMetadataDeleteTarget,
type CatalogMetadataDeleteCounts,
type CatalogRelationship,
@@ -19,6 +22,10 @@ import {
type DatabaseTestResult,
type ObservedCatalogTable,
type ObservedSchemaSnapshot,
type DescriptionGenerationEvent,
type DescriptionGenerationRun,
type DescriptionGenerationRunUpdate,
type DescriptionGenerationScope,
type TableSyncRepositoryResult,
type WorkspaceDatabase,
} from "./types.js";
@@ -33,6 +40,8 @@ export class MemoryCatalogRepository implements CatalogRepository {
private readonly tables = new Map<string, CatalogTable>();
private readonly columns = new Map<string, CatalogColumn>();
private readonly relationships = new Map<string, CatalogRelationship>();
private readonly descriptionGenerationRuns = new Map<string, DescriptionGenerationRun>();
private readonly descriptionGenerationEvents = new Map<string, DescriptionGenerationEvent[]>();
private readonly syncRuns = new Map<string, CatalogSyncRun>();
private readonly syncEvents = new Map<string, CatalogSyncEvent[]>();
@@ -122,6 +131,11 @@ export class MemoryCatalogRepository implements CatalogRepository {
for (const [relationshipId, relationship] of this.relationships) {
if (relationship.databaseId === id) this.relationships.delete(relationshipId);
}
for (const [runId, run] of this.descriptionGenerationRuns) {
if (run.databaseId !== id) continue;
this.descriptionGenerationRuns.delete(runId);
this.descriptionGenerationEvents.delete(runId);
}
return this.records.delete(id);
}
async listTables(databaseId: string): Promise<CatalogTable[]> {
@@ -196,6 +210,166 @@ export class MemoryCatalogRepository implements CatalogRepository {
return structuredClone(updated);
}
async consolidateGeneratedDescriptions(
databaseId: string,
target: CatalogDescriptionTarget,
targetIds: readonly string[],
): Promise<CatalogDescriptionConsolidationCounts | undefined> {
const selectedTargetIds = [...new Set(targetIds)];
if (!this.records.has(databaseId) || selectedTargetIds.length === 0) {
return undefined;
}
const now = new Date().toISOString();
if (target === "tables") {
const targets = selectedTargetIds.map((id) => this.tables.get(id));
if (targets.some((table) => !table || table.databaseId !== databaseId)) return undefined;
const copied = targets.filter((table) => Boolean(table!.generatedDescription?.trim()));
for (const table of copied) {
this.tables.set(table!.id, {
...table!,
description: table!.generatedDescription,
version: table!.version + 1,
updatedAt: now,
});
}
return { copied: copied.length, skipped: targets.length - copied.length };
}
const targets = selectedTargetIds.map((id) => this.columns.get(id));
if (targets.some((column) => (
!column || this.tables.get(column.tableId)?.databaseId !== databaseId
))) return undefined;
const copied = targets.filter((column) => Boolean(column!.generatedDescription?.trim()));
for (const column of copied) {
this.columns.set(column!.id, {
...column!,
description: column!.generatedDescription,
version: column!.version + 1,
updatedAt: now,
});
}
return { copied: copied.length, skipped: targets.length - copied.length };
}
async createDescriptionGenerationRun(
databaseId: string,
scope: DescriptionGenerationScope,
modelId: string,
language: DescriptionGenerationRun["language"],
total: number,
): Promise<DescriptionGenerationRun> {
if ([...this.descriptionGenerationRuns.values()].some((run) => (
run.status === "queued" || run.status === "running"
))) {
throw new DescriptionGenerationRunActiveError("A description generation run is already active");
}
const now = new Date().toISOString();
const run: DescriptionGenerationRun = {
id: randomUUID(),
databaseId,
scope,
modelId,
language,
status: "queued",
total,
processed: 0,
generated: 0,
nonGeneratable: 0,
failed: 0,
createdAt: now,
startedAt: null,
updatedAt: now,
finishedAt: null,
errorSummary: null,
};
this.descriptionGenerationRuns.set(run.id, run);
return structuredClone(run);
}
async getDescriptionGenerationRun(runId: string): Promise<DescriptionGenerationRun | undefined> {
const run = this.descriptionGenerationRuns.get(runId);
return run ? structuredClone(run) : undefined;
}
async listDescriptionGenerationRuns(limit = 50): Promise<DescriptionGenerationRun[]> {
return [...this.descriptionGenerationRuns.values()]
.sort((a, b) => b.createdAt.localeCompare(a.createdAt) || b.id.localeCompare(a.id))
.slice(0, limit)
.map((run) => structuredClone(run));
}
async getActiveDescriptionGenerationRun(): Promise<DescriptionGenerationRun | undefined> {
const run = [...this.descriptionGenerationRuns.values()]
.filter((candidate) => candidate.status === "queued" || candidate.status === "running")
.sort((a, b) => b.createdAt.localeCompare(a.createdAt))[0];
return run ? structuredClone(run) : undefined;
}
async interruptActiveDescriptionGenerationRuns(
errorSummary: string,
): Promise<DescriptionGenerationRun[]> {
const interrupted: DescriptionGenerationRun[] = [];
for (const run of this.descriptionGenerationRuns.values()) {
if (run.status !== "queued" && run.status !== "running") continue;
const now = new Date().toISOString();
const updated: DescriptionGenerationRun = {
...run,
status: "interrupted",
updatedAt: now,
finishedAt: now,
errorSummary,
};
this.descriptionGenerationRuns.set(run.id, updated);
interrupted.push(structuredClone(updated));
}
return interrupted;
}
async updateDescriptionGenerationRun(
runId: string,
update: DescriptionGenerationRunUpdate,
): Promise<DescriptionGenerationRun | undefined> {
const current = this.descriptionGenerationRuns.get(runId);
if (!current) return undefined;
const updated = {
...current,
...structuredClone(update),
updatedAt: new Date().toISOString(),
};
this.descriptionGenerationRuns.set(runId, updated);
return structuredClone(updated);
}
async appendDescriptionGenerationEvent(
runId: string,
level: DescriptionGenerationEvent["level"],
message: string,
): Promise<DescriptionGenerationEvent> {
if (!this.descriptionGenerationRuns.has(runId)) {
throw new CatalogConflictError("Description Generation Run does not exist");
}
const events = this.descriptionGenerationEvents.get(runId) ?? [];
const event: DescriptionGenerationEvent = {
runId,
sequence: events.length + 1,
level,
message,
createdAt: new Date().toISOString(),
};
events.push(event);
this.descriptionGenerationEvents.set(runId, events);
return structuredClone(event);
}
async listDescriptionGenerationEvents(
runId: string,
afterSequence = 0,
): Promise<DescriptionGenerationEvent[]> {
return (this.descriptionGenerationEvents.get(runId) ?? [])
.filter((event) => event.sequence > afterSequence)
.map((event) => structuredClone(event));
}
async listRelationships(databaseId: string): Promise<CatalogRelationship[]> {
return [...this.relationships.values()].filter((relationship) => relationship.databaseId === databaseId)
.sort((a, b) => `${a.sourceTableName}.${a.constraintName}`.localeCompare(`${b.sourceTableName}.${b.constraintName}`))
@@ -0,0 +1,235 @@
import {
closeSync, constants, fstatSync, lstatSync, openSync, readFileSync,
type Stats,
} from "node:fs";
import { parseAllDocuments } from "yaml";
import { z } from "zod";
import {
loadSecretBundle,
METADATA_GENERATION_SECRET_KEYS,
} from "../config/secret-bundle.js";
const MAX_INSTALLATION_BYTES = 1024 * 1024;
const RUNTIME_INSTALLATION_FILE = "/run/thothii-installation/thothii-installation.yaml";
const modelId = z.string().regex(/^[a-z][a-z0-9._-]{0,63}$/);
const apiKeyEnvironment = z.enum(METADATA_GENERATION_SECRET_KEYS);
const endpointSchema = z.object({
baseUrl: z.string().min(1).max(2048).refine((value) => {
try {
const url = new URL(value);
return (url.protocol === "http:" || url.protocol === "https:")
&& url.username === "" && url.password === "" && url.search === "" && url.hash === "";
} catch {
return false;
}
}),
apiVersion: z.string().regex(/^[A-Za-z0-9][A-Za-z0-9._-]{0,127}$/).optional(),
}).strict();
const configuredModelSchema = z.object({
id: modelId,
label: z.string().min(1).max(128).refine((value) => value.trim() === value && !/\p{Cc}/u.test(value)),
litellm: z.object({
provider: z.string().regex(/^[A-Za-z0-9][A-Za-z0-9._-]{0,63}$/),
model: z.string().regex(/^[A-Za-z0-9][A-Za-z0-9._:/-]{0,255}$/),
disableThinking: z.literal(true).optional(),
endpoint: endpointSchema.optional(),
}).strict(),
apiKeyEnv: apiKeyEnvironment.optional(),
}).strict().superRefine((value, context) => {
if (value.apiKeyEnv === undefined && value.litellm.endpoint === undefined) {
context.addIssue({
code: z.ZodIssueCode.custom,
path: ["apiKeyEnv"],
message: "keyless models require an explicit endpoint",
});
}
if (value.litellm.disableThinking === true && value.litellm.endpoint === undefined) {
context.addIssue({
code: z.ZodIssueCode.custom,
path: ["litellm", "disableThinking"],
message: "thinking may be disabled only for an explicit endpoint",
});
}
});
const metadataGenerationSchema = z.object({
default: modelId.optional(),
models: z.array(configuredModelSchema).max(64).default([]),
}).strict();
const installationSchema = z.object({
metadataGeneration: metadataGenerationSchema.optional(),
}).passthrough();
export interface MetadataGenerationModelChoice {
id: string;
label: string;
}
export interface MetadataGenerationModelCatalog {
models: MetadataGenerationModelChoice[];
default: string | null;
}
export interface ResolvedMetadataGenerationModel {
readonly id: string;
readonly provider: string;
readonly model: string;
readonly disableThinking?: true;
readonly endpoint?: Readonly<{ baseUrl: string; apiVersion?: string }>;
readonly apiKeyEnv?: string;
readonly apiKey?: string;
}
export class MetadataGenerationModelUnavailableError extends Error {
constructor() {
super("metadata-generation model is unavailable");
this.name = "MetadataGenerationModelUnavailableError";
}
}
/** The complete interface callers need: safe discovery plus fail-closed runtime resolution. */
export interface MetadataGenerationModels {
catalog(): MetadataGenerationModelCatalog;
resolve(selection: string): ResolvedMetadataGenerationModel;
}
class RestartLoadedMetadataGenerationModels implements MetadataGenerationModels {
readonly #models: ReadonlyMap<string, ResolvedMetadataGenerationModel>;
readonly #catalog: MetadataGenerationModelCatalog;
constructor(
models: ReadonlyMap<string, ResolvedMetadataGenerationModel> = new Map(),
defaultModel: string | null = null,
choices: MetadataGenerationModelChoice[] = [],
) {
this.#models = models;
this.#catalog = {
models: choices.map((choice) => ({ ...choice })),
default: defaultModel,
};
}
catalog(): MetadataGenerationModelCatalog {
return {
models: this.#catalog.models.map((choice) => ({ ...choice })),
default: this.#catalog.default,
};
}
resolve(selection: string): ResolvedMetadataGenerationModel {
const model = typeof selection === "string" ? this.#models.get(selection) : undefined;
if (!model) throw new MetadataGenerationModelUnavailableError();
return model;
}
}
function invalid(message = "metadata-generation configuration is invalid"): Error {
return new Error(message);
}
function protectedInstallationStat(file: string, info: Stats): boolean {
const mode = info.mode & 0o777;
if (!info.isFile() || info.isSymbolicLink() || info.nlink !== 1
|| info.size < 1 || info.size > MAX_INSTALLATION_BYTES) return false;
if (file === RUNTIME_INSTALLATION_FILE && info.uid === 0 && mode === 0o444) return true;
return info.uid === (process.getuid?.() ?? info.uid) && (mode === 0o400 || mode === 0o600);
}
function readProtectedInstallation(file: string): string {
let descriptor: number | undefined;
try {
const before = lstatSync(file);
if (!protectedInstallationStat(file, before)) throw new Error("unavailable");
descriptor = openSync(file, constants.O_RDONLY | constants.O_NOFOLLOW);
const opened = fstatSync(descriptor);
if (!protectedInstallationStat(file, opened)
|| before.dev !== opened.dev || before.ino !== opened.ino) throw new Error("unavailable");
const source = readFileSync(descriptor, "utf8");
const after = fstatSync(descriptor);
const current = lstatSync(file);
if (!protectedInstallationStat(file, after) || !protectedInstallationStat(file, current)
|| opened.dev !== after.dev || opened.ino !== after.ino
|| opened.dev !== current.dev || opened.ino !== current.ino) throw new Error("unavailable");
return source;
} finally {
if (descriptor !== undefined) try { closeSync(descriptor); } catch { /* sanitized below */ }
}
}
function readInstallation(file: string): unknown {
try {
const documents = parseAllDocuments(readProtectedInstallation(file), { uniqueKeys: true });
if (documents.length !== 1) throw invalid("metadata-generation installation must contain one YAML document");
const document = documents[0];
if (document.errors.length > 0 || document.warnings.length > 0) {
throw invalid("metadata-generation installation contains invalid YAML");
}
return document.toJSON();
} catch (error) {
if (error instanceof Error && error.message.startsWith("metadata-generation")) throw error;
throw invalid("metadata-generation installation is unavailable");
}
}
export function loadMetadataGenerationModels(options: {
installationFile?: string;
secretsFile?: string;
}): MetadataGenerationModels {
if (!options.installationFile) return new RestartLoadedMetadataGenerationModels();
const installation = installationSchema.safeParse(readInstallation(options.installationFile));
if (!installation.success) throw invalid();
const configured = installation.data.metadataGeneration;
if (!configured || configured.models.length === 0) {
if (configured?.default !== undefined) throw invalid("metadata-generation default does not identify a configured model");
return new RestartLoadedMetadataGenerationModels();
}
if (!configured.default) throw invalid("metadata-generation default is required when models are configured");
const seen = new Set<string>();
for (const model of configured.models) {
if (seen.has(model.id)) throw invalid(`metadata-generation model id "${model.id}" is duplicated`);
seen.add(model.id);
}
if (!seen.has(configured.default)) {
throw invalid(`metadata-generation default "${configured.default}" is not configured`);
}
const requiresSecrets = configured.models.some((model) => model.apiKeyEnv !== undefined);
let secrets: ReadonlyMap<string, string> = new Map();
if (requiresSecrets) {
if (!options.secretsFile) throw invalid("metadata-generation keyed models require THT_SECRETS_FILE");
try {
secrets = loadSecretBundle(options.secretsFile);
} catch {
throw invalid("metadata-generation secrets are unavailable");
}
}
const models = new Map<string, ResolvedMetadataGenerationModel>();
for (const configuredModel of configured.models) {
let apiKey: string | undefined;
if (configuredModel.apiKeyEnv !== undefined) {
apiKey = secrets.get(configuredModel.apiKeyEnv);
if (!apiKey) {
throw invalid(`metadata-generation model "${configuredModel.id}" secret "${configuredModel.apiKeyEnv}" is missing`);
}
if (apiKey.length > 16 * 1024 || /\s/u.test(apiKey)) {
throw invalid(`metadata-generation model "${configuredModel.id}" secret "${configuredModel.apiKeyEnv}" is unusable`);
}
}
models.set(configuredModel.id, Object.freeze({
id: configuredModel.id,
provider: configuredModel.litellm.provider,
model: configuredModel.litellm.model,
...(configuredModel.litellm.disableThinking === true ? { disableThinking: true as const } : {}),
...(configuredModel.litellm.endpoint === undefined
? {}
: { endpoint: Object.freeze({ ...configuredModel.litellm.endpoint }) }),
...(configuredModel.apiKeyEnv === undefined
? {}
: { apiKeyEnv: configuredModel.apiKeyEnv, apiKey }),
}));
}
return new RestartLoadedMetadataGenerationModels(
models,
configured.default,
configured.models.map(({ id, label }) => ({ id, label })),
);
}
+2
View File
@@ -7,6 +7,7 @@ import * as initialMigration from "./migrations/001_workspace_databases.js";
import * as catalogTablesMigration from "./migrations/002_catalog_tables.js";
import * as catalogSchemaSyncMigration from "./migrations/003_catalog_schema_sync.js";
import * as catalogRuntimeSequencePrivilegesMigration from "./migrations/004_catalog_runtime_sequence_privileges.js";
import * as descriptionGenerationRunsMigration from "./migrations/005_description_generation_runs.js";
const connectionString = process.env.THT_CATALOG_MIGRATOR_DATABASE_URL;
const host = process.env.THT_CATALOG_DB_HOST;
@@ -36,6 +37,7 @@ const provider: MigrationProvider = {
"002_catalog_tables": catalogTablesMigration,
"003_catalog_schema_sync": catalogSchemaSyncMigration,
"004_catalog_runtime_sequence_privileges": catalogRuntimeSequencePrivilegesMigration,
"005_description_generation_runs": descriptionGenerationRunsMigration,
};
},
};
@@ -0,0 +1,87 @@
import { sql, type Kysely } from "kysely";
import type { CatalogDatabase } from "../repository.js";
export async function up(db: Kysely<CatalogDatabase>): Promise<void> {
await db.schema.createTable("description_generation_runs")
.addColumn("id", "uuid", (column) => column.primaryKey())
.addColumn("database_id", "uuid", (column) => column.notNull()
.references("workspace_databases.id").onDelete("cascade"))
.addColumn("scope", "text", (column) => column.notNull())
.addColumn("model_id", "text", (column) => column.notNull())
.addColumn("language", "text", (column) => column.notNull())
.addColumn("status", "text", (column) => column.notNull())
.addColumn("total", "integer", (column) => column.notNull())
.addColumn("processed", "integer", (column) => column.notNull().defaultTo(0))
.addColumn("generated", "integer", (column) => column.notNull().defaultTo(0))
.addColumn("non_generatable", "integer", (column) => column.notNull().defaultTo(0))
.addColumn("failed", "integer", (column) => column.notNull().defaultTo(0))
.addColumn("created_at", "timestamptz", (column) => column.notNull().defaultTo(sql`now()`))
.addColumn("started_at", "timestamptz")
.addColumn("updated_at", "timestamptz", (column) => column.notNull().defaultTo(sql`now()`))
.addColumn("finished_at", "timestamptz")
.addColumn("error_summary", "text")
.addCheckConstraint(
"description_generation_runs_scope_check",
sql`scope in ('selected_columns', 'selected_tables', 'all', 'missing')`,
)
.addCheckConstraint(
"description_generation_runs_model_id_check",
sql`model_id ~ '^[a-z][a-z0-9._-]{0,63}$'`,
)
.addCheckConstraint(
"description_generation_runs_language_check",
sql`language in ('en', 'it')`,
)
.addCheckConstraint(
"description_generation_runs_status_check",
sql`status in (
'queued', 'running', 'completed', 'completed_with_errors',
'cancelled', 'failed', 'interrupted'
)`,
)
.addCheckConstraint(
"description_generation_runs_counters_check",
sql`total > 0
and processed between 0 and total
and generated >= 0
and non_generatable >= 0
and failed >= 0
and generated + non_generatable + failed <= processed`,
)
.addCheckConstraint(
"description_generation_runs_error_summary_check",
sql`error_summary is null or char_length(error_summary) between 1 and 2000`,
)
.execute();
await db.schema.createIndex("description_generation_runs_database_created_idx")
.on("description_generation_runs")
.columns(["database_id", "created_at"])
.execute();
await sql`CREATE UNIQUE INDEX description_generation_runs_one_active
ON description_generation_runs ((true))
WHERE status IN ('queued', 'running')`.execute(db);
await db.schema.createTable("description_generation_events")
.addColumn("run_id", "uuid", (column) => column.notNull()
.references("description_generation_runs.id").onDelete("cascade"))
.addColumn("sequence", "integer", (column) => column.notNull())
.addColumn("level", "text", (column) => column.notNull())
.addColumn("message", "text", (column) => column.notNull())
.addColumn("created_at", "timestamptz", (column) => column.notNull().defaultTo(sql`now()`))
.addPrimaryKeyConstraint("description_generation_events_pkey", ["run_id", "sequence"])
.addCheckConstraint("description_generation_events_sequence_check", sql`sequence > 0`)
.addCheckConstraint(
"description_generation_events_level_check",
sql`level in ('info', 'warning', 'error')`,
)
.addCheckConstraint(
"description_generation_events_message_check",
sql`char_length(message) between 1 and 2000`,
)
.execute();
}
export async function down(db: Kysely<CatalogDatabase>): Promise<void> {
await db.schema.dropTable("description_generation_events").execute();
await db.schema.dropTable("description_generation_runs").execute();
}
+150
View File
@@ -0,0 +1,150 @@
import { spawn } from "node:child_process";
import { z } from "zod";
import type { ResolvedMetadataGenerationModel } from "./metadata-generation-models.js";
const MAX_HELPER_OUTPUT_BYTES = 64 * 1024;
const helperOutputSchema = z.discriminatedUnion("ok", [
z.object({ ok: z.literal(true), content: z.string() }).strict(),
z.object({ ok: z.literal(false), error: z.literal("provider_failure") }).strict(),
]);
export interface ModelCompletionMessage {
role: "system" | "user";
content: string;
}
export interface ModelCompletionRequest {
model: ResolvedMetadataGenerationModel;
messages: readonly ModelCompletionMessage[];
signal: AbortSignal;
}
/** The provider boundary used by Description Generation. */
export interface ModelCompleter {
complete(request: ModelCompletionRequest): Promise<string>;
}
export class ModelCompletionProviderError extends Error {
constructor() {
super("model completion failed");
this.name = "ModelCompletionProviderError";
}
}
export class ModelCompletionCancelledError extends Error {
constructor() {
super("model completion cancelled");
this.name = "ModelCompletionCancelledError";
}
}
export class PythonModelCompleter implements ModelCompleter {
constructor(private readonly options: {
pythonExecutable: string;
cwd: string;
helperModule?: string;
timeoutMs?: number;
terminationGraceMs?: number;
}) {}
async complete(request: ModelCompletionRequest): Promise<string> {
if (request.signal.aborted) throw new ModelCompletionCancelledError();
const payload = {
model: `${request.model.provider}/${request.model.model}`,
...(request.model.apiKey === undefined ? {} : { api_key: request.model.apiKey }),
messages: request.messages.map((message) => ({ ...message })),
...(request.model.endpoint?.baseUrl === undefined
? {}
: { api_base: request.model.endpoint.baseUrl }),
...(request.model.endpoint?.apiVersion === undefined
? {}
: { api_version: request.model.endpoint.apiVersion }),
...(request.model.disableThinking === true ? { disable_thinking: true } : {}),
};
return await new Promise<string>((resolve, reject) => {
const child = spawn(
this.options.pythonExecutable,
["-m", this.options.helperModule ?? "tht.internal.litellm_completion"],
{
cwd: this.options.cwd,
stdio: ["pipe", "pipe", "pipe"],
},
);
let stdout = "";
let settled = false;
let timeout: ReturnType<typeof setTimeout> | undefined;
let killFallback: ReturnType<typeof setTimeout> | undefined;
let terminatingWith: Error | undefined;
const cleanup = () => {
if (timeout) clearTimeout(timeout);
if (killFallback) clearTimeout(killFallback);
request.signal.removeEventListener("abort", cancel);
};
const fail = (error: Error = new ModelCompletionProviderError()) => {
if (settled) return;
settled = true;
cleanup();
reject(error);
};
const terminate = (error: Error) => {
if (settled || terminatingWith) return;
terminatingWith = error;
if (timeout) clearTimeout(timeout);
try {
child.kill("SIGTERM");
} catch {
fail(error);
return;
}
killFallback = setTimeout(() => {
if (settled || child.exitCode !== null || child.signalCode !== null) return;
try {
child.kill("SIGKILL");
} catch {
fail(error);
}
}, this.options.terminationGraceMs ?? 250);
};
const cancel = () => {
terminate(new ModelCompletionCancelledError());
};
timeout = setTimeout(() => {
terminate(new ModelCompletionProviderError());
}, this.options.timeoutMs ?? 120_000);
request.signal.addEventListener("abort", cancel, { once: true });
if (request.signal.aborted) cancel();
child.stdout.setEncoding("utf8");
child.stdout.on("data", (chunk: string) => {
if (terminatingWith) return;
stdout += chunk;
if (Buffer.byteLength(stdout, "utf8") > MAX_HELPER_OUTPUT_BYTES) {
terminate(new ModelCompletionProviderError());
}
});
// Helper and provider diagnostics are deliberately not copied into application logs.
child.stderr.resume();
child.once("error", () => fail(terminatingWith ?? new ModelCompletionProviderError()));
child.once("close", (code) => {
if (settled) return;
if (terminatingWith) return fail(terminatingWith);
try {
if (code !== 0) return fail();
const output = helperOutputSchema.parse(JSON.parse(stdout));
if (!output.ok) return fail();
settled = true;
cleanup();
resolve(output.content);
} catch {
fail();
}
});
child.stdin.once("error", () => {
if (terminatingWith) return;
terminate(new ModelCompletionProviderError());
});
child.stdin.end(JSON.stringify(payload));
});
}
}
+17 -5
View File
@@ -1,22 +1,34 @@
import { CatalogOperationInProgressError } from "./types.js";
/** Serializes connection tests, synchronization, and metadata cleanup for each catalog database. */
/** Serializes mutable catalog operations for each Workspace Database. */
export class CatalogOperationCoordinator {
private readonly active = new Set<string>();
private readonly active = new Map<
string,
{ token: symbol; owner: "catalog_operation" | "description_generation" }
>();
reserve(databaseId: string): () => void {
reserve(
databaseId: string,
owner: "catalog_operation" | "description_generation" = "catalog_operation",
): () => void {
if (this.active.has(databaseId)) {
throw new CatalogOperationInProgressError("A database operation is already in progress");
}
this.active.add(databaseId);
const token = Symbol(databaseId);
this.active.set(databaseId, { token, owner });
let released = false;
return () => {
if (released) return;
released = true;
this.active.delete(databaseId);
if (this.active.get(databaseId)?.token === token) this.active.delete(databaseId);
};
}
/** Administrative recovery for a reservation whose owning local operation is no longer live. */
releaseStale(databaseId: string, owner: "description_generation"): void {
if (this.active.get(databaseId)?.owner === owner) this.active.delete(databaseId);
}
async run<T>(databaseId: string, operation: () => Promise<T>): Promise<T> {
const release = this.reserve(databaseId);
try {
+5 -1
View File
@@ -180,13 +180,17 @@ export class ConcreteCatalogPostgresAccess implements CatalogPostgresAccess {
materialized.release();
};
const abort = () => { void close(); };
signal.addEventListener("abort", abort, { once: true });
try {
if (signal.aborted) throw new CatalogConnectorError("PostgreSQL connector aborted");
const passwordFile = materialized.files.get(CATALOG_SECRET_IDS.password);
if (!passwordFile) throw new CatalogConnectorError("Database password is not configured");
const password = await readFile(passwordFile, "utf8");
if (signal.aborted) throw new CatalogConnectorError("PostgreSQL connector aborted");
const tlsCaFile = materialized.files.get(CATALOG_SECRET_IDS.tlsCa);
const tlsCa = tlsCaFile ? await readFile(tlsCaFile, "utf8") : undefined;
if (signal.aborted) throw new CatalogConnectorError("PostgreSQL connector aborted");
let host: string;
let port: number;
@@ -243,8 +247,8 @@ export class ConcreteCatalogPostgresAccess implements CatalogPostgresAccess {
connectionTimeoutMillis: this.connectTimeoutMs,
...(stream ? { stream: () => stream } : {}),
});
signal.addEventListener("abort", abort, { once: true });
await client.connect();
if (signal.aborted) throw new CatalogConnectorError("PostgreSQL connector aborted");
return {
query: async (sql, values) => await client!.query(sql, [...values]),
end: close,
+260
View File
@@ -15,7 +15,10 @@ import {
CatalogConflictError,
CatalogConnectorError,
CatalogUnavailableError,
DescriptionGenerationRunActiveError,
type CatalogColumn,
type CatalogDescriptionConsolidationCounts,
type CatalogDescriptionTarget,
type CatalogDatabaseMetadataDeleteTarget,
type CatalogMetadataDeleteCounts,
type CatalogRelationship,
@@ -31,6 +34,10 @@ import {
type DatabaseBinding,
type DatabaseConfigurationInput,
type DatabaseTestResult,
type DescriptionGenerationEvent,
type DescriptionGenerationRun,
type DescriptionGenerationRunUpdate,
type DescriptionGenerationScope,
type ObservedCatalogTable,
type ObservedSchemaSnapshot,
type TableSyncRepositoryResult,
@@ -131,6 +138,33 @@ interface CatalogRelationshipColumnTable {
targetColumnId: string;
}
interface DescriptionGenerationRunTable {
id: string;
databaseId: string;
scope: DescriptionGenerationScope;
modelId: string;
language: DescriptionGenerationRun["language"];
status: DescriptionGenerationRun["status"];
total: number;
processed: number;
generated: number;
nonGeneratable: number;
failed: number;
createdAt: Timestamp;
startedAt: Timestamp | null;
updatedAt: Timestamp;
finishedAt: Timestamp | null;
errorSummary: string | null;
}
interface DescriptionGenerationEventTable {
runId: string;
sequence: number;
level: DescriptionGenerationEvent["level"];
message: string;
createdAt: Timestamp;
}
interface CatalogSyncRunTable {
id: string;
databaseId: string;
@@ -176,6 +210,8 @@ export interface CatalogDatabase {
catalogColumns: CatalogColumnTable;
catalogRelationships: CatalogRelationshipTable;
catalogRelationshipColumns: CatalogRelationshipColumnTable;
descriptionGenerationRuns: DescriptionGenerationRunTable;
descriptionGenerationEvents: DescriptionGenerationEventTable;
catalogSyncRuns: CatalogSyncRunTable;
catalogSyncEvents: CatalogSyncEventTable;
}
@@ -281,6 +317,25 @@ function serializeSyncEvent(row: Selectable<CatalogSyncEventTable>): CatalogSync
return { ...row, id: Number(row.id), data: row.data ?? {}, createdAt: new Date(row.createdAt).toISOString() };
}
function serializeDescriptionGenerationRun(
row: Selectable<DescriptionGenerationRunTable>,
): DescriptionGenerationRun {
const stamp = (value: Date | string | null) => value === null ? null : new Date(value).toISOString();
return {
...row,
createdAt: new Date(row.createdAt).toISOString(),
startedAt: stamp(row.startedAt),
updatedAt: new Date(row.updatedAt).toISOString(),
finishedAt: stamp(row.finishedAt),
};
}
function serializeDescriptionGenerationEvent(
row: Selectable<DescriptionGenerationEventTable>,
): DescriptionGenerationEvent {
return { ...row, createdAt: new Date(row.createdAt).toISOString() };
}
function bindingValues(databaseId: string, binding: DatabaseBinding) {
return {
databaseId,
@@ -520,6 +575,202 @@ export class KyselyCatalogRepository implements CatalogRepository {
return row ? await this.getColumn(databaseId, tableId, row.id) : undefined;
}
async consolidateGeneratedDescriptions(
databaseId: string,
target: CatalogDescriptionTarget,
targetIds: readonly string[],
): Promise<CatalogDescriptionConsolidationCounts | undefined> {
const selectedTargetIds = [...new Set(targetIds)];
if (selectedTargetIds.length === 0) return undefined;
return await this.db.transaction().execute(async (trx) => {
const database = await trx.selectFrom("workspaceDatabases").select("id")
.where("id", "=", databaseId).forUpdate().executeTakeFirst();
if (!database) return undefined;
if (target === "tables") {
const rows = await trx.selectFrom("catalogTables")
.select(["id", "generatedDescription"])
.where("databaseId", "=", databaseId)
.where("id", "in", selectedTargetIds)
.orderBy("id")
.forUpdate()
.execute();
if (rows.length !== selectedTargetIds.length) return undefined;
const copiedIds = rows
.filter((row) => Boolean(row.generatedDescription?.trim()))
.map((row) => row.id);
if (copiedIds.length > 0) {
await trx.updateTable("catalogTables").set({
description: sql`generated_description`,
version: sql`version + 1`,
updatedAt: sql`now()`,
}).where("id", "in", copiedIds).execute();
}
return {
copied: copiedIds.length,
skipped: selectedTargetIds.length - copiedIds.length,
};
}
const tableRows = await trx.selectFrom("catalogTables").select("id")
.where("databaseId", "=", databaseId).execute();
const rows = tableRows.length === 0 ? [] : await trx.selectFrom("catalogColumns")
.select(["id", "generatedDescription"])
.where("tableId", "in", tableRows.map((table) => table.id))
.where("id", "in", selectedTargetIds)
.orderBy("id")
.forUpdate()
.execute();
if (rows.length !== selectedTargetIds.length) return undefined;
const copiedIds = rows
.filter((row) => Boolean(row.generatedDescription?.trim()))
.map((row) => row.id);
if (copiedIds.length > 0) {
await trx.updateTable("catalogColumns").set({
description: sql`generated_description`,
version: sql`version + 1`,
updatedAt: sql`now()`,
}).where("id", "in", copiedIds).execute();
}
return {
copied: copiedIds.length,
skipped: selectedTargetIds.length - copiedIds.length,
};
});
}
async createDescriptionGenerationRun(
databaseId: string,
scope: DescriptionGenerationScope,
modelId: string,
language: DescriptionGenerationRun["language"],
total: number,
): Promise<DescriptionGenerationRun> {
try {
const row = await this.db.insertInto("descriptionGenerationRuns").values({
id: randomUUID(),
databaseId,
scope,
modelId,
language,
status: "queued",
total,
processed: 0,
generated: 0,
nonGeneratable: 0,
failed: 0,
startedAt: null,
finishedAt: null,
errorSummary: null,
}).returningAll().executeTakeFirstOrThrow();
return serializeDescriptionGenerationRun(row);
} catch (error: any) {
if (error?.code === "23505"
&& error?.constraint === "description_generation_runs_one_active") {
throw new DescriptionGenerationRunActiveError(
"A description generation run is already active",
);
}
throw error;
}
}
async getDescriptionGenerationRun(
runId: string,
): Promise<DescriptionGenerationRun | undefined> {
const row = await this.db.selectFrom("descriptionGenerationRuns")
.selectAll()
.where("id", "=", runId)
.executeTakeFirst();
return row ? serializeDescriptionGenerationRun(row) : undefined;
}
async listDescriptionGenerationRuns(limit = 50): Promise<DescriptionGenerationRun[]> {
const rows = await this.db.selectFrom("descriptionGenerationRuns")
.selectAll()
.orderBy("createdAt", "desc")
.orderBy("id", "desc")
.limit(limit)
.execute();
return rows.map(serializeDescriptionGenerationRun);
}
async getActiveDescriptionGenerationRun(): Promise<DescriptionGenerationRun | undefined> {
const row = await this.db.selectFrom("descriptionGenerationRuns")
.selectAll()
.where("status", "in", ["queued", "running"])
.orderBy("createdAt", "desc")
.executeTakeFirst();
return row ? serializeDescriptionGenerationRun(row) : undefined;
}
async interruptActiveDescriptionGenerationRuns(
errorSummary: string,
): Promise<DescriptionGenerationRun[]> {
const rows = await this.db.updateTable("descriptionGenerationRuns")
.set({
status: "interrupted",
finishedAt: sql`now()`,
updatedAt: sql`now()`,
errorSummary,
})
.where("status", "in", ["queued", "running"])
.returningAll()
.execute();
return rows.map(serializeDescriptionGenerationRun);
}
async updateDescriptionGenerationRun(
runId: string,
update: DescriptionGenerationRunUpdate,
): Promise<DescriptionGenerationRun | undefined> {
const values: any = { ...update, updatedAt: sql`now()` };
const row = await this.db.updateTable("descriptionGenerationRuns")
.set(values)
.where("id", "=", runId)
.returningAll()
.executeTakeFirst();
return row ? serializeDescriptionGenerationRun(row) : undefined;
}
async appendDescriptionGenerationEvent(
runId: string,
level: DescriptionGenerationEvent["level"],
message: string,
): Promise<DescriptionGenerationEvent> {
return await this.db.transaction().execute(async (trx) => {
const run = await trx.selectFrom("descriptionGenerationRuns")
.select("id")
.where("id", "=", runId)
.forUpdate()
.executeTakeFirst();
if (!run) throw new CatalogConflictError("Description Generation Run does not exist");
const current = await trx.selectFrom("descriptionGenerationEvents")
.select(sql<number>`coalesce(max(sequence), 0)::int`.as("sequence"))
.where("runId", "=", runId)
.executeTakeFirst();
const row = await trx.insertInto("descriptionGenerationEvents").values({
runId,
sequence: Number(current?.sequence ?? 0) + 1,
level,
message,
}).returningAll().executeTakeFirstOrThrow();
return serializeDescriptionGenerationEvent(row);
});
}
async listDescriptionGenerationEvents(
runId: string,
afterSequence = 0,
): Promise<DescriptionGenerationEvent[]> {
const rows = await this.db.selectFrom("descriptionGenerationEvents")
.selectAll()
.where("runId", "=", runId)
.where("sequence", ">", afterSequence)
.orderBy("sequence")
.execute();
return rows.map(serializeDescriptionGenerationEvent);
}
async listRelationships(databaseId: string): Promise<CatalogRelationship[]> {
const rows = await this.db.selectFrom("catalogRelationships as relationship")
.innerJoin("catalogTables as sourceTable", "sourceTable.id", "relationship.sourceTableId")
@@ -1102,6 +1353,15 @@ export class UnavailableCatalogRepository implements CatalogRepository {
async listColumns(): Promise<CatalogColumn[]> { return this.fail(); }
async getColumn(): Promise<CatalogColumn | undefined> { return this.fail(); }
async updateColumnMetadata(): Promise<CatalogColumn | undefined> { return this.fail(); }
async consolidateGeneratedDescriptions(): Promise<CatalogDescriptionConsolidationCounts | undefined> { return this.fail(); }
async createDescriptionGenerationRun(): Promise<DescriptionGenerationRun> { return this.fail(); }
async getDescriptionGenerationRun(): Promise<DescriptionGenerationRun | undefined> { return this.fail(); }
async listDescriptionGenerationRuns(): Promise<DescriptionGenerationRun[]> { return this.fail(); }
async getActiveDescriptionGenerationRun(): Promise<DescriptionGenerationRun | undefined> { return this.fail(); }
async interruptActiveDescriptionGenerationRuns(): Promise<DescriptionGenerationRun[]> { return this.fail(); }
async updateDescriptionGenerationRun(): Promise<DescriptionGenerationRun | undefined> { return this.fail(); }
async appendDescriptionGenerationEvent(): Promise<DescriptionGenerationEvent> { return this.fail(); }
async listDescriptionGenerationEvents(): Promise<DescriptionGenerationEvent[]> { return this.fail(); }
async listRelationships(): Promise<CatalogRelationship[]> { return this.fail(); }
async deleteDatabaseMetadata(): Promise<CatalogMetadataDeleteCounts | undefined> { return this.fail(); }
async deleteTableMetadata(): Promise<CatalogMetadataDeleteCounts | undefined> { return this.fail(); }
+90
View File
@@ -135,6 +135,7 @@ export interface CatalogRelationship {
export type CatalogDatabaseMetadataDeleteTarget = "tables" | "relationships";
export type CatalogTableMetadataDeleteTarget = "columns" | "relationships";
export type CatalogDescriptionTarget = "tables" | "columns";
export interface CatalogMetadataDeleteCounts {
tables: number;
@@ -142,6 +143,63 @@ export interface CatalogMetadataDeleteCounts {
relationships: number;
}
export interface CatalogDescriptionConsolidationCounts {
copied: number;
skipped: number;
}
export type DescriptionGenerationScope =
| "selected_columns"
| "selected_tables"
| "all"
| "missing";
export type DescriptionGenerationStatus =
| "queued"
| "running"
| "completed"
| "completed_with_errors"
| "cancelled"
| "failed"
| "interrupted";
export interface DescriptionGenerationRun {
id: string;
databaseId: string;
scope: DescriptionGenerationScope;
modelId: string;
language: "en" | "it";
status: DescriptionGenerationStatus;
total: number;
processed: number;
generated: number;
nonGeneratable: number;
failed: number;
createdAt: string;
startedAt: string | null;
updatedAt: string;
finishedAt: string | null;
errorSummary: string | null;
}
export interface DescriptionGenerationRunUpdate {
status?: DescriptionGenerationStatus;
processed?: number;
generated?: number;
nonGeneratable?: number;
failed?: number;
startedAt?: string | null;
finishedAt?: string | null;
errorSummary?: string | null;
}
export interface DescriptionGenerationEvent {
runId: string;
sequence: number;
level: "info" | "warning" | "error";
message: string;
createdAt: string;
}
export interface ObservedRelationshipColumn {
position: number;
sourceColumnName: string;
@@ -292,6 +350,37 @@ export interface CatalogRepository {
description: string | null,
generatedDescription: string | null,
): Promise<CatalogColumn | undefined>;
consolidateGeneratedDescriptions(
databaseId: string,
target: CatalogDescriptionTarget,
targetIds: readonly string[],
): Promise<CatalogDescriptionConsolidationCounts | undefined>;
createDescriptionGenerationRun(
databaseId: string,
scope: DescriptionGenerationScope,
modelId: string,
language: DescriptionGenerationRun["language"],
total: number,
): Promise<DescriptionGenerationRun>;
getDescriptionGenerationRun(runId: string): Promise<DescriptionGenerationRun | undefined>;
listDescriptionGenerationRuns(limit?: number): Promise<DescriptionGenerationRun[]>;
getActiveDescriptionGenerationRun(): Promise<DescriptionGenerationRun | undefined>;
interruptActiveDescriptionGenerationRuns(
errorSummary: string,
): Promise<DescriptionGenerationRun[]>;
updateDescriptionGenerationRun(
runId: string,
update: DescriptionGenerationRunUpdate,
): Promise<DescriptionGenerationRun | undefined>;
appendDescriptionGenerationEvent(
runId: string,
level: DescriptionGenerationEvent["level"],
message: string,
): Promise<DescriptionGenerationEvent>;
listDescriptionGenerationEvents(
runId: string,
afterSequence?: number,
): Promise<DescriptionGenerationEvent[]>;
listRelationships(databaseId: string): Promise<CatalogRelationship[]>;
deleteDatabaseMetadata(
databaseIds: readonly string[],
@@ -347,6 +436,7 @@ export interface CatalogRepository {
}
export class CatalogConflictError extends Error {}
export class DescriptionGenerationRunActiveError extends CatalogConflictError {}
export class CatalogUnavailableError extends Error {}
export class CatalogOperationInProgressError extends Error {}
export class CatalogConnectorError extends Error {}