Skip to content

Garry's Opinionated OpenClaw/Hermes Agent Brain

TypeScript

fix: rows-behind primes only metafile blobs no larger than the head it keeps, so a batch never loads large blobs whole

garrytanffcd596009d5
diff --git a/src/core/persistence/rows-behind.ts b/src/core/persistence/rows-behind.ts
index f586aaa5df..f524a0ee33 100644
--- a/src/core/persistence/rows-behind.ts
+++ b/src/core/persistence/rows-behind.ts
@@ -39,6 +39,7 @@ import { prepareRepairPublication, repairScreenConfig } from './file-repair.ts';
import { SYNC_READ_BOUND } from './sync-blobs.ts';
import { heldGitPaths, readSyncImportProvenances } from './sync-holds.ts';
import { loadSyncEligibility } from './sync-discovery.ts';
+import { METAFILE_HEAD_BYTES } from '../sync.ts';
import { isSourceDbOnlySlug } from './source-storage.ts';
export const ROWS_BEHIND_BATCH = 256;
@@ -156,7 +157,7 @@ export async function scanRowsBehind(engine: BrainEngine, sourceId: string,
if (blob.mode !== '100644' && blob.mode !== '100755') { skip(key.id, 'not_regular_file'); continue; }
listed.push({ key, rel, blob });
}
- metafileHeads.prime(listed.map(row => ({ path: gitPath(row.rel), oid: row.blob.oid, size: row.blob.size })));
+ metafileHeads.prime(listed.filter(row => row.blob.size <= METAFILE_HEAD_BYTES).map(row => ({ path: gitPath(row.rel), oid: row.blob.oid, size: row.blob.size })));
const rows = listed.filter(row => {
if (eligible(gitPath(row.rel))) return true;
skip(row.key.id, 'not_eligible');
diff --git a/test/rows-behind-checkpoint-8071.test.ts b/test/rows-behind-checkpoint-8071.test.ts
index 0dceca697f..9a965da5c7 100644
--- a/test/rows-behind-checkpoint-8071.test.ts
+++ b/test/rows-behind-checkpoint-8071.test.ts
@@ -301,16 +301,19 @@ for (const backend of testBackends()) describe(`[${backend}] #8071: rows behind
expect(await doctor(engine)).toMatchObject({ name: CHECK, status: 'ok', readiness_state: 'unknown' });
}, { databaseUrl, setup }), 240_000);
- test('nested typed index pages are screened with one batched head read per batch, not a Git read per page', () => managedBrain(async ({ engine, root }) => {
+ test('nested typed index pages are screened with one batched head read per batch, not a Git read per page; a large one is read alone', () => managedBrain(async ({ engine, root }) => {
for (const dir of ['notes/s1', 'notes/s2', 'notes/s3']) { mkdirSync(join(root, dir), { recursive: true }); writeFileSync(join(root, dir, 'index.md'), note(dir, `The index of ${dir}.`)); }
+ mkdirSync(join(root, 'notes/big'), { recursive: true });
+ writeFileSync(join(root, 'notes/big/index.md'), note('notes/big', 'A long index line.\n'.repeat(5_000)));
commit(root, 'indexes');
expect((await performManagedSync(engine, OPTS)).status).toBe('first_sync');
const trace = join(tmpdir(), `gbrain-rows-behind-trace-${randomUUID()}.log`);
try {
const scan = await withEnv({ GIT_TRACE: trace }, () => scanRowsBehind(engine, 'default'));
- expect(scan).toMatchObject({ status: 'known', complete: true, behind: [], pages: 8 });
- // The eligibility screen reads the index pages' heads from the batch's checkpoint blobs, never one lookup per page.
- expect(readFileSync(trace, 'utf8').split('\n').filter(line => line.includes('cat-file') && line.includes('--batch-check'))).toEqual([]);
+ expect(scan).toMatchObject({ status: 'known', complete: true, behind: [], pages: 9 });
+ // The small index pages' heads come from one batched read of their checkpoint blobs, never a lookup per page. A blob
+ // larger than the head kept is not loaded into the batch's read; its head is read on its own (one lookup).
+ expect(readFileSync(trace, 'utf8').split('\n').filter(line => line.includes('cat-file') && line.includes('--batch-check'))).toHaveLength(1);
} finally { rmSync(trace, { force: true }); }
}, { databaseUrl, setup }), 240_000);