rust/greptimedb
1
0
Fork 0
mirror of https://github.com/GreptimeTeam/greptimedb.git synced 2026-10-29 07:06:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

perf: avoid duplicate OTLP histogram conversion (#9880)

Browse Source 
Signed-off-by: shuiyisong <xixing.sys@gmail.com>
This commit is contained in:
shuiyisong
2026-10-23 08:05:38 +00:00
committed by GitHub
parent 815da8a6ef
commit afbe77e31f
3 changed files with 333 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/servers/benches/otlp_metrics.rs
+227 -1
View File
@@ -12,16 +12,242 @@
// See the License for the specific language governing permissions and
// limitations under the License.
//! Measures OTLP request-to-row conversion, excluding request cloning and storage.
//! Save a baseline with `cargo bench -p servers --bench otlp_metrics -- --save-baseline before`,
//! then compare with `cargo bench -p servers --bench otlp_metrics -- --baseline before`.
#[path = "otlp_metrics/fixtures.rs"]
mod fixtures;
use std::hint::black_box;
use std::time::Duration;
use api::v1::value::ValueData;
use common_catalog::consts::SEMANTIC_GRAPH_WINDOW_NANOS;
use common_query::prelude::{greptime_timestamp, set_default_prefix};
use criterion::{BatchSize, BenchmarkId, Criterion, Throughput, criterion_group, criterion_main};
use otel_arrow_rust::proto::opentelemetry::collector::metrics::v1::ExportMetricsServiceRequest;
use otel_arrow_rust::proto::opentelemetry::common::v1::{AnyValue, KeyValue, any_value};
use otel_arrow_rust::proto::opentelemetry::metrics::v1::{
AggregationTemporality, ExponentialHistogram, ExponentialHistogramDataPoint, Gauge, Histogram,
HistogramDataPoint, Metric, NumberDataPoint, ResourceMetrics, ScopeMetrics,
exponential_histogram_data_point::Buckets, metric, number_data_point,
};
use otel_arrow_rust::proto::opentelemetry::resource::v1::Resource;
use servers::otlp::metrics::to_grpc_insert_requests;
use session::protocol_ctx::OtlpMetricCtx;
const POINTS: usize = 128;
fn attribute(key: &str, value: &str) -> KeyValue {
KeyValue {
key: key.to_string(),
value: Some(AnyValue {
value: Some(any_value::Value::StringValue(value.to_string())),
}),
}
}
fn timestamp(index: usize) -> u64 {
((index % 4 + 1) as i64 * SEMANTIC_GRAPH_WINDOW_NANOS) as u64 + index as u64 * 1_000_000
}
fn point_attributes() -> Vec<KeyValue> {
vec![
attribute("http.request.method", "GET"),
attribute("route", "/api"),
]
}
fn request(data: metric::Data) -> ExportMetricsServiceRequest {
ExportMetricsServiceRequest {
resource_metrics: vec![ResourceMetrics {
resource: Some(Resource {
attributes: vec![
attribute("service.name", "api"),
attribute("service.namespace", "shop"),
attribute("service.instance.id", "instance-1"),
attribute("host.id", "host-1"),
attribute("k8s.pod.name", "api-1"),
],
..Default::default()
}),
scope_metrics: vec![ScopeMetrics {
metrics: vec![Metric {
name: "request_duration".to_string(),
data: Some(data),
..Default::default()
}],
..Default::default()
}],
..Default::default()
}],
}
}
fn exponential(bucket_counts: Vec<u64>, scale: i32, reject: bool) -> metric::Data {
let count = 2 + 2 * bucket_counts.iter().sum::<u64>();
metric::Data::ExponentialHistogram(ExponentialHistogram {
aggregation_temporality: AggregationTemporality::Cumulative as i32,
data_points: (0..POINTS)
.map(|index| ExponentialHistogramDataPoint {
attributes: point_attributes(),
start_time_unix_nano: 1,
time_unix_nano: timestamp(index),
count: count + u64::from(reject && index % 4 == 3),
sum: Some(100.0),
scale,
zero_count: 2,
positive: Some(Buckets {
offset: -128,
bucket_counts: bucket_counts.clone(),
}),
negative: Some(Buckets {
offset: -128,
bucket_counts: bucket_counts.clone(),
}),
zero_threshold: 1e-6,
..Default::default()
})
.collect(),
})
}
fn bench_otlp_metrics(c: &mut Criterion) {
set_default_prefix(None).unwrap();
let gauge = metric::Data::Gauge(Gauge {
data_points: (0..POINTS)
.map(|index| NumberDataPoint {
attributes: point_attributes(),
time_unix_nano: timestamp(index),
value: Some(number_data_point::Value::AsDouble(1.0)),
..Default::default()
})
.collect(),
});
let histogram = metric::Data::Histogram(Histogram {
aggregation_temporality: AggregationTemporality::Delta as i32,
data_points: (0..POINTS)
.map(|index| HistogramDataPoint {
attributes: point_attributes(),
start_time_unix_nano: 1,
time_unix_nano: timestamp(index),
count: 17,
sum: Some(100.0),
bucket_counts: vec![1; 17],
explicit_bounds: (0..16).map(|index| index as f64).collect(),
..Default::default()
})
.collect(),
});
let workloads = [
("gauge", gauge, POINTS, POINTS),
("classic_16", histogram, POINTS, POINTS * 19),
(
"exponential_16",
exponential(vec![1; 16], 0, false),
POINTS,
POINTS,
),
(
"exponential_256",
exponential(vec![1; 256], 0, false),
POINTS,
POINTS,
),
(
"exponential_sparse_256",
exponential((0..256).map(|i| u64::from(i % 16 == 0)).collect(), 0, false),
POINTS,
POINTS,
),
(
"exponential_downscale_256",
exponential(vec![1; 256], 10, false),
POINTS,
POINTS,
),
(
"exponential_reject_quarter_256",
exponential(vec![1; 256], 0, true),
POINTS * 3 / 4,
POINTS * 3 / 4,
),
];
let mut group = c.benchmark_group("otlp_metrics");
group.sample_size(30);
group.warm_up_time(Duration::from_secs(1));
group.measurement_time(Duration::from_secs(3));
group.throughput(Throughput::Elements(POINTS as u64));
for (name, data, accepted, rows) in workloads {
let request = request(data);
for resource_info in [false, true] {
let ctx = OtlpMetricCtx {
resource_info,
..Default::default()
};
let conversion = to_grpc_insert_requests(request.clone(), &mut ctx.clone()).unwrap();
assert_eq!(
conversion.outcome.accepted_data_points, accepted as i64,
"{name}"
);
assert_eq!(
conversion.outcome.rejected_data_points,
(POINTS - accepted) as i64,
"{name}"
);
assert_eq!(conversion.rows, rows, "{name}");
assert_eq!(conversion.resource_info.is_some(), resource_info, "{name}");
if let Some(descriptor) = conversion.resource_info {
let descriptor_rows = descriptor.inserts[0].rows.as_ref().unwrap();
let ts_index = descriptor_rows
.schema
.iter()
.position(|column| column.column_name == greptime_timestamp())
.unwrap();
let expected = (0..4)
.filter(|window| accepted == POINTS || *window != 3)
.map(|window| {
Some(ValueData::TimestampMillisecondValue(
(timestamp(POINTS - 4 + window) / 1_000_000) as i64,
))
})
.collect::<Vec<_>>();
let actual = descriptor_rows
.rows
.iter()
.map(|row| row.values[ts_index].value_data.clone())
.collect::<Vec<_>>();
assert_eq!(actual, expected, "{name}");
}
group.bench_with_input(
BenchmarkId::new(
name,
if resource_info {
"descriptor_on"
} else {
"descriptor_off"
},
),
&request,
|b, request| {
b.iter_batched(
|| (request.clone(), ctx.clone()),
|(request, mut ctx)| {
black_box(
to_grpc_insert_requests(black_box(request), &mut ctx).unwrap(),
)
},
BatchSize::LargeInput,
);
},
);
}
}
group.finish();
}
fn bench_conversion(c: &mut Criterion) {
let fixtures = fixtures::fixtures();
for (group_name, delta) in [
@@ -79,6 +305,6 @@ fn bench_conversion(c: &mut Criterion) {
criterion_group! {
name = benches;
config = Criterion::default().sample_size(20).warm_up_time(Duration::from_secs(1)).measurement_time(Duration::from_secs(3));
targets = bench_conversion
targets = bench_conversion, bench_otlp_metrics
}
criterion_main!(benches);
src/servers/src/otlp/metrics.rs
+89
View File
@@ -1591,6 +1591,95 @@ mod tests {
}
}
#[test]
fn test_conversion_describes_only_accepted_exponential_histogram_windows() {
set_default_prefix(None).unwrap();
let window = common_catalog::consts::SEMANTIC_GRAPH_WINDOW_NANOS as u64;
let mut valid = exponential_point();
valid.time_unix_nano = window + 1_000_000;
let mut invalid_count = valid.clone();
invalid_count.time_unix_nano += 8_000_000;
invalid_count.count += 1;
let mut invalid_scale = valid.clone();
invalid_scale.time_unix_nano += window;
invalid_scale.scale = -5;
// OTLP bucket conversion succeeds, but the shared native encoder rejects this index.
let invalid_native = ExponentialHistogramDataPoint {
time_unix_nano: 3 * window + 1_000_000,
count: 1,
positive: Some(exponential_buckets(2000, vec![1])),
..Default::default()
};
assert!(
exponential_histogram_value(&invalid_native, AggregationTemporality::Cumulative)
.unwrap_err()
.contains("out of range")
);
let stale = ExponentialHistogramDataPoint {
time_unix_nano: 4 * window + 1_000_000,
count: u64::MAX,
scale: i32::MIN,
flags: DataPointFlags::NoRecordedValueMask as u32,
..Default::default()
};
let invalid_time = ExponentialHistogramDataPoint {
time_unix_nano: 5 * window + 1_000_000,
start_time_unix_nano: 5 * window + 2_000_000,
..stale.clone()
};
let rejected = vec![invalid_count, invalid_scale, invalid_native, invalid_time];
let mut mixed = rejected.clone();
mixed.extend([valid, stale]);
for (points, expected) in [
(rejected, vec![]),
(
mixed,
vec![
(window / 1_000_000 + 1) as i64,
(4 * window / 1_000_000 + 1) as i64,
],
),
] {
let mut request = metrics_request(vec![exponential_metric(
"latency",
points,
AggregationTemporality::Cumulative,
)]);
request.resource_metrics[0].resource = Some(Resource {
attributes: vec![keyvalue("service.name", "api")],
..Default::default()
});
let conversion = to_grpc_insert_requests(request, &mut descriptor_ctx()).unwrap();
assert_eq!(conversion.outcome.rejected_data_points, 4);
assert_eq!(
conversion.outcome.accepted_data_points,
expected.len() as i64
);
assert_eq!(conversion.rows, expected.len());
assert_eq!(conversion.resource_info.is_some(), !expected.is_empty());
let actual = conversion
.resource_info
.map(|descriptor| {
let rows = descriptor.inserts[0].rows.as_ref().unwrap();
let ts_index = rows
.schema
.iter()
.position(|column| column.column_name == greptime_timestamp())
.unwrap();
rows.rows
.iter()
.map(|row| match row.values[ts_index].value_data {
Some(ValueData::TimestampMillisecondValue(ts)) => ts,
_ => panic!("expected descriptor timestamp"),
})
.collect::<Vec<_>>()
})
.unwrap_or_default();
assert_eq!(actual, expected);
}
}
#[test]
fn test_conversion_skips_descriptor_on_metric_name_collision() {
set_default_prefix(None).unwrap();
src/servers/src/otlp/metrics/resource_info.rs
+17
View File
@@ -217,6 +217,15 @@ mod tests {
.all(|(k, _)| k != "os.type" && k != "service.instance.id")
);
let mut reordered = attrs.clone();
reordered.reverse();
data.observe_times(&reordered, [200, 75]);
assert_eq!(data.rows.len(), 1);
assert_eq!(
data.rows.values().next().unwrap().values().next(),
Some(&200)
);
data.observe_times(&[kv("host.id", "h-2")], [10]);
assert_eq!(data.rows.len(), 2);
@@ -225,6 +234,14 @@ mod tests {
assert!(empty.into_row_insert_requests().unwrap().is_none());
}
#[test]
fn observe_without_accepted_points_does_not_emit_rows() {
let mut data = ResourceInfoData::default();
data.observe_times(&[kv("service.name", "api")], []);
assert!(data.is_empty());
assert!(data.into_row_insert_requests().unwrap().is_none());
}
/// Earlier windows would keep their metric rows but lose their entities.
#[test]
fn observe_keeps_one_row_per_graph_window() {