Line data Source code
1 : use std::collections::{HashSet, VecDeque};
2 : use std::future::Future;
3 : use std::num::NonZeroUsize;
4 : use std::pin::Pin;
5 : use std::sync::atomic::{AtomicU64, Ordering};
6 : use std::sync::{Arc, Mutex};
7 : use std::time::{Duration, Instant};
8 :
9 : use anyhow::Context;
10 : use camino::Utf8PathBuf;
11 : use pageserver_api::key::Key;
12 : use pageserver_api::keyspace::KeySpaceAccum;
13 : use pageserver_api::models::{PagestreamGetPageRequest, PagestreamRequest};
14 : use pageserver_api::shard::TenantShardId;
15 : use rand::prelude::*;
16 : use tokio::task::JoinSet;
17 : use tokio_util::sync::CancellationToken;
18 : use tracing::info;
19 : use utils::id::TenantTimelineId;
20 : use utils::lsn::Lsn;
21 :
22 : use crate::util::tokio_thread_local_stats::AllThreadLocalStats;
23 : use crate::util::{request_stats, tokio_thread_local_stats};
24 :
25 : /// GetPage@LatestLSN, uniformly distributed across the compute-accessible keyspace.
26 : #[derive(clap::Parser)]
27 : pub(crate) struct Args {
28 : #[clap(long, default_value = "http://localhost:9898")]
29 0 : mgmt_api_endpoint: String,
30 : #[clap(long, default_value = "postgres://postgres@localhost:64000")]
31 0 : page_service_connstring: String,
32 : #[clap(long)]
33 : pageserver_jwt: Option<String>,
34 : #[clap(long, default_value = "1")]
35 0 : num_clients: NonZeroUsize,
36 : #[clap(long)]
37 : runtime: Option<humantime::Duration>,
38 : /// Each client sends requests at the given rate.
39 : ///
40 : /// If a request takes too long and we should be issuing a new request already,
41 : /// we skip that request and account it as `MISSED`.
42 : #[clap(long)]
43 : per_client_rate: Option<usize>,
44 : /// Probability for sending `latest=true` in the request (uniform distribution).
45 : #[clap(long, default_value = "1")]
46 0 : req_latest_probability: f64,
47 : #[clap(long)]
48 : limit_to_first_n_targets: Option<usize>,
49 : /// For large pageserver installations, enumerating the keyspace takes a lot of time.
50 : /// If specified, the specified path is used to maintain a cache of the keyspace enumeration result.
51 : /// The cache is tagged and auto-invalided by the tenant/timeline ids only.
52 : /// It doesn't get invalidated if the keyspace changes under the hood, e.g., due to new ingested data or compaction.
53 : #[clap(long)]
54 : keyspace_cache: Option<Utf8PathBuf>,
55 : /// Before starting the benchmark, live-reconfigure the pageserver to use the given
56 : /// [`pageserver_api::models::virtual_file::IoEngineKind`].
57 : #[clap(long)]
58 : set_io_engine: Option<pageserver_api::models::virtual_file::IoEngineKind>,
59 :
60 : /// Before starting the benchmark, live-reconfigure the pageserver to use specified io mode (buffered vs. direct).
61 : #[clap(long)]
62 : set_io_mode: Option<pageserver_api::models::virtual_file::IoMode>,
63 :
64 : /// Queue depth generated in each client.
65 : #[clap(long, default_value = "1")]
66 0 : queue_depth: NonZeroUsize,
67 :
68 0 : targets: Option<Vec<TenantTimelineId>>,
69 : }
70 :
71 : /// State shared by all clients
72 : #[derive(Debug)]
73 : struct SharedState {
74 : start_work_barrier: tokio::sync::Barrier,
75 : live_stats: LiveStats,
76 : }
77 :
78 : #[derive(Debug, Default)]
79 : struct LiveStats {
80 : completed_requests: AtomicU64,
81 : missed: AtomicU64,
82 : }
83 :
84 : impl LiveStats {
85 0 : fn request_done(&self) {
86 0 : self.completed_requests.fetch_add(1, Ordering::Relaxed);
87 0 : }
88 0 : fn missed(&self, n: u64) {
89 0 : self.missed.fetch_add(n, Ordering::Relaxed);
90 0 : }
91 : }
92 :
93 0 : #[derive(Clone, serde::Serialize, serde::Deserialize)]
94 : struct KeyRange {
95 : timeline: TenantTimelineId,
96 : timeline_lsn: Lsn,
97 : start: i128,
98 : end: i128,
99 : }
100 :
101 : impl KeyRange {
102 0 : fn len(&self) -> i128 {
103 0 : self.end - self.start
104 0 : }
105 : }
106 :
107 : #[derive(PartialEq, Eq, Hash, Copy, Clone)]
108 : struct WorkerId {
109 : timeline: TenantTimelineId,
110 : num_client: usize, // from 0..args.num_clients
111 : }
112 :
113 : #[derive(serde::Serialize)]
114 : struct Output {
115 : total: request_stats::Output,
116 : }
117 :
118 : tokio_thread_local_stats::declare!(STATS: request_stats::Stats);
119 :
120 0 : pub(crate) fn main(args: Args) -> anyhow::Result<()> {
121 0 : tokio_thread_local_stats::main!(STATS, move |thread_local_stats| {
122 0 : main_impl(args, thread_local_stats)
123 0 : })
124 0 : }
125 :
126 0 : async fn main_impl(
127 0 : args: Args,
128 0 : all_thread_local_stats: AllThreadLocalStats<request_stats::Stats>,
129 0 : ) -> anyhow::Result<()> {
130 0 : let args: &'static Args = Box::leak(Box::new(args));
131 0 :
132 0 : let mgmt_api_client = Arc::new(pageserver_client::mgmt_api::Client::new(
133 0 : reqwest::Client::new(), // TODO: support ssl_ca_file for https APIs in pagebench.
134 0 : args.mgmt_api_endpoint.clone(),
135 0 : args.pageserver_jwt.as_deref(),
136 0 : ));
137 :
138 0 : if let Some(engine_str) = &args.set_io_engine {
139 0 : mgmt_api_client.put_io_engine(engine_str).await?;
140 0 : }
141 :
142 0 : if let Some(mode) = &args.set_io_mode {
143 0 : mgmt_api_client.put_io_mode(mode).await?;
144 0 : }
145 :
146 : // discover targets
147 0 : let timelines: Vec<TenantTimelineId> = crate::util::cli::targets::discover(
148 0 : &mgmt_api_client,
149 0 : crate::util::cli::targets::Spec {
150 0 : limit_to_first_n_targets: args.limit_to_first_n_targets,
151 0 : targets: args.targets.clone(),
152 0 : },
153 0 : )
154 0 : .await?;
155 :
156 0 : #[derive(serde::Deserialize)]
157 : struct KeyspaceCacheDe {
158 : tag: Vec<TenantTimelineId>,
159 : data: Vec<KeyRange>,
160 : }
161 : #[derive(serde::Serialize)]
162 : struct KeyspaceCacheSer<'a> {
163 : tag: &'a [TenantTimelineId],
164 : data: &'a [KeyRange],
165 : }
166 0 : let cache = args
167 0 : .keyspace_cache
168 0 : .as_ref()
169 0 : .map(|keyspace_cache_file| {
170 0 : let contents = match std::fs::read(keyspace_cache_file) {
171 0 : Err(e) if e.kind() == std::io::ErrorKind::NotFound => {
172 0 : return anyhow::Ok(None);
173 : }
174 0 : x => x.context("read keyspace cache file")?,
175 : };
176 0 : let cache: KeyspaceCacheDe =
177 0 : serde_json::from_slice(&contents).context("deserialize cache file")?;
178 0 : let tag_ok = HashSet::<TenantTimelineId>::from_iter(cache.tag.into_iter())
179 0 : == HashSet::from_iter(timelines.iter().cloned());
180 0 : info!("keyspace cache file matches tag: {tag_ok}");
181 0 : anyhow::Ok(if tag_ok { Some(cache.data) } else { None })
182 0 : })
183 0 : .transpose()?
184 0 : .flatten();
185 0 : let all_ranges: Vec<KeyRange> = if let Some(cached) = cache {
186 0 : info!("using keyspace cache file");
187 0 : cached
188 : } else {
189 0 : let mut js = JoinSet::new();
190 0 : for timeline in &timelines {
191 0 : js.spawn({
192 0 : let mgmt_api_client = Arc::clone(&mgmt_api_client);
193 0 : let timeline = *timeline;
194 0 : async move {
195 0 : let partitioning = mgmt_api_client
196 0 : .keyspace(
197 0 : TenantShardId::unsharded(timeline.tenant_id),
198 0 : timeline.timeline_id,
199 0 : )
200 0 : .await?;
201 0 : let lsn = partitioning.at_lsn;
202 0 : let start = Instant::now();
203 0 : let mut filtered = KeySpaceAccum::new();
204 : // let's hope this is inlined and vectorized...
205 : // TODO: turn this loop into a is_rel_block_range() function.
206 0 : for r in partitioning.keys.ranges.iter() {
207 0 : let mut i = r.start;
208 0 : while i != r.end {
209 0 : if i.is_rel_block_key() {
210 0 : filtered.add_key(i);
211 0 : }
212 0 : i = i.next();
213 : }
214 : }
215 0 : let filtered = filtered.to_keyspace();
216 0 : let filter_duration = start.elapsed();
217 0 :
218 0 : anyhow::Ok((
219 0 : filter_duration,
220 0 : filtered.ranges.into_iter().map(move |r| KeyRange {
221 0 : timeline,
222 0 : timeline_lsn: lsn,
223 0 : start: r.start.to_i128(),
224 0 : end: r.end.to_i128(),
225 0 : }),
226 0 : ))
227 0 : }
228 0 : });
229 0 : }
230 0 : let mut total_filter_duration = Duration::from_secs(0);
231 0 : let mut all_ranges: Vec<KeyRange> = Vec::new();
232 0 : while let Some(res) = js.join_next().await {
233 0 : let (filter_duration, range) = res.unwrap().unwrap();
234 0 : all_ranges.extend(range);
235 0 : total_filter_duration += filter_duration;
236 0 : }
237 0 : info!("filter duration: {}", total_filter_duration.as_secs_f64());
238 0 : if let Some(cachefile) = args.keyspace_cache.as_ref() {
239 0 : let cache = KeyspaceCacheSer {
240 0 : tag: &timelines,
241 0 : data: &all_ranges,
242 0 : };
243 0 : let bytes = serde_json::to_vec(&cache).context("serialize keyspace for cache file")?;
244 0 : std::fs::write(cachefile, bytes).context("write keyspace cache file to disk")?;
245 0 : info!("successfully wrote keyspace cache file");
246 0 : }
247 0 : all_ranges
248 : };
249 :
250 0 : let num_live_stats_dump = 1;
251 0 : let num_work_sender_tasks = args.num_clients.get() * timelines.len();
252 0 : let num_main_impl = 1;
253 0 :
254 0 : let shared_state = Arc::new(SharedState {
255 0 : start_work_barrier: tokio::sync::Barrier::new(
256 0 : num_live_stats_dump + num_work_sender_tasks + num_main_impl,
257 0 : ),
258 0 : live_stats: LiveStats::default(),
259 0 : });
260 0 : let cancel = CancellationToken::new();
261 0 :
262 0 : let ss = shared_state.clone();
263 0 : tokio::spawn({
264 0 : async move {
265 0 : ss.start_work_barrier.wait().await;
266 : loop {
267 0 : let start = std::time::Instant::now();
268 0 : tokio::time::sleep(std::time::Duration::from_secs(1)).await;
269 0 : let stats = &ss.live_stats;
270 0 : let completed_requests = stats.completed_requests.swap(0, Ordering::Relaxed);
271 0 : let missed = stats.missed.swap(0, Ordering::Relaxed);
272 0 : let elapsed = start.elapsed();
273 0 : info!(
274 0 : "RPS: {:.0} MISSED: {:.0}",
275 0 : completed_requests as f64 / elapsed.as_secs_f64(),
276 0 : missed as f64 / elapsed.as_secs_f64()
277 : );
278 : }
279 0 : }
280 0 : });
281 0 :
282 0 : let rps_period = args
283 0 : .per_client_rate
284 0 : .map(|rps_limit| Duration::from_secs_f64(1.0 / (rps_limit as f64)));
285 0 : let make_worker: &dyn Fn(WorkerId) -> Pin<Box<dyn Send + Future<Output = ()>>> = &|worker_id| {
286 0 : let ss = shared_state.clone();
287 0 : let cancel = cancel.clone();
288 0 : let ranges: Vec<KeyRange> = all_ranges
289 0 : .iter()
290 0 : .filter(|r| r.timeline == worker_id.timeline)
291 0 : .cloned()
292 0 : .collect();
293 0 : let weights =
294 0 : rand::distributions::weighted::WeightedIndex::new(ranges.iter().map(|v| v.len()))
295 0 : .unwrap();
296 0 :
297 0 : Box::pin(async move {
298 0 : client_libpq(args, worker_id, ss, cancel, rps_period, ranges, weights).await
299 0 : })
300 0 : };
301 :
302 0 : info!("spawning workers");
303 0 : let mut workers = JoinSet::new();
304 0 : for timeline in timelines.iter().cloned() {
305 0 : for num_client in 0..args.num_clients.get() {
306 0 : let worker_id = WorkerId {
307 0 : timeline,
308 0 : num_client,
309 0 : };
310 0 : workers.spawn(make_worker(worker_id));
311 0 : }
312 : }
313 0 : let workers = async move {
314 0 : while let Some(res) = workers.join_next().await {
315 0 : res.unwrap();
316 0 : }
317 0 : };
318 :
319 0 : info!("waiting for everything to become ready");
320 0 : shared_state.start_work_barrier.wait().await;
321 0 : info!("work started");
322 0 : if let Some(runtime) = args.runtime {
323 0 : tokio::time::sleep(runtime.into()).await;
324 0 : info!("runtime over, signalling cancellation");
325 0 : cancel.cancel();
326 0 : workers.await;
327 0 : info!("work sender exited");
328 : } else {
329 0 : workers.await;
330 0 : unreachable!("work sender never terminates");
331 : }
332 :
333 0 : let output = Output {
334 : total: {
335 0 : let mut agg_stats = request_stats::Stats::new();
336 0 : for stats in all_thread_local_stats.lock().unwrap().iter() {
337 0 : let stats = stats.lock().unwrap();
338 0 : agg_stats.add(&stats);
339 0 : }
340 0 : agg_stats.output()
341 0 : },
342 0 : };
343 0 :
344 0 : let output = serde_json::to_string_pretty(&output).unwrap();
345 0 : println!("{output}");
346 0 :
347 0 : anyhow::Ok(())
348 0 : }
349 :
350 0 : async fn client_libpq(
351 0 : args: &Args,
352 0 : worker_id: WorkerId,
353 0 : shared_state: Arc<SharedState>,
354 0 : cancel: CancellationToken,
355 0 : rps_period: Option<Duration>,
356 0 : ranges: Vec<KeyRange>,
357 0 : weights: rand::distributions::weighted::WeightedIndex<i128>,
358 0 : ) {
359 0 : let client = pageserver_client::page_service::Client::new(args.page_service_connstring.clone())
360 0 : .await
361 0 : .unwrap();
362 0 : let mut client = client
363 0 : .pagestream(worker_id.timeline.tenant_id, worker_id.timeline.timeline_id)
364 0 : .await
365 0 : .unwrap();
366 0 :
367 0 : shared_state.start_work_barrier.wait().await;
368 0 : let client_start = Instant::now();
369 0 : let mut ticks_processed = 0;
370 0 : let mut inflight = VecDeque::new();
371 0 : while !cancel.is_cancelled() {
372 : // Detect if a request took longer than the RPS rate
373 0 : if let Some(period) = &rps_period {
374 0 : let periods_passed_until_now =
375 0 : usize::try_from(client_start.elapsed().as_micros() / period.as_micros()).unwrap();
376 0 :
377 0 : if periods_passed_until_now > ticks_processed {
378 0 : shared_state
379 0 : .live_stats
380 0 : .missed((periods_passed_until_now - ticks_processed) as u64);
381 0 : }
382 0 : ticks_processed = periods_passed_until_now;
383 0 : }
384 :
385 0 : while inflight.len() < args.queue_depth.get() {
386 0 : let start = Instant::now();
387 0 : let req = {
388 0 : let mut rng = rand::thread_rng();
389 0 : let r = &ranges[weights.sample(&mut rng)];
390 0 : let key: i128 = rng.gen_range(r.start..r.end);
391 0 : let key = Key::from_i128(key);
392 0 : assert!(key.is_rel_block_key());
393 0 : let (rel_tag, block_no) = key
394 0 : .to_rel_block()
395 0 : .expect("we filter non-rel-block keys out above");
396 0 : PagestreamGetPageRequest {
397 0 : hdr: PagestreamRequest {
398 0 : reqid: 0,
399 0 : request_lsn: if rng.gen_bool(args.req_latest_probability) {
400 0 : Lsn::MAX
401 : } else {
402 0 : r.timeline_lsn
403 : },
404 0 : not_modified_since: r.timeline_lsn,
405 0 : },
406 0 : rel: rel_tag,
407 0 : blkno: block_no,
408 0 : }
409 0 : };
410 0 : client.getpage_send(req).await.unwrap();
411 0 : inflight.push_back(start);
412 : }
413 :
414 0 : let start = inflight.pop_front().unwrap();
415 0 : client.getpage_recv().await.unwrap();
416 0 : let end = Instant::now();
417 0 : shared_state.live_stats.request_done();
418 0 : ticks_processed += 1;
419 0 : STATS.with(|stats| {
420 0 : stats
421 0 : .borrow()
422 0 : .lock()
423 0 : .unwrap()
424 0 : .observe(end.duration_since(start))
425 0 : .unwrap();
426 0 : });
427 :
428 0 : if let Some(period) = &rps_period {
429 0 : let next_at = client_start
430 0 : + Duration::from_micros(
431 0 : (ticks_processed) as u64 * u64::try_from(period.as_micros()).unwrap(),
432 0 : );
433 0 : tokio::time::sleep_until(next_at.into()).await;
434 0 : }
435 : }
436 0 : }
|
