Sometimes replica may be readonly for a long time, and this will make
some INSERT queries fail, but it does not make sense to INSERT into
readonly replica, so let's ignore them.
But note, that this will require to extend TableStatus (not extend, but
introduce new version), that will have is_readonly field.
Also before background INSERT into Distributed does not uses
getManyChecked() which means that they do not request TableStatus
packet, while now they would, though this is minor (just a note).
v2: Add a note about max_replica_delay_for_distributed_queries for INSERT
v3: Skip read-only replicas for async INSERT into Distributed
v4: Remove extra @insert parameter for ConnectionPool*::get*
It make sense only when the table name had passed --
ConnectionPoolWithFailover::getManyChecked()
v5: rebase on top LoggerPtr
v6: rebase
v7: rebase
v8: move TryResult::is_readonly into the end
Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
Right now the receive_timeout for INSERT works as a timeout for
receiving data block, however this is not very convenient, since
sometimes server may not send data for quite some time (i.e. due to in
order aggregation), Ping packets is there for a reason (also Progress
and ProfileEvents as well, though the purpose is different).
Initially this special handling of receive_timeout had been added in
6a5ef9be83 ("dbms: fixed error with
hanging INSERTs [#METR-16514]"), but the behaviour was different, since
that time the receivePacket() was outside loop, only poll() was there,
and that was the workaround for poll() timeout (which does not triggers
the socket timeout).
But in fabd7193bd ("Code cleanups and
improvements"), receivePacket() had been moved into the loop, and so
this changed the behaviour of the timeout to current one.
Though all of this will not help for INSERT queries anyway, since there
are no Ping packets for them. Yet.
Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
