Hyukjin Kwon created SPARK-61847:
------------------------------------

             Summary: Bound the TLS handshake of a direct connection by 
grpcio's 20 s connect deadline, as through a proxy
                 Key: SPARK-61847
                 URL: https://issues.apache.org/jira/browse/SPARK-61847
             Project: Spark
          Issue Type: Bug
          Components: Connect
    Affects Versions: connect-rust-4.4.0
            Reporter: Hyukjin Kwon


SPARK-61773 made a connection through an HTTP proxy bounded as a whole by 
grpcio's 20 s connect deadline: the TCP connection to the proxy, the tunnel and 
the TLS handshake inside it, with a retried UNAVAILABLE. It left the direct 
path as it was, with the same pre-existing gap: tonic's connect_lazy bounds 
only the TCP connection, so a server that accepts the TCP connection and never 
answers the TLS handshake holds the call until its own deadline (10 minutes for 
a Config request), and the process's exit waits on it too. PySpark's grpcio 
gives up after 20 s with UNAVAILABLE ("Handshake read failed"), which its retry 
policy retries.

Bound the whole direct connection, TCP and TLS, by the same 20 s, with tonic's 
own HTTP connector, keeping the 10 s bound of the TCP connection.




--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to