[pbs-devel] [PATCH v2 proxmox-backup 2/3] api: reader: handle reader client disconnects

[Christian Ebner]

Currently, if a reader client disconnects after finishing its work,
the connection will be closed by the client without notifying the
server. The future handling the connection on then server side will
then return with a connection error, and in consequence the reader
worker task will log with error state. This can cause confusion [0],
as this is not an error but normal behaviour.

Instead of failing, provide an api endpoint for the client to signal
it has finished operation. The server sets the connection environment
state accordingly, and the connection error is suppressed if the flag
has been set. This follows the same logic used for the backup writer,
introduced by commit b428af97 ("backup: avoid Transport endpoint is
not connected error").

Report in the community forum:
[0] https://forum.proxmox.com/threads/158306/

Signed-off-by: Christian Ebner <c.ebner at proxmox.com>
---
changes since version 1:
- Use the same approach as used for the backup writer, as the
  connection graceful_shutdown did not resolve the issue at hand.

 src/api2/reader/environment.rs | 20 +++++++++++++++++++-
 src/api2/reader/mod.rs         | 30 +++++++++++++++++++++++++++---
 2 files changed, 46 insertions(+), 4 deletions(-)

diff --git a/src/api2/reader/environment.rs b/src/api2/reader/environment.rs
index 3b2f06f43..3cdc8e394 100644
--- a/src/api2/reader/environment.rs
+++ b/src/api2/reader/environment.rs
@@ -1,5 +1,5 @@
 use std::collections::HashSet;
-use std::sync::{Arc, RwLock};
+use std::sync::{Arc, Mutex, RwLock};
 
 use serde_json::{json, Value};
 
@@ -24,6 +24,11 @@ pub struct ReaderEnvironment {
     pub datastore: Arc<DataStore>,
     pub backup_dir: BackupDir,
     allowed_chunks: Arc<RwLock<HashSet<[u8; 32]>>>,
+    connection_state: Arc<Mutex<ConnectionState>>,
+}
+
+struct ConnectionState {
+    client_finished: bool,
 }
 
 impl ReaderEnvironment {
@@ -44,6 +49,9 @@ impl ReaderEnvironment {
             formatter: JSON_FORMATTER,
             backup_dir,
             allowed_chunks: Arc::new(RwLock::new(HashSet::new())),
+            connection_state: Arc::new(Mutex::new(ConnectionState {
+                client_finished: false,
+            })),
         }
     }
 
@@ -69,6 +77,16 @@ impl ReaderEnvironment {
     pub fn check_chunk_access(&self, digest: [u8; 32]) -> bool {
         self.allowed_chunks.read().unwrap().contains(&digest)
     }
+
+    pub(crate) fn client_finished(&self) -> bool {
+        let state = self.connection_state.lock().unwrap();
+        state.client_finished
+    }
+
+    pub(crate) fn finish(&self) {
+        let mut state = self.connection_state.lock().unwrap();
+        state.client_finished = true;
+    }
 }
 
 impl RpcEnvironment for ReaderEnvironment {
diff --git a/src/api2/reader/mod.rs b/src/api2/reader/mod.rs
index 50f80de43..cb53f6b5e 100644
--- a/src/api2/reader/mod.rs
+++ b/src/api2/reader/mod.rs
@@ -192,9 +192,16 @@ fn upgrade_to_backup_reader_protocol(
                     http.http2_initial_connection_window_size(window_size);
                     http.http2_max_frame_size(4 * 1024 * 1024);
 
-                    http.serve_connection(conn, service)
-                        .map_err(Error::from)
-                        .await
+                    if let Err(err) = http.serve_connection(conn, service).await {
+                        // Avoid  Transport endpoint is not connected (os error 107)
+                        // fixme: find a better way to test for that error
+                        if !(err.to_string().starts_with("connection error")
+                            && env2.client_finished())
+                        {
+                            return Err(Error::from(err));
+                        }
+                    }
+                    Ok(())
                 };
 
                 futures::select! {
@@ -228,6 +235,7 @@ const READER_API_SUBDIRS: SubdirMap = &[
         "download",
         &Router::new().download(&API_METHOD_DOWNLOAD_FILE),
     ),
+    ("finish", &Router::new().post(&API_METHOD_FINISH)),
     ("speedtest", &Router::new().download(&API_METHOD_SPEEDTEST)),
 ];
 
@@ -347,6 +355,22 @@ fn download_chunk(
     .boxed()
 }
 
+#[sortable]
+pub const API_METHOD_FINISH: ApiMethod = ApiMethod::new(
+    &ApiHandler::Sync(&finish),
+    &ObjectSchema::new("Signal the reader instance is finished", &[]),
+);
+
+fn finish(
+    _param: Value,
+    _info: &ApiMethod,
+    rpcenv: &mut dyn RpcEnvironment,
+) -> Result<Value, Error> {
+    let env: &ReaderEnvironment = rpcenv.as_ref();
+    env.finish();
+    Ok(Value::Null)
+}
+
 /* this is too slow
 fn download_chunk_old(
     _parts: Parts,
-- 
2.39.5