Line data    Source code

       1              : //!
       2              : //!   WAL service listens for client connections and
       3              : //!   receive WAL from wal_proposer and send it to WAL receivers
       4              : //!
       5              : use anyhow::{Context, Result};
       6              : use postgres_backend::QueryError;
       7              : use safekeeper_api::models::ConnectionId;
       8              : use std::sync::Arc;
       9              : use std::time::Duration;
      10              : use tokio::net::TcpStream;
      11              : use tokio_io_timeout::TimeoutReader;
      12              : use tokio_util::sync::CancellationToken;
      13              : use tracing::*;
      14              : use utils::{auth::Scope, measured_stream::MeasuredStream};
      15              : 
      16              : use crate::metrics::TrafficMetrics;
      17              : use crate::SafeKeeperConf;
      18              : use crate::{handler::SafekeeperPostgresHandler, GlobalTimelines};
      19              : use postgres_backend::{AuthType, PostgresBackend};
      20              : 
      21              : /// Accept incoming TCP connections and spawn them into a background thread.
      22              : ///
      23              : /// allowed_auth_scope is either SafekeeperData (wide JWT tokens giving access
      24              : /// to any tenant are allowed) or Tenant (only tokens giving access to specific
      25              : /// tenant are allowed). Doesn't matter if auth is disabled in conf.
      26            0 : pub async fn task_main(
      27            0 :     conf: Arc<SafeKeeperConf>,
      28            0 :     pg_listener: std::net::TcpListener,
      29            0 :     allowed_auth_scope: Scope,
      30            0 :     global_timelines: Arc<GlobalTimelines>,
      31            0 : ) -> anyhow::Result<()> {
      32            0 :     // Tokio's from_std won't do this for us, per its comment.
      33            0 :     pg_listener.set_nonblocking(true)?;
      34              : 
      35            0 :     let listener = tokio::net::TcpListener::from_std(pg_listener)?;
      36            0 :     let mut connection_count: ConnectionCount = 0;
      37              : 
      38              :     loop {
      39            0 :         let (socket, peer_addr) = listener.accept().await.context("accept")?;
      40            0 :         debug!("accepted connection from {}", peer_addr);
      41            0 :         let conf = conf.clone();
      42            0 :         let conn_id = issue_connection_id(&mut connection_count);
      43            0 :         let global_timelines = global_timelines.clone();
      44            0 :         tokio::spawn(
      45            0 :             async move {
      46            0 :                 if let Err(err) = handle_socket(socket, conf, conn_id, allowed_auth_scope, global_timelines).await {
      47            0 :                     error!("connection handler exited: {}", err);
      48            0 :                 }
      49            0 :             }
      50            0 :             .instrument(info_span!("", cid = %conn_id, ttid = field::Empty, application_name = field::Empty, shard = field::Empty)),
      51              :         );
      52              :     }
      53            0 : }
      54              : 
      55              : /// This is run by `task_main` above, inside a background thread.
      56              : ///
      57            0 : async fn handle_socket(
      58            0 :     socket: TcpStream,
      59            0 :     conf: Arc<SafeKeeperConf>,
      60            0 :     conn_id: ConnectionId,
      61            0 :     allowed_auth_scope: Scope,
      62            0 :     global_timelines: Arc<GlobalTimelines>,
      63            0 : ) -> Result<(), QueryError> {
      64            0 :     socket.set_nodelay(true)?;
      65            0 :     let peer_addr = socket.peer_addr()?;
      66              : 
      67              :     // Set timeout on reading from the socket. It prevents hanged up connection
      68              :     // if client suddenly disappears. Note that TCP_KEEPALIVE is not enabled by
      69              :     // default, and tokio doesn't provide ability to set it out of the box.
      70            0 :     let mut socket = TimeoutReader::new(socket);
      71            0 :     let wal_service_timeout = Duration::from_secs(60 * 10);
      72            0 :     socket.set_timeout(Some(wal_service_timeout));
      73            0 :     // pin! is here because TimeoutReader (due to storing sleep future inside)
      74            0 :     // is not Unpin, and all pgbackend/framed/tokio dependencies require stream
      75            0 :     // to be Unpin. Which is reasonable, as indeed something like TimeoutReader
      76            0 :     // shouldn't be moved.
      77            0 :     let socket = std::pin::pin!(socket);
      78            0 : 
      79            0 :     let traffic_metrics = TrafficMetrics::new();
      80            0 :     if let Some(current_az) = conf.availability_zone.as_deref() {
      81            0 :         traffic_metrics.set_sk_az(current_az);
      82            0 :     }
      83              : 
      84            0 :     let socket = MeasuredStream::new(
      85            0 :         socket,
      86            0 :         |cnt| {
      87            0 :             traffic_metrics.observe_read(cnt);
      88            0 :         },
      89            0 :         |cnt| {
      90            0 :             traffic_metrics.observe_write(cnt);
      91            0 :         },
      92            0 :     );
      93              : 
      94            0 :     let auth_key = match allowed_auth_scope {
      95            0 :         Scope::Tenant => conf.pg_tenant_only_auth.clone(),
      96            0 :         _ => conf.pg_auth.clone(),
      97              :     };
      98            0 :     let auth_type = match auth_key {
      99            0 :         None => AuthType::Trust,
     100            0 :         Some(_) => AuthType::NeonJWT,
     101              :     };
     102            0 :     let auth_pair = auth_key.map(|key| (allowed_auth_scope, key));
     103            0 :     let mut conn_handler = SafekeeperPostgresHandler::new(
     104            0 :         conf,
     105            0 :         conn_id,
     106            0 :         Some(traffic_metrics.clone()),
     107            0 :         auth_pair,
     108            0 :         global_timelines,
     109            0 :     );
     110            0 :     let pgbackend = PostgresBackend::new_from_io(socket, peer_addr, auth_type, None)?;
     111              :     // libpq protocol between safekeeper and walproposer / pageserver
     112              :     // We don't use shutdown.
     113            0 :     pgbackend
     114            0 :         .run(&mut conn_handler, &CancellationToken::new())
     115            0 :         .await
     116            0 : }
     117              : 
     118              : pub type ConnectionCount = u32;
     119              : 
     120            0 : pub fn issue_connection_id(count: &mut ConnectionCount) -> ConnectionId {
     121            0 :     *count = count.wrapping_add(1);
     122            0 :     *count
     123            0 : }