From 5dda371c2b75213bb3fa286cc7ba612980379613 Mon Sep 17 00:00:00 2001
From: Anna Khanova <32508607+khanova@users.noreply.github.com>
Date: Wed, 24 Apr 2024 15:13:18 +0200
Subject: [PATCH] Fix a bug with retries (#7494)

## Problem

## Summary of changes

By default, it's 5s retry.
---
 proxy/src/bin/proxy.rs       |  7 +++++--
 proxy/src/cache/endpoints.rs | 12 ++++++++++--
 2 files changed, 15 insertions(+), 4 deletions(-)

diff --git a/proxy/src/bin/proxy.rs b/proxy/src/bin/proxy.rs
index 7df320fd42..760ccf40d4 100644
--- a/proxy/src/bin/proxy.rs
+++ b/proxy/src/bin/proxy.rs
@@ -403,7 +403,7 @@ async fn main() -> anyhow::Result<()> {
         maintenance_tasks.spawn(usage_metrics::task_main(metrics_config));
         client_tasks.spawn(usage_metrics::task_backup(
             &metrics_config.backup_metric_collection_config,
-            cancellation_token,
+            cancellation_token.clone(),
         ));
     }
 
@@ -423,7 +423,10 @@ async fn main() -> anyhow::Result<()> {
                 let cache = api.caches.endpoints_cache.clone();
                 let con = regional_redis_client;
                 let span = tracing::info_span!("endpoints_cache");
-                maintenance_tasks.spawn(async move { cache.do_read(con).await }.instrument(span));
+                maintenance_tasks.spawn(
+                    async move { cache.do_read(con, cancellation_token.clone()).await }
+                        .instrument(span),
+                );
             }
         }
     }
diff --git a/proxy/src/cache/endpoints.rs b/proxy/src/cache/endpoints.rs
index 2aa1986d5e..02511e6ff7 100644
--- a/proxy/src/cache/endpoints.rs
+++ b/proxy/src/cache/endpoints.rs
@@ -4,6 +4,7 @@ use std::{
         atomic::{AtomicBool, Ordering},
         Arc,
     },
+    time::Duration,
 };
 
 use dashmap::DashSet;
@@ -13,6 +14,7 @@ use redis::{
 };
 use serde::Deserialize;
 use tokio::sync::Mutex;
+use tokio_util::sync::CancellationToken;
 use tracing::info;
 
 use crate::{
@@ -111,16 +113,22 @@ impl EndpointsCache {
     pub async fn do_read(
         &self,
         mut con: ConnectionWithCredentialsProvider,
+        cancellation_token: CancellationToken,
     ) -> anyhow::Result<Infallible> {
         let mut last_id = "0-0".to_string();
         loop {
-            self.ready.store(false, Ordering::Release);
             if let Err(e) = con.connect().await {
                 tracing::error!("error connecting to redis: {:?}", e);
-                continue;
+                self.ready.store(false, Ordering::Release);
             }
             if let Err(e) = self.read_from_stream(&mut con, &mut last_id).await {
                 tracing::error!("error reading from redis: {:?}", e);
+                self.ready.store(false, Ordering::Release);
+            }
+            if cancellation_token.is_cancelled() {
+                info!("cancellation token is cancelled, exiting");
+                tokio::time::sleep(Duration::from_secs(60 * 60 * 24 * 7)).await;
+                // 1 week.
             }
             tokio::time::sleep(self.config.retry_interval).await;
         }