removed many more metrics, still compiles

This commit is contained in:
BodoBolero
2025-04-23 16:12:01 +02:00
parent a12369be43
commit ce1e575db1
29 changed files with 207 additions and 1878 deletions

View File

@@ -76,14 +76,7 @@ pub fn gather() -> Vec<prometheus::proto::MetricFamily> {
mfs
}
static DISK_IO_BYTES: Lazy<IntGaugeVec> = Lazy::new(|| {
register_int_gauge_vec!(
"libmetrics_disk_io_bytes_total",
"Bytes written and read from disk, grouped by the operation (read|write)",
&["io_operation"]
)
.expect("Failed to register disk i/o bytes int gauge vec")
});
static MAXRSS_KB: Lazy<IntGauge> = Lazy::new(|| {
register_int_gauge!(
@@ -261,12 +254,7 @@ const BYTES_IN_BLOCK: i64 = 512;
fn update_rusage_metrics() {
let rusage_stats = get_rusage_stats();
DISK_IO_BYTES
.with_label_values(&["read"])
.set(rusage_stats.ru_inblock * BYTES_IN_BLOCK);
DISK_IO_BYTES
.with_label_values(&["write"])
.set(rusage_stats.ru_oublock * BYTES_IN_BLOCK);
// On macOS, the unit of maxrss is bytes; on Linux, it's kilobytes. https://stackoverflow.com/a/59915669
#[cfg(target_os = "macos")]
@@ -357,10 +345,7 @@ impl<P: Atomic> GenericCounterPairVec<P> {
self.get_metric_with_label_values(vals).unwrap()
}
pub fn remove_label_values(&self, res: &mut [prometheus::Result<()>; 2], vals: &[&str]) {
res[0] = self.inc.remove_label_values(vals);
res[1] = self.dec.remove_label_values(vals);
}
}
impl<P: Atomic> GenericCounterPair<P> {

View File

@@ -23,7 +23,6 @@ use futures::future::Either;
use futures::stream::Stream;
use futures_util::{StreamExt, TryStreamExt};
use http_types::{StatusCode, Url};
use scopeguard::ScopeGuard;
use tokio_util::sync::CancellationToken;
use tracing::debug;
use utils::backoff;
@@ -32,7 +31,7 @@ use utils::backoff::exponential_backoff_duration_seconds;
use super::REMOTE_STORAGE_PREFIX_SEPARATOR;
use crate::config::AzureConfig;
use crate::error::Cancelled;
use crate::metrics::{AttemptOutcome, RequestKind, start_measuring_requests};
use crate::metrics::RequestKind;
use crate::{
ConcurrencyLimiter, Download, DownloadError, DownloadKind, DownloadOpts, Listing, ListingMode,
ListingObject, RemotePath, RemoteStorage, StorageMetadata, TimeTravelError, TimeoutOrCancel,
@@ -165,7 +164,7 @@ impl AzureBlobStorage {
let mut last_modified = None;
let mut metadata = HashMap::new();
let started_at = start_measuring_requests(kind);
let download = async {
let response = builder
@@ -237,19 +236,8 @@ impl AzureBlobStorage {
TimeoutOrCancel::Cancel => return Err(DownloadError::Cancelled),
},
};
let started_at = ScopeGuard::into_inner(started_at);
let outcome = match &download {
Ok(_) => AttemptOutcome::Ok,
// At this level in the stack 404 and 304 responses do not indicate an error.
// There's expected cases when a blob may not exist or hasn't been modified since
// the last get (e.g. probing for timeline indices and heatmap downloads).
// Callers should handle errors if they are unexpected.
Err(DownloadError::NotFound | DownloadError::Unmodified) => AttemptOutcome::Ok,
Err(_) => AttemptOutcome::Err,
};
crate::metrics::BUCKET_METRICS
.req_seconds
.observe_elapsed(kind, outcome, started_at);
download
}
@@ -431,7 +419,7 @@ impl RemoteStorage for AzureBlobStorage {
let kind = RequestKind::Head;
let _permit = self.permit(kind, cancel).await?;
let started_at = start_measuring_requests(kind);
let blob_client = self.client.blob_client(self.relative_path_to_name(key));
let properties_future = blob_client.get_properties().into_future();
@@ -443,12 +431,9 @@ impl RemoteStorage for AzureBlobStorage {
_ = cancel.cancelled() => return Err(TimeoutOrCancel::Cancel.into()),
};
if let Ok(inner) = &res {
// do not incl. timeouts as errors in metrics but cancellations
let started_at = ScopeGuard::into_inner(started_at);
crate::metrics::BUCKET_METRICS
.req_seconds
.observe_elapsed(kind, inner, started_at);
if let Ok(_inner) = &res {
}
let data = match res {
@@ -476,7 +461,7 @@ impl RemoteStorage for AzureBlobStorage {
let kind = RequestKind::Put;
let _permit = self.permit(kind, cancel).await?;
let started_at = start_measuring_requests(kind);
let op = async {
let blob_client = self.client.blob_client(self.relative_path_to_name(to));
@@ -509,14 +494,7 @@ impl RemoteStorage for AzureBlobStorage {
_ = cancel.cancelled() => return Err(TimeoutOrCancel::Cancel.into()),
};
let outcome = match res {
Ok(_) => AttemptOutcome::Ok,
Err(_) => AttemptOutcome::Err,
};
let started_at = ScopeGuard::into_inner(started_at);
crate::metrics::BUCKET_METRICS
.req_seconds
.observe_elapsed(kind, outcome, started_at);
res
}
@@ -562,7 +540,7 @@ impl RemoteStorage for AzureBlobStorage {
) -> anyhow::Result<()> {
let kind = RequestKind::Delete;
let _permit = self.permit(kind, cancel).await?;
let started_at = start_measuring_requests(kind);
let op = async {
// TODO batch requests are not supported by the SDK
@@ -628,10 +606,8 @@ impl RemoteStorage for AzureBlobStorage {
_ = cancel.cancelled() => return Err(TimeoutOrCancel::Cancel.into()),
};
let started_at = ScopeGuard::into_inner(started_at);
crate::metrics::BUCKET_METRICS
.req_seconds
.observe_elapsed(kind, &res, started_at);
res
}
@@ -647,7 +623,7 @@ impl RemoteStorage for AzureBlobStorage {
) -> anyhow::Result<()> {
let kind = RequestKind::Copy;
let _permit = self.permit(kind, cancel).await?;
let started_at = start_measuring_requests(kind);
let timeout = tokio::time::sleep(self.timeout);
@@ -701,10 +677,8 @@ impl RemoteStorage for AzureBlobStorage {
},
};
let started_at = ScopeGuard::into_inner(started_at);
crate::metrics::BUCKET_METRICS
.req_seconds
.observe_elapsed(kind, &res, started_at);
res
}

View File

@@ -1,9 +1,7 @@
use metrics::{
Histogram, IntCounter, register_histogram_vec, register_int_counter, register_int_counter_vec,
};
use once_cell::sync::Lazy;
pub(super) static BUCKET_METRICS: Lazy<BucketMetrics> = Lazy::new(Default::default);
#[derive(Clone, Copy, Debug)]
pub(crate) enum RequestKind {
@@ -16,62 +14,9 @@ pub(crate) enum RequestKind {
Head = 6,
}
use RequestKind::*;
use scopeguard::ScopeGuard;
impl RequestKind {
const fn as_str(&self) -> &'static str {
match self {
Get => "get_object",
Put => "put_object",
Delete => "delete_object",
List => "list_objects",
Copy => "copy_object",
TimeTravel => "time_travel_recover",
Head => "head_object",
}
}
const fn as_index(&self) -> usize {
*self as usize
}
}
const REQUEST_KIND_COUNT: usize = 7;
pub(crate) struct RequestTyped<C>([C; REQUEST_KIND_COUNT]);
impl<C> RequestTyped<C> {
pub(crate) fn get(&self, kind: RequestKind) -> &C {
&self.0[kind.as_index()]
}
fn build_with(mut f: impl FnMut(RequestKind) -> C) -> Self {
use RequestKind::*;
let mut it = [Get, Put, Delete, List, Copy, TimeTravel, Head].into_iter();
let arr = std::array::from_fn::<C, REQUEST_KIND_COUNT, _>(|index| {
let next = it.next().unwrap();
assert_eq!(index, next.as_index());
f(next)
});
if let Some(next) = it.next() {
panic!("unexpected {next:?}");
}
RequestTyped(arr)
}
}
impl RequestTyped<Histogram> {
pub(crate) fn observe_elapsed(&self, kind: RequestKind, started_at: std::time::Instant) {
self.get(kind).observe(started_at.elapsed().as_secs_f64())
}
}
pub(crate) struct PassFailCancelledRequestTyped<C> {
success: RequestTyped<C>,
fail: RequestTyped<C>,
cancelled: RequestTyped<C>,
}
#[derive(Debug, Clone, Copy)]
pub(crate) enum AttemptOutcome {
@@ -89,138 +34,22 @@ impl<T, E> From<&Result<T, E>> for AttemptOutcome {
}
}
impl AttemptOutcome {
pub(crate) fn as_str(&self) -> &'static str {
match self {
AttemptOutcome::Ok => "ok",
AttemptOutcome::Err => "err",
AttemptOutcome::Cancelled => "cancelled",
}
}
}
impl<C> PassFailCancelledRequestTyped<C> {
pub(crate) fn get(&self, kind: RequestKind, outcome: AttemptOutcome) -> &C {
let target = match outcome {
AttemptOutcome::Ok => &self.success,
AttemptOutcome::Err => &self.fail,
AttemptOutcome::Cancelled => &self.cancelled,
};
target.get(kind)
}
fn build_with(mut f: impl FnMut(RequestKind, AttemptOutcome) -> C) -> Self {
let success = RequestTyped::build_with(|kind| f(kind, AttemptOutcome::Ok));
let fail = RequestTyped::build_with(|kind| f(kind, AttemptOutcome::Err));
let cancelled = RequestTyped::build_with(|kind| f(kind, AttemptOutcome::Cancelled));
PassFailCancelledRequestTyped {
success,
fail,
cancelled,
}
}
}
impl PassFailCancelledRequestTyped<Histogram> {
pub(crate) fn observe_elapsed(
&self,
kind: RequestKind,
outcome: impl Into<AttemptOutcome>,
started_at: std::time::Instant,
) {
self.get(kind, outcome.into())
.observe(started_at.elapsed().as_secs_f64())
}
}
/// On drop (cancellation) count towards [`BucketMetrics::cancelled_waits`].
pub(crate) fn start_counting_cancelled_wait(
kind: RequestKind,
) -> ScopeGuard<std::time::Instant, impl FnOnce(std::time::Instant), scopeguard::OnSuccess> {
scopeguard::guard_on_success(std::time::Instant::now(), move |_| {
crate::metrics::BUCKET_METRICS
.cancelled_waits
.get(kind)
.inc()
})
}
/// On drop (cancellation) add time to [`BucketMetrics::req_seconds`].
pub(crate) fn start_measuring_requests(
kind: RequestKind,
_kind: RequestKind,
) -> ScopeGuard<std::time::Instant, impl FnOnce(std::time::Instant), scopeguard::OnSuccess> {
scopeguard::guard_on_success(std::time::Instant::now(), move |started_at| {
crate::metrics::BUCKET_METRICS.req_seconds.observe_elapsed(
kind,
AttemptOutcome::Cancelled,
started_at,
)
scopeguard::guard_on_success(std::time::Instant::now(), move |_started_at| {
})
}
pub(crate) struct BucketMetrics {
/// Full request duration until successful completion, error or cancellation.
pub(crate) req_seconds: PassFailCancelledRequestTyped<Histogram>,
/// Total amount of seconds waited on queue.
pub(crate) wait_seconds: RequestTyped<Histogram>,
/// Track how many semaphore awaits were cancelled per request type.
///
/// This is in case cancellations are happening more than expected.
pub(crate) cancelled_waits: RequestTyped<IntCounter>,
/// Total amount of deleted objects in batches or single requests.
pub(crate) deleted_objects_total: IntCounter,
}
impl Default for BucketMetrics {
fn default() -> Self {
// first bucket 100 microseconds to count requests that do not need to wait at all
// and get a permit immediately
let buckets = [0.0001, 0.01, 0.10, 0.5, 1.0, 5.0, 10.0, 50.0, 100.0];
let req_seconds = register_histogram_vec!(
"remote_storage_s3_request_seconds",
"Seconds to complete a request",
&["request_type", "result"],
buckets.to_vec(),
)
.unwrap();
let req_seconds = PassFailCancelledRequestTyped::build_with(|kind, outcome| {
req_seconds.with_label_values(&[kind.as_str(), outcome.as_str()])
});
let wait_seconds = register_histogram_vec!(
"remote_storage_s3_wait_seconds",
"Seconds rate limited",
&["request_type"],
buckets.to_vec(),
)
.unwrap();
let wait_seconds =
RequestTyped::build_with(|kind| wait_seconds.with_label_values(&[kind.as_str()]));
let cancelled_waits = register_int_counter_vec!(
"remote_storage_s3_cancelled_waits_total",
"Times a semaphore wait has been cancelled per request type",
&["request_type"],
)
.unwrap();
let cancelled_waits =
RequestTyped::build_with(|kind| cancelled_waits.with_label_values(&[kind.as_str()]));
let deleted_objects_total = register_int_counter!(
"remote_storage_s3_deleted_objects_total",
"Amount of deleted objects in total",
)
.unwrap();
Self {
req_seconds,
wait_seconds,
cancelled_waits,
deleted_objects_total,
}
}
}

View File

@@ -41,7 +41,7 @@ use super::StorageMetadata;
use crate::config::S3Config;
use crate::error::Cancelled;
pub(super) use crate::metrics::RequestKind;
use crate::metrics::{AttemptOutcome, start_counting_cancelled_wait, start_measuring_requests};
use crate::metrics::{AttemptOutcome, start_measuring_requests};
use crate::support::PermitCarrying;
use crate::{
ConcurrencyLimiter, Download, DownloadError, DownloadOpts, Listing, ListingMode, ListingObject,
@@ -199,7 +199,7 @@ impl S3Bucket {
kind: RequestKind,
cancel: &CancellationToken,
) -> Result<tokio::sync::SemaphorePermit<'_>, Cancelled> {
let started_at = start_counting_cancelled_wait(kind);
let acquire = self.concurrency_limiter.acquire(kind);
let permit = tokio::select! {
@@ -207,10 +207,8 @@ impl S3Bucket {
_ = cancel.cancelled() => return Err(Cancelled),
};
let started_at = ScopeGuard::into_inner(started_at);
crate::metrics::BUCKET_METRICS
.wait_seconds
.observe_elapsed(kind, started_at);
Ok(permit)
}
@@ -220,7 +218,7 @@ impl S3Bucket {
kind: RequestKind,
cancel: &CancellationToken,
) -> Result<tokio::sync::OwnedSemaphorePermit, Cancelled> {
let started_at = start_counting_cancelled_wait(kind);
let acquire = self.concurrency_limiter.acquire_owned(kind);
let permit = tokio::select! {
@@ -228,10 +226,8 @@ impl S3Bucket {
_ = cancel.cancelled() => return Err(Cancelled),
};
let started_at = ScopeGuard::into_inner(started_at);
crate::metrics::BUCKET_METRICS
.wait_seconds
.observe_elapsed(kind, started_at);
Ok(permit)
}
@@ -273,11 +269,7 @@ impl S3Bucket {
// Count this in the AttemptOutcome::Ok bucket, because 404 is not
// an error: we expect to sometimes fetch an object and find it missing,
// e.g. when probing for timeline indices.
crate::metrics::BUCKET_METRICS.req_seconds.observe_elapsed(
kind,
AttemptOutcome::Ok,
started_at,
);
return Err(DownloadError::NotFound);
}
Err(SdkError::ServiceError(e))
@@ -287,19 +279,11 @@ impl S3Bucket {
if e.raw().status().as_u16() == StatusCode::NotModified =>
{
// Count an unmodified file as a success.
crate::metrics::BUCKET_METRICS.req_seconds.observe_elapsed(
kind,
AttemptOutcome::Ok,
started_at,
);
return Err(DownloadError::Unmodified);
}
Err(e) => {
crate::metrics::BUCKET_METRICS.req_seconds.observe_elapsed(
kind,
AttemptOutcome::Err,
started_at,
);
return Err(DownloadError::Other(
anyhow::Error::new(e).context("download s3 object"),
@@ -346,11 +330,11 @@ impl S3Bucket {
delete_objects: &[ObjectIdentifier],
cancel: &CancellationToken,
) -> anyhow::Result<()> {
let kind = RequestKind::Delete;
let mut cancel = std::pin::pin!(cancel.cancelled());
for chunk in delete_objects.chunks(MAX_KEYS_PER_DELETE_S3) {
let started_at = start_measuring_requests(kind);
let req = self
.client
@@ -370,15 +354,10 @@ impl S3Bucket {
_ = &mut cancel => return Err(TimeoutOrCancel::Cancel.into()),
};
let started_at = ScopeGuard::into_inner(started_at);
crate::metrics::BUCKET_METRICS
.req_seconds
.observe_elapsed(kind, &resp, started_at);
let resp = resp.context("request deletion")?;
crate::metrics::BUCKET_METRICS
.deleted_objects_total
.inc_by(chunk.len() as u64);
if let Some(errors) = resp.errors {
// Log a bounded number of the errors within the response:
@@ -445,8 +424,8 @@ pin_project_lite::pin_project! {
}
impl<S> PinnedDrop for TimedDownload<S> {
fn drop(mut this: Pin<&mut Self>) {
crate::metrics::BUCKET_METRICS.req_seconds.observe_elapsed(RequestKind::Get, this.outcome, this.started_at);
fn drop(mut _this: Pin<&mut Self>) {
}
}
}
@@ -511,7 +490,7 @@ impl RemoteStorage for S3Bucket {
let mut continuation_token = None;
'outer: loop {
let started_at = start_measuring_requests(kind);
// min of two Options, returning Some if one is value and another is
// None (None is smaller than anything, so plain min doesn't work).
@@ -544,11 +523,9 @@ impl RemoteStorage for S3Bucket {
.context("Failed to list S3 prefixes")
.map_err(DownloadError::Other);
let started_at = ScopeGuard::into_inner(started_at);
crate::metrics::BUCKET_METRICS
.req_seconds
.observe_elapsed(kind, &response, started_at);
let response = match response {
Ok(response) => response,
@@ -629,7 +606,7 @@ impl RemoteStorage for S3Bucket {
let kind = RequestKind::Head;
let _permit = self.permit(kind, cancel).await?;
let started_at = start_measuring_requests(kind);
let head_future = self
.client
@@ -648,30 +625,18 @@ impl RemoteStorage for S3Bucket {
let res = res.map_err(|_e| DownloadError::Timeout)?;
// do not incl. timeouts as errors in metrics but cancellations
let started_at = ScopeGuard::into_inner(started_at);
crate::metrics::BUCKET_METRICS
.req_seconds
.observe_elapsed(kind, &res, started_at);
let data = match res {
Ok(object_output) => object_output,
Err(SdkError::ServiceError(e)) if matches!(e.err(), HeadObjectError::NotFound(_)) => {
// Count this in the AttemptOutcome::Ok bucket, because 404 is not
// an error: we expect to sometimes fetch an object and find it missing,
// e.g. when probing for timeline indices.
crate::metrics::BUCKET_METRICS.req_seconds.observe_elapsed(
kind,
AttemptOutcome::Ok,
started_at,
);
return Err(DownloadError::NotFound);
}
Err(e) => {
crate::metrics::BUCKET_METRICS.req_seconds.observe_elapsed(
kind,
AttemptOutcome::Err,
started_at,
);
return Err(DownloadError::Other(
anyhow::Error::new(e).context("s3 head object"),
@@ -704,7 +669,7 @@ impl RemoteStorage for S3Bucket {
let kind = RequestKind::Put;
let _permit = self.permit(kind, cancel).await?;
let started_at = start_measuring_requests(kind);
let body = StreamBody::new(from.map(|x| x.map(Frame::data)));
let bytes_stream = ByteStream::new(SdkBody::from_body_1_x(body));
@@ -727,12 +692,10 @@ impl RemoteStorage for S3Bucket {
_ = cancel.cancelled() => return Err(TimeoutOrCancel::Cancel.into()),
};
if let Ok(inner) = &res {
if let Ok(_inner) = &res {
// do not incl. timeouts as errors in metrics but cancellations
let started_at = ScopeGuard::into_inner(started_at);
crate::metrics::BUCKET_METRICS
.req_seconds
.observe_elapsed(kind, inner, started_at);
}
match res {
@@ -753,7 +716,7 @@ impl RemoteStorage for S3Bucket {
let timeout = tokio::time::sleep(self.timeout);
let started_at = start_measuring_requests(kind);
// we need to specify bucket_name as a prefix
let copy_source = format!(
@@ -777,10 +740,8 @@ impl RemoteStorage for S3Bucket {
_ = cancel.cancelled() => return Err(TimeoutOrCancel::Cancel.into()),
};
let started_at = ScopeGuard::into_inner(started_at);
crate::metrics::BUCKET_METRICS
.req_seconds
.observe_elapsed(kind, &res, started_at);
res?;