zib/Veskslop-t
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Serialize walks and backfill, split the ETA into start and run

Browse Source
This commit is contained in:
Fisto
2026-09-17 10:15:15 +02:00
parent 5b8e78940a
commit 7bbb45b79b
5 changed files with 212 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files
media-scrape/src/backfill.ts
+50 -2
View File
@@ -57,6 +57,12 @@ export interface BackfillJobStatus {
perMinute: number;
/** Estimated ms to clear the pending rows. Null when there is nothing to do. */
etaMs: number | null;
/** Ms until this job starts, waiting for the jobs ahead of it. Null when not scheduled. */
etaStartMs: number | null;
/** Ms this job needs once it starts. */
etaRunMs: number | null;
/** Start plus run: when this job is expected to finish. */
etaTotalMs: number | null;
}
export interface BackfillTotals {
@@ -68,6 +74,9 @@ export interface BackfillTotals {
bytes: number;
perMinute: number;
etaMs: number | null;
etaStartMs: number | null;
etaRunMs: number | null;
etaTotalMs: number | null;
byBackend: Record<BackendName, { pending: number; cached: number; failed: number; total: number }>;
}
@@ -107,6 +116,8 @@ const defaultSleep = (ms: number) => new Promise<void>(resolve => setTimeout(res
export class BackfillQueue {
private readonly jobs = new Map<string, Job>();
/** Start order: jobs claimed by the scheduler in this sequence, smallest first at start(). */
private readonly order: string[] = [];
private active: Promise<void> = Promise.resolve();
private limitedUntil = 0;
@@ -217,10 +228,32 @@ export class BackfillQueue {
startedAt: tracked?.startedAt ? new Date(tracked.startedAt).toISOString() : null,
finishedAt: tracked?.finishedAt ? new Date(tracked.finishedAt).toISOString() : null,
perMinute: Math.round(perMinute * 10) / 10,
etaMs
etaMs,
etaStartMs: null,
etaRunMs: null,
etaTotalMs: null
});
}
// Waiting times follow the execution order: each job starts after the ones ahead of it.
const executionOrder = [...this.order];
for (const job of jobs) {
if (!executionOrder.includes(job.chatId)) executionOrder.push(job.chatId);
}
let ahead = 0;
for (const chatId of executionOrder) {
const job = jobs.find(candidate => candidate.chatId === chatId);
if (!job) continue;
const scheduled = job.state === "running" || job.state === "queued" || job.state === "paused";
const runMs = job.pending > 0 ? (job.pending / Math.max(0.01, job.perMinute)) * 60_000 : 0;
job.etaStartMs = scheduled ? Math.round(ahead) : null;
job.etaRunMs = Math.round(runMs);
job.etaTotalMs = job.etaStartMs === null ? null : Math.round(ahead + runMs);
if (scheduled) ahead += runMs;
}
jobs.sort((a, b) => b.pending - a.pending);
// Totals span every scrape on disk, not just the jobs the list shows.
@@ -268,6 +301,9 @@ export class BackfillQueue {
bytes,
perMinute: Math.round(rate * 10) / 10,
etaMs: pending > 0 && rate > 0 ? Math.round((pending / rate) * 60_000) : null,
etaStartMs: pending > 0 ? 0 : null,
etaRunMs: pending > 0 && rate > 0 ? Math.round((pending / rate) * 60_000) : null,
etaTotalMs: pending > 0 && rate > 0 ? Math.round((pending / rate) * 60_000) : null,
byBackend
},
settings,
@@ -318,6 +354,7 @@ export class BackfillQueue {
tracked.startedAt = this.now();
tracked.finishedAt = null;
this.jobs.set(job.chatId, tracked);
if (!this.order.includes(job.chatId)) this.order.push(job.chatId);
this.options.log?.(`backfill: started ${job.chatId} (${job.pending} pending)`);
}
@@ -350,7 +387,11 @@ export class BackfillQueue {
private schedule(): void {
this.active = this.active.then(async () => {
for (;;) {
const job = [...this.jobs.values()].find(
const ordered = this.order
.map(chatId => this.jobs.get(chatId))
.filter((candidate): candidate is Job => candidate !== undefined);
const rest = [...this.jobs.values()].filter(candidate => !this.order.includes(candidate.chatId));
const job = [...ordered, ...rest].find(
candidate => candidate.state === "running" && !candidate.claimed
);
if (!job) return;
@@ -383,6 +424,13 @@ export class BackfillQueue {
this.limitedUntil = 0;
}
// A walk holds the media-traffic lock for its whole run. Wait it out without
// blocking a pause request.
while (!job.stop && this.engine.isWalking()) {
await this.sleep(5_000);
}
if (job.stop) break;
const settings = this.options.getSettings();
const result: PreviewBatchResult = await this.engine.previewBatch(
member.scrapeId,
media-scrape/src/engine.ts
+37
View File
@@ -116,6 +116,11 @@ export class ScrapeEngine {
private readonly pageTimes: number[] = [];
private rateLimitUntil = 0;
private processingPromise: Promise<void> | null = null;
/**
* One media-traffic lock: a walk holds it for its whole run, a backfill batch for the batch.
* Walks and backfill therefore never fetch at the same time.
*/
private trafficLock: Promise<void> = Promise.resolve();
constructor(private readonly options: ScrapeEngineOptions) {}
@@ -186,6 +191,32 @@ export class ScrapeEngine {
}
}
/** True while a walk is draining. The backfill queue waits on this before a batch. */
isWalking(): boolean {
return this.processingPromise !== null;
}
/** Resolves when every queued walk has finished. */
async whenIdle(): Promise<void> {
while (this.processingPromise) await this.processingPromise;
}
/**
* Take the media-traffic lock. Resolves with the release function. Walks take it for a whole
* run; backfill batches for one batch, so the two never overlap and the pacing settings hold
* for the total traffic, not per action.
*/
async acquireTraffic(): Promise<() => void> {
let release!: () => void;
const next = new Promise<void>(resolve => {
release = resolve;
});
const previous = this.trafficLock;
this.trafficLock = previous.then(() => next);
await previous;
return release;
}
get(id: string): ScrapeMeta | null {
const active = this.active.get(id);
if (active) return { ...active };
@@ -248,12 +279,14 @@ export class ScrapeEngine {
const dir = join(this.options.rootDir, scrapeId);
const store = new ScrapeStore(dir);
const releaseTraffic = await this.acquireTraffic();
const before = meta.previewed ?? 0;
try {
await this.runPreviews(store, dir, meta);
} finally {
store.close();
releaseTraffic();
}
return (meta.previewed ?? 0) - before;
@@ -283,6 +316,7 @@ export class ScrapeEngine {
const pageSize = this.options.pageSize ?? 100;
const perCycle = this.options.download?.previewBatchSize ?? 25;
let pagesSincePreview = 0;
const releaseTraffic = await this.acquireTraffic();
try {
if (this.stopRequested.has(meta.id)) {
@@ -388,6 +422,7 @@ export class ScrapeEngine {
this.persist(dir, meta);
store.close();
this.active.set(meta.id, meta);
releaseTraffic();
}
}
@@ -514,11 +549,13 @@ export class ScrapeEngine {
const meta = this.active.get(scrapeId) ?? readScrapeMeta(dir);
if (!meta) return { cached: 0, skipped: 0, failed: 0, ratelimited: 0, remaining: 0, stopped: false };
const releaseTraffic = await this.acquireTraffic();
const store = new ScrapeStore(dir);
try {
return await this.previewRows(store, dir, meta, Math.max(1, limit), this.previewOptions(meta));
} finally {
store.close();
releaseTraffic();
}
}
media-scrape/test/run.ts
+97 -8
View File
@@ -95,6 +95,26 @@ class FakeBackend implements ScrapeBackend {
}
}
/** A backend whose first page waits for `release()`, so a test can stop the walk mid-page. */
function gatedBackend(page: FetchPage): { backend: ScrapeBackend; release: () => void } {
let release!: () => void;
const gate = new Promise<void>(resolve => (release = resolve));
let first = true;
const backend: ScrapeBackend = {
name: "puppet",
async fetchPage() {
if (first) {
first = false;
await gate;
}
return page;
}
};
return { backend, release };
}
function engineFor(root: string, backend: ScrapeBackend, extra: Record<string, unknown> = {}) {
return new ScrapeEngine({
rootDir: root,
@@ -276,14 +296,15 @@ void (async () => {
});
await test("engine stop pauses after the current page", async () => {
const backend = new FakeBackend([
{ messages: [message("300"), message("301")] },
{ messages: [message("299")] }
]);
const { backend, release } = gatedBackend({ messages: [message("300"), message("301")] });
const engine = engineFor(root, backend);
const started = await engine.start({ channelId: "126", guildId: null, mode: "links", backend });
// Stop while the first page is in flight; the page lands, then the walk pauses.
await new Promise(resolve => setImmediate(resolve));
engine.stop(started.id);
release();
await engine.processQueue();
const meta = engine.get(started.id);
@@ -291,6 +312,20 @@ void (async () => {
assert.equal(meta?.pages, 1);
});
await test("an immediate stop fetches nothing", async () => {
const backend = new FakeBackend([{ messages: [message("310")] }]);
const engine = engineFor(root, backend);
const started = await engine.start({ channelId: "132", guildId: null, mode: "links", backend });
engine.stop(started.id);
await engine.processQueue();
const meta = engine.get(started.id);
assert.equal(meta?.state, "paused");
assert.equal(meta?.pages, 0);
assert.equal(backend.calls.length, 0);
});
await test("puppet backend passes before and limit to the core", async () => {
const calls: Array<Record<string, unknown>> = [];
const backend = new PuppetBackend(async (channelId, limit, options) => {
@@ -1100,14 +1135,15 @@ void (async () => {
});
await test("engine resumes a paused scrape from its cursor", async () => {
const first = new FakeBackend([
{ messages: [message("500"), message("499")] },
{ messages: [message("498")] }
]);
const { backend: first, release } = gatedBackend({ messages: [message("500"), message("499")] });
const engine = engineFor(root, first);
const started = await engine.start({ channelId: "140", guildId: null, mode: "links", backend: first });
// Stop while the first page is in flight so the cursor is stored, then pauses.
await new Promise(resolve => setImmediate(resolve));
engine.stop(started.id);
release();
await engine.processQueue();
let meta = engine.get(started.id);
@@ -1345,17 +1381,70 @@ void (async () => {
sleep: async () => {}
});
const small = new FakeBackend([
{
messages: [
message("662", {
attachments: [attachment({ filename: "small.png", contentType: "image/png" })]
})
]
}
]);
await engine.start({ channelId: "203", guildId: null, mode: "links", backend: small });
await engine.processQueue();
const status = queue.list();
const job = status.jobs.find(candidate => candidate.chatId === "200");
const smallJob = status.jobs.find(candidate => candidate.chatId === "203");
assert.ok(job);
assert.ok(smallJob);
assert.equal(job.members.length, 2);
assert.equal(job.pending, 3);
assert.ok((job.etaMs ?? 0) > 0);
// Start everything: smallest first, so 203 runs and 200 waits behind it.
queue.start();
const scheduled = queue.list();
const firstJob = scheduled.jobs.find(candidate => candidate.chatId === "203");
const secondJob = scheduled.jobs.find(candidate => candidate.chatId === "200");
assert.ok(firstJob);
assert.ok(secondJob);
assert.equal(firstJob.etaStartMs, 0);
assert.ok((firstJob.etaRunMs ?? 0) > 0);
assert.equal(secondJob.etaTotalMs, (secondJob.etaStartMs ?? 0) + (secondJob.etaRunMs ?? 0));
assert.ok((secondJob.etaStartMs ?? 0) >= (firstJob.etaRunMs ?? 0));
assert.ok(status.totals.byBackend.puppet.pending >= 3);
assert.equal(status.totals.byBackend.slopbot.pending, 0);
assert.ok(status.totals.total >= 3);
assert.equal(job.members.some(member => member.scrapeId === first.id), true);
assert.equal(job.members.some(member => member.scrapeId === second.id), true);
queue.pause();
});
await test("media traffic serializes walks and backfill batches", async () => {
const { backend, release } = gatedBackend({ messages: [message("320")] });
const engine = engineFor(root, backend);
await engine.start({ channelId: "133", guildId: null, mode: "links", backend });
await new Promise(resolve => setImmediate(resolve));
assert.equal(engine.isWalking(), true);
const order: string[] = [];
const queued = engine.acquireTraffic().then(releaseSecond => {
order.push("batch");
releaseSecond();
});
// While the walk holds the lock, the batch cannot start.
await new Promise(resolve => setTimeout(resolve, 5));
assert.deepEqual(order, []);
release();
await engine.processQueue();
await queued;
assert.deepEqual(order, ["batch"]);
assert.equal(engine.isWalking(), false);
});
await test("backfill queue processes batches and finishes", async () => {
poc/public/media.css
+5
View File
@@ -745,3 +745,8 @@
background: color-mix(in srgb, var(--color-neutral-400) 18%, transparent);
color: var(--color-neutral-300);
}
.vk-bot-eta {
cursor: help;
text-decoration: underline dotted color-mix(in srgb, var(--color-neutral-400) 60%, transparent);
}
poc/public/media.js
+23 -7
View File
@@ -552,6 +552,15 @@
const bot = backend.slopbot ?? { pending: 0, cached: 0, failed: 0, total: 0 };
const puppet = backend.puppet ?? { pending: 0, cached: 0, failed: 0, total: 0 };
const etaLine = el(
"div",
"vk-bot-summary-line vk-bot-eta",
`ETA ${formatEta(totals.etaTotalMs ?? totals.etaMs)} at ${totals.perMinute ?? 0} rows/min ` +
`(settings estimate ${status.estimatedPerMinute ?? 0}/min)`
);
etaLine.title =
`first job starts now \u00b7 queue drains in ${formatEta(totals.etaRunMs ?? totals.etaMs)}`;
backfillSummary.replaceChildren(
el(
"div",
@@ -560,12 +569,7 @@
`${totals.failed ?? 0} failed \u00b7 ${totals.downloaded ?? 0} downloaded \u00b7 ` +
`${formatBytes(totals.bytes ?? 0)} stored`
),
el(
"div",
"vk-bot-summary-line",
`ETA ${formatEta(totals.etaMs)} at ${totals.perMinute ?? 0} rows/min ` +
`(settings estimate ${status.estimatedPerMinute ?? 0}/min)`
),
etaLine,
el(
"div",
"vk-bot-summary-line vk-dim",
@@ -590,8 +594,20 @@
const parts = [`${job.pending} pending`];
if (job.cached) parts.push(`${job.cached} cached`);
if (job.failed) parts.push(`${job.failed} failed`);
if (job.pending > 0) parts.push(`ETA ${formatEta(job.etaMs)}`);
head.append(el("span", "vk-bot-job-meta", parts.join(" \u00b7 ")));
if (job.pending > 0) {
const eta = el(
"span",
"vk-bot-job-meta vk-bot-eta",
`ETA ${formatEta(job.etaTotalMs ?? job.etaMs)}`
);
eta.title =
`starts in ${formatEta(job.etaStartMs ?? 0)} \u00b7 ` +
`runs for ${formatEta(job.etaRunMs ?? job.etaMs)}`;
head.append(eta);
}
row.append(head);
for (const member of job.members) {