rust/neon
1
0
Fork 0
mirror of https://github.com/neondatabase/neon.git synced 2026-01-09 14:32:57 +00:00
Code Issues Packages Projects Releases Wiki Activity

Proxy: remove fail fast logic to connect to compute (#6759)

Browse Source 
## Problem

Flaky tests

## Summary of changes

Remove failfast logic
This commit is contained in:
Anna Khanova
2024-02-14 19:43:52 +01:00
committed by GitHub
parent a2d0d44b42
commit c7538a2c20
2 changed files with 17 additions and 54 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
proxy/src/proxy/connect_compute.rs
View File

@@ -122,25 +122,24 @@ where
error!(error = ?err, "could not connect to compute node");
let node_info =
if err.get_error_kind() == crate::error::ErrorKind::Postgres || !node_info.cached() {
// If the error is Postgres, that means that we managed to connect to the compute node, but there was an error.
// Do not need to retrieve a new node_info, just return the old one.
if !err.should_retry(num_retries) {
return Err(err.into());
}
node_info
} else {
// if we failed to connect, it's likely that the compute node was suspended, wake a new compute node
info!("compute node's state has likely changed; requesting a wake-up");
ctx.latency_timer.cache_miss();
let old_node_info = invalidate_cache(node_info);
let mut node_info = wake_compute(&mut num_retries, ctx, user_info).await?;
node_info.reuse_settings(old_node_info);
let node_info = if !node_info.cached() {
// If we just recieved this from cplane and dodn't get it from cache, we shouldn't retry.
// Do not need to retrieve a new node_info, just return the old one.
if !err.should_retry(num_retries) {
return Err(err.into());
}
node_info
} else {
// if we failed to connect, it's likely that the compute node was suspended, wake a new compute node
info!("compute node's state has likely changed; requesting a wake-up");
ctx.latency_timer.cache_miss();
let old_node_info = invalidate_cache(node_info);
let mut node_info = wake_compute(&mut num_retries, ctx, user_info).await?;
node_info.reuse_settings(old_node_info);
mechanism.update_connect_config(&mut node_info.config);
node_info
};
mechanism.update_connect_config(&mut node_info.config);
node_info
};
// now that we have a new node, try connect to it repeatedly.
// this can error for a few reasons, for instance:

36
proxy/src/proxy/tests.rs
View File

@@ -375,8 +375,6 @@ enum ConnectAction {
Connect,
Retry,
Fail,
RetryPg,
FailPg,
}
#[derive(Clone)]
@@ -466,14 +464,6 @@ impl ConnectMechanism for TestConnectMechanism {
retryable: false,
kind: ErrorKind::Compute,
}),
ConnectAction::FailPg => Err(TestConnectError {
retryable: false,
kind: ErrorKind::Postgres,
}),
ConnectAction::RetryPg => Err(TestConnectError {
retryable: true,
kind: ErrorKind::Postgres,
}),
x => panic!("expecting action {:?}, connect is called instead", x),
}
}
@@ -572,32 +562,6 @@ async fn connect_to_compute_retry() {
mechanism.verify();
}
#[tokio::test]
async fn connect_to_compute_retry_pg() {
let _ = env_logger::try_init();
use ConnectAction::*;
let mut ctx = RequestMonitoring::test();
let mechanism = TestConnectMechanism::new(vec![Wake, RetryPg, Connect]);
let user_info = helper_create_connect_info(&mechanism);
connect_to_compute(&mut ctx, &mechanism, &user_info, false)
.await
.unwrap();
mechanism.verify();
}
#[tokio::test]
async fn connect_to_compute_fail_pg() {
let _ = env_logger::try_init();
use ConnectAction::*;
let mut ctx = RequestMonitoring::test();
let mechanism = TestConnectMechanism::new(vec![Wake, FailPg]);
let user_info = helper_create_connect_info(&mechanism);
connect_to_compute(&mut ctx, &mechanism, &user_info, false)
.await
.unwrap_err();
mechanism.verify();
}
/// Test that we don't retry if the error is not retryable.
#[tokio::test]
async fn connect_to_compute_non_retry_1() {