fix: page call_logs cleanup queries to avoid startup OOM on large DBs (#5618) (#5675)

This commit is contained in:
Diego Rodrigues de Sa e Souza
2026-06-30 13:16:17 -03:00
committed by GitHub
parent 20b19bff87
commit 96a33053bc
4 changed files with 170 additions and 15 deletions

View File

@@ -9,6 +9,7 @@ import fs from "node:fs";
import path from "node:path";
import type { RequestPipelinePayloads } from "@omniroute/open-sse/utils/requestLogger.ts";
import { getDbInstance } from "../db/core";
import { collectReferencedArtifacts, selectCallLogIdsBefore } from "./callLogsBoundedQueries";
import { getRequestDetailLogByCallLogId } from "../db/detailedLogs";
import { shouldPersistToDisk } from "./migrations";
import {
@@ -415,14 +416,8 @@ function clearArtifactReference(relativePath: string, nextState: CallLogDetailSt
}
function listReferencedArtifacts() {
const db = getDbInstance();
const rows = db
.prepare("SELECT artifact_relpath FROM call_logs WHERE artifact_relpath IS NOT NULL")
.all() as Array<{ artifact_relpath: string | null }>;
return new Set(
rows.map((row) => row.artifact_relpath).filter((value): value is string => Boolean(value))
);
// #5618: paged to avoid an unbounded `.all()` OOM on large call_logs tables.
return collectReferencedArtifacts();
}
// #5217: SQLite caps a statement at SQLITE_MAX_VARIABLE_NUMBER bound params
@@ -510,13 +505,18 @@ export function cleanupOverflowCallLogFiles(baseDir = CALL_LOGS_DIR, maxEntries?
}
export function deleteCallLogsBefore(cutoff: string): DeleteResult {
const db = getDbInstance();
const ids = db
.prepare("SELECT id FROM call_logs WHERE timestamp < ? ORDER BY timestamp ASC")
.all(cutoff)
.map((row) => String((row as { id: string }).id));
return deleteCallLogRowsByIds(ids);
// #5618: page the id selection so a large backlog never loads in one `.all()`.
let deletedRows = 0;
let deletedArtifacts = 0;
for (;;) {
const ids = selectCallLogIdsBefore(cutoff);
if (ids.length === 0) break;
const result = deleteCallLogRowsByIds(ids);
deletedRows += result.deletedRows;
deletedArtifacts += result.deletedArtifacts;
if (result.deletedRows === 0) break;
}
return { deletedRows, deletedArtifacts };
}
export function trimCallLogsToMaxRows(maxRows = getCallLogsTableMaxRows()) {

View File

@@ -0,0 +1,42 @@
import { getDbInstance } from "../db/core";
// #5618 — node:sqlite's StatementSync.all() materializes the ENTIRE result set as
// JS objects at once. On a large storage.sqlite (~170 MB+) an unbounded
// `SELECT … FROM call_logs` over the whole table blows the V8 heap during the
// startup cleanup pass (`cleanupExpiredLogs` → `rotateCallLogs`), crashing the
// daemon before it binds. These helpers page through call_logs in bounded chunks
// so peak memory stays flat regardless of table size.
const CALL_LOG_QUERY_PAGE = 5000;
/**
* Collect every non-null `artifact_relpath` referenced by call_logs, paging with
* LIMIT/OFFSET so a huge table never loads into memory in one `.all()`.
*/
export function collectReferencedArtifacts(): Set<string> {
const db = getDbInstance();
const referenced = new Set<string>();
const stmt = db.prepare(
"SELECT artifact_relpath FROM call_logs WHERE artifact_relpath IS NOT NULL LIMIT ? OFFSET ?"
);
for (let offset = 0; ; offset += CALL_LOG_QUERY_PAGE) {
const rows = stmt.all(CALL_LOG_QUERY_PAGE, offset) as Array<{ artifact_relpath: string | null }>;
for (const row of rows) {
if (typeof row.artifact_relpath === "string") referenced.add(row.artifact_relpath);
}
if (rows.length < CALL_LOG_QUERY_PAGE) break;
}
return referenced;
}
/**
* Select one bounded page of call_log ids older than `cutoff` (oldest first).
* Callers loop until it returns an empty page, deleting each batch, so the id
* list never grows to the full retention backlog at once.
*/
export function selectCallLogIdsBefore(cutoff: string, limit = CALL_LOG_QUERY_PAGE): string[] {
const db = getDbInstance();
const rows = db
.prepare("SELECT id FROM call_logs WHERE timestamp < ? ORDER BY timestamp ASC LIMIT ?")
.all(cutoff, limit) as Array<{ id: string }>;
return rows.map((row) => String(row.id));
}