Fix broken beacon chain metrics, add slot clock metrics

This commit is contained in:
Paul Hauner
2019-08-12 17:44:47 +10:00
parent 95a320817e
commit d7c546844c
9 changed files with 68 additions and 28 deletions

View File

@@ -1,3 +1,4 @@
#![recursion_limit = "128"] // For lazy-static
#[macro_use]
extern crate lazy_static;

View File

@@ -1,6 +1,6 @@
use crate::{BeaconChain, BeaconChainTypes};
pub use lighthouse_metrics::*;
use types::{BeaconState, Epoch, EthSpec, Hash256, Slot};
use types::{BeaconState, Epoch, Hash256, Slot};
lazy_static! {
/*
@@ -140,17 +140,6 @@ lazy_static! {
*/
pub static ref PERSIST_CHAIN: Result<Histogram> =
try_create_histogram("beacon_persist_chain", "Time taken to update the canonical head");
}
// Lazy-static is split so we don't reach the crate-level recursion limit.
lazy_static! {
/*
* Slot Clock
*/
pub static ref PRESENT_SLOT: Result<IntGauge> =
try_create_int_gauge("beacon_present_slot", "The present slot, according to system time");
pub static ref PRESENT_EPOCH: Result<IntGauge> =
try_create_int_gauge("beacon_present_epoch", "The present epoch, according to system time");
/*
* Chain Head
@@ -194,21 +183,6 @@ lazy_static! {
/// Scrape the `beacon_chain` for metrics that are not constantly updated (e.g., the present slot,
/// head state info, etc) and update the Prometheus `DEFAULT_REGISTRY`.
pub fn scrape_for_metrics<T: BeaconChainTypes>(beacon_chain: &BeaconChain<T>) {
set_gauge_by_slot(
&PRESENT_SLOT,
beacon_chain
.read_slot_clock()
.unwrap_or_else(|| Slot::new(0)),
);
set_gauge_by_epoch(
&PRESENT_EPOCH,
beacon_chain
.read_slot_clock()
.map(|s| s.epoch(T::EthSpec::slots_per_epoch()))
.unwrap_or_else(|| Epoch::new(0)),
);
scrape_head_state::<T>(
&beacon_chain.head().beacon_state,
beacon_chain.head().beacon_state_root,

View File

@@ -26,3 +26,4 @@ tokio = "0.1.17"
url = "2.0"
lazy_static = "1.3.0"
lighthouse_metrics = { path = "../../eth2/utils/lighthouse_metrics" }
slot_clock = { path = "../../eth2/utils/slot_clock" }

View File

@@ -39,6 +39,23 @@ pub fn get_prometheus<T: BeaconChainTypes + 'static>(req: Request<Body>) -> ApiR
.get::<DBPath>()
.ok_or_else(|| ApiError::ServerError("DBPath extension missing".to_string()))?;
// There are two categories of metrics:
//
// - Dynamically updated: things like histograms and event counters that are updated on the
// fly.
// - Statically updated: things which are only updated at the time of the scrape (used where we
// can avoid cluttering up code with metrics calls).
//
// The `prometheus` crate has a `DEFAULT_REGISTRY` global singleton (via `lazy_static`) which
// keeps the state of all the metrics. Dynamically updated things will already be up-to-date in
// the registry (because they update themselves) however statically updated things need to be
// "scraped".
//
// We proceed by, first updating all the static metrics using `scrape_for_metrics(..)`. Then,
// using `prometheus::gather(..)` to collect the global `DEFAULT_REGISTRY` metrics into a
// string that can be returned via HTTP.
slot_clock::scrape_for_metrics::<T::EthSpec, T::SlotClock>(&beacon_chain.slot_clock);
store::scrape_for_metrics(&db_path);
beacon_chain::scrape_for_metrics(&beacon_chain);