From 071e30cc530f2c89cd9bc07bffc07e9cb1a595aa Mon Sep 17 00:00:00 2001
From: anastasia <lubennikovaav@gmail.com>
Date: Wed, 27 Oct 2021 20:37:11 +0300
Subject: [PATCH] Expose TENANT_THREADS_COUNT metric to observe number of
 currently active checkpointer and GC threads

---
 pageserver/src/tenant_threads.rs | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)

diff --git a/pageserver/src/tenant_threads.rs b/pageserver/src/tenant_threads.rs
index 3dc824180b..d7d06fdef8 100644
--- a/pageserver/src/tenant_threads.rs
+++ b/pageserver/src/tenant_threads.rs
@@ -11,6 +11,7 @@ use std::sync::Mutex;
 use std::thread::JoinHandle;
 use std::time::Duration;
 use tracing::*;
+use zenith_metrics::{register_int_gauge_vec, IntGaugeVec};
 use zenith_utils::zid::ZTenantId;
 
 struct TenantHandleEntry {
@@ -25,6 +26,15 @@ lazy_static! {
         Mutex::new(HashMap::new());
 }
 
+lazy_static! {
+    static ref TENANT_THREADS_COUNT: IntGaugeVec = register_int_gauge_vec!(
+        "tenant_threads_count",
+        "Number of live tenant threads",
+        &["tenant_thread_type"]
+    )
+    .expect("failed to define a metric");
+}
+
 pub fn start_tenant_threads(conf: &'static PageServerConf, tenantid: ZTenantId) {
     //ensure that old threads are stopeed
     wait_for_tenant_threads_to_stop(tenantid);
@@ -69,6 +79,12 @@ pub fn wait_for_tenant_threads_to_stop(tenantid: ZTenantId) {
 /// Checkpointer thread's main loop
 ///
 fn checkpoint_loop(tenantid: ZTenantId, conf: &'static PageServerConf) -> Result<()> {
+    let gauge = TENANT_THREADS_COUNT.with_label_values(&["checkpointer"]);
+    gauge.inc();
+    scopeguard::defer! {
+        gauge.dec();
+    }
+
     loop {
         if tenant_mgr::get_tenant_state(tenantid) != TenantState::Active {
             break;
@@ -95,6 +111,12 @@ fn checkpoint_loop(tenantid: ZTenantId, conf: &'static PageServerConf) -> Result
 /// GC thread's main loop
 ///
 fn gc_loop(tenantid: ZTenantId, conf: &'static PageServerConf) -> Result<()> {
+    let gauge = TENANT_THREADS_COUNT.with_label_values(&["gc"]);
+    gauge.inc();
+    scopeguard::defer! {
+        gauge.dec();
+    }
+
     loop {
         if tenant_mgr::get_tenant_state(tenantid) != TenantState::Active {
             break;