rust/neon
1
0
Fork 0
mirror of https://github.com/neondatabase/neon.git synced 2026-01-06 21:12:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

pagebench: avoid noise about CopyFail in PS logs (#6392)

Browse Source 
Before this patch, pagebench get-page-latest-lsn would sometimes cause
noisy errors in pageserver log about `CopyFail` protocol message.

refs https://github.com/neondatabase/neon/issues/6390
This commit is contained in:
Christian Schwarz
2024-01-18 19:50:42 +01:00
committed by GitHub
parent 00936d19e1
commit e8f773387d
2 changed files with 48 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
pageserver/client/src/page_service.rs
View File

@@ -108,9 +108,32 @@ pub struct RelTagBlockNo {
}
impl PagestreamClient {
pub async fn shutdown(mut self) {
let _ = self.cancel_on_client_drop.take();
self.conn_task.await.unwrap();
pub async fn shutdown(self) {
let Self {
copy_both,
cancel_on_client_drop: cancel_conn_task,
conn_task,
} = self;
// The `copy_both` contains internal channel sender, the receiver of which is polled by `conn_task`.
// When `conn_task` observes the sender has been dropped, it sends a `FeMessage::CopyFail` into the connection.
// (see https://github.com/neondatabase/rust-postgres/blob/2005bf79573b8add5cf205b52a2b208e356cc8b0/tokio-postgres/src/copy_both.rs#L56).
//
// If we drop(copy_both) first, but then immediately drop the `cancel_on_client_drop`,
// the CopyFail mesage only makes it to the socket sometimes (i.e., it's a race).
//
// Further, the pageserver makes a lot of noise when it receives CopyFail.
// Computes don't send it in practice, they just hard-close the connection.
//
// So, let's behave like the computes and suppress the CopyFail as follows:
// kill the socket first, then drop copy_both.
//
// See also: https://www.postgresql.org/docs/current/protocol-flow.html#PROTOCOL-COPY
//
// NB: page_service doesn't have a use case to exit the `pagestream` mode currently.
// => https://github.com/neondatabase/neon/issues/6390
let _ = cancel_conn_task.unwrap();
conn_task.await.unwrap();
drop(copy_both);
}
pub async fn getpage(