metrics: convert some metrics to pageserver-level (#4490)

## Problem

Some metrics are better to be observed at page-server level. Otherwise,
as we have a lot of tenants in production, we cannot do a sum b/c
Prometheus has limit on how many time series we can aggregate. This also
helps reduce metrics scraping size.

## Summary of changes

Some integration tests are likely not to pass as it will check the
existence of some metrics. Waiting for CI complete and fix them.

Metrics downgraded: page cache hit (where we are likely to have a
page-server level page cache in the future instead of per-tenant), and
reconstruct time (this would better be tenant-level, as we have one pg
replayer for each tenant, but now we make it page-server level as we do
not need that fine-grained data).

---------

Signed-off-by: Alex Chi <iskyzh@gmail.com>
This commit is contained in:
Alex Chi Z
2023-06-14 17:12:34 -04:00
committed by GitHub
parent 94f315d490
commit 2252c5c282
4 changed files with 25 additions and 42 deletions

View File

@@ -1,4 +1,3 @@
use metrics::core::{AtomicU64, GenericCounter};
use metrics::{
register_counter_vec, register_histogram, register_histogram_vec, register_int_counter,
register_int_counter_vec, register_int_gauge, register_int_gauge_vec, register_uint_gauge_vec,
@@ -95,21 +94,19 @@ static READ_NUM_FS_LAYERS: Lazy<HistogramVec> = Lazy::new(|| {
});
// Metrics collected on operations on the storage repository.
static RECONSTRUCT_TIME: Lazy<HistogramVec> = Lazy::new(|| {
register_histogram_vec!(
pub static RECONSTRUCT_TIME: Lazy<Histogram> = Lazy::new(|| {
register_histogram!(
"pageserver_getpage_reconstruct_seconds",
"Time spent in reconstruct_value",
&["tenant_id", "timeline_id"],
"Time spent in reconstruct_value (reconstruct a page from deltas)",
CRITICAL_OP_BUCKETS.into(),
)
.expect("failed to define a metric")
});
static MATERIALIZED_PAGE_CACHE_HIT_DIRECT: Lazy<IntCounterVec> = Lazy::new(|| {
register_int_counter_vec!(
pub static MATERIALIZED_PAGE_CACHE_HIT_DIRECT: Lazy<IntCounter> = Lazy::new(|| {
register_int_counter!(
"pageserver_materialized_cache_hits_direct_total",
"Number of cache hits from materialized page cache without redo",
&["tenant_id", "timeline_id"]
)
.expect("failed to define a metric")
});
@@ -124,11 +121,10 @@ static GET_RECONSTRUCT_DATA_TIME: Lazy<HistogramVec> = Lazy::new(|| {
.expect("failed to define a metric")
});
static MATERIALIZED_PAGE_CACHE_HIT: Lazy<IntCounterVec> = Lazy::new(|| {
register_int_counter_vec!(
pub static MATERIALIZED_PAGE_CACHE_HIT: Lazy<IntCounter> = Lazy::new(|| {
register_int_counter!(
"pageserver_materialized_cache_hits_total",
"Number of cache hits from materialized page cache",
&["tenant_id", "timeline_id"]
)
.expect("failed to define a metric")
});
@@ -752,10 +748,7 @@ impl StorageTimeMetrics {
pub struct TimelineMetrics {
tenant_id: String,
timeline_id: String,
pub reconstruct_time_histo: Histogram,
pub get_reconstruct_data_time_histo: Histogram,
pub materialized_page_cache_hit_counter: GenericCounter<AtomicU64>,
pub materialized_page_cache_hit_upon_request_counter: GenericCounter<AtomicU64>,
pub flush_time_histo: StorageTimeMetrics,
pub compact_time_histo: StorageTimeMetrics,
pub create_images_time_histo: StorageTimeMetrics,
@@ -783,15 +776,9 @@ impl TimelineMetrics {
) -> Self {
let tenant_id = tenant_id.to_string();
let timeline_id = timeline_id.to_string();
let reconstruct_time_histo = RECONSTRUCT_TIME
.get_metric_with_label_values(&[&tenant_id, &timeline_id])
.unwrap();
let get_reconstruct_data_time_histo = GET_RECONSTRUCT_DATA_TIME
.get_metric_with_label_values(&[&tenant_id, &timeline_id])
.unwrap();
let materialized_page_cache_hit_counter = MATERIALIZED_PAGE_CACHE_HIT
.get_metric_with_label_values(&[&tenant_id, &timeline_id])
.unwrap();
let flush_time_histo =
StorageTimeMetrics::new(StorageTimeOperation::LayerFlush, &tenant_id, &timeline_id);
let compact_time_histo =
@@ -833,19 +820,18 @@ impl TimelineMetrics {
let read_num_fs_layers = READ_NUM_FS_LAYERS
.get_metric_with_label_values(&[&tenant_id, &timeline_id])
.unwrap();
let materialized_page_cache_hit_upon_request_counter = MATERIALIZED_PAGE_CACHE_HIT_DIRECT
.get_metric_with_label_values(&[&tenant_id, &timeline_id])
.unwrap();
let evictions_with_low_residence_duration =
evictions_with_low_residence_duration_builder.build(&tenant_id, &timeline_id);
// TODO(chi): remove this once we remove Lazy for all metrics. Otherwise this will not appear in the exporter
// and integration test will error.
MATERIALIZED_PAGE_CACHE_HIT_DIRECT.get();
MATERIALIZED_PAGE_CACHE_HIT.get();
TimelineMetrics {
tenant_id,
timeline_id,
reconstruct_time_histo,
get_reconstruct_data_time_histo,
materialized_page_cache_hit_counter,
materialized_page_cache_hit_upon_request_counter,
flush_time_histo,
compact_time_histo,
create_images_time_histo,
@@ -872,10 +858,7 @@ impl Drop for TimelineMetrics {
fn drop(&mut self) {
let tenant_id = &self.tenant_id;
let timeline_id = &self.timeline_id;
let _ = RECONSTRUCT_TIME.remove_label_values(&[tenant_id, timeline_id]);
let _ = GET_RECONSTRUCT_DATA_TIME.remove_label_values(&[tenant_id, timeline_id]);
let _ = MATERIALIZED_PAGE_CACHE_HIT.remove_label_values(&[tenant_id, timeline_id]);
let _ = MATERIALIZED_PAGE_CACHE_HIT_DIRECT.remove_label_values(&[tenant_id, timeline_id]);
let _ = LAST_RECORD_LSN.remove_label_values(&[tenant_id, timeline_id]);
let _ = WAIT_LSN_TIME.remove_label_values(&[tenant_id, timeline_id]);
let _ = RESIDENT_PHYSICAL_SIZE.remove_label_values(&[tenant_id, timeline_id]);

View File

@@ -47,7 +47,10 @@ use crate::tenant::{
use crate::config::PageServerConf;
use crate::keyspace::{KeyPartitioning, KeySpace, KeySpaceRandomAccum};
use crate::metrics::{TimelineMetrics, UNEXPECTED_ONDEMAND_DOWNLOADS};
use crate::metrics::{
TimelineMetrics, MATERIALIZED_PAGE_CACHE_HIT, MATERIALIZED_PAGE_CACHE_HIT_DIRECT,
RECONSTRUCT_TIME, UNEXPECTED_ONDEMAND_DOWNLOADS,
};
use crate::pgdatadir_mapping::LsnForTimestamp;
use crate::pgdatadir_mapping::{is_rel_fsm_block_key, is_rel_vm_block_key};
use crate::pgdatadir_mapping::{BlockNumber, CalculateLogicalSizeError};
@@ -539,9 +542,7 @@ impl Timeline {
match cached_lsn.cmp(&lsn) {
Ordering::Less => {} // there might be WAL between cached_lsn and lsn, we need to check
Ordering::Equal => {
self.metrics
.materialized_page_cache_hit_upon_request_counter
.inc();
MATERIALIZED_PAGE_CACHE_HIT_DIRECT.inc();
return Ok(cached_img); // exact LSN match, return the image
}
Ordering::Greater => {
@@ -563,8 +564,7 @@ impl Timeline {
.await?;
timer.stop_and_record();
self.metrics
.reconstruct_time_histo
RECONSTRUCT_TIME
.observe_closure_duration(|| self.reconstruct_value(key, lsn, reconstruct_state))
}
@@ -2387,7 +2387,7 @@ impl Timeline {
ValueReconstructResult::Continue => {
// If we reached an earlier cached page image, we're done.
if cont_lsn == cached_lsn + 1 {
self.metrics.materialized_page_cache_hit_counter.inc_by(1);
MATERIALIZED_PAGE_CACHE_HIT.inc_by(1);
return Ok(());
}
if prev_lsn <= cont_lsn {

View File

@@ -57,14 +57,16 @@ PAGESERVER_GLOBAL_METRICS: Tuple[str, ...] = (
"libmetrics_launch_timestamp",
"libmetrics_build_info",
"libmetrics_tracing_event_count_total",
"pageserver_materialized_cache_hits_total",
"pageserver_materialized_cache_hits_direct_total",
"pageserver_getpage_reconstruct_seconds_bucket",
"pageserver_getpage_reconstruct_seconds_count",
"pageserver_getpage_reconstruct_seconds_sum",
)
PAGESERVER_PER_TENANT_METRICS: Tuple[str, ...] = (
"pageserver_current_logical_size",
"pageserver_resident_physical_size",
"pageserver_getpage_reconstruct_seconds_bucket",
"pageserver_getpage_reconstruct_seconds_count",
"pageserver_getpage_reconstruct_seconds_sum",
"pageserver_getpage_get_reconstruct_data_seconds_bucket",
"pageserver_getpage_get_reconstruct_data_seconds_count",
"pageserver_getpage_get_reconstruct_data_seconds_sum",
@@ -73,8 +75,6 @@ PAGESERVER_PER_TENANT_METRICS: Tuple[str, ...] = (
"pageserver_io_operations_seconds_count",
"pageserver_io_operations_seconds_sum",
"pageserver_last_record_lsn",
"pageserver_materialized_cache_hits_total",
"pageserver_materialized_cache_hits_direct_total",
"pageserver_read_num_fs_layers_bucket",
"pageserver_read_num_fs_layers_count",
"pageserver_read_num_fs_layers_sum",

View File

@@ -213,7 +213,7 @@ def test_metrics_normal_work(neon_env_builder: NeonEnvBuilder):
# Test (a subset of) pageserver global metrics
for metric in PAGESERVER_GLOBAL_METRICS:
ps_samples = ps_metrics.query_all(metric, {})
assert len(ps_samples) > 0
assert len(ps_samples) > 0, f"expected at least one sample for {metric}"
for sample in ps_samples:
labels = ",".join([f'{key}="{value}"' for key, value in sample.labels.items()])
log.info(f"{sample.name}{{{labels}}} {sample.value}")