Evgeny Stanilovsky created IGNITE-23458:
-------------------------------------------
Summary: Sql. Worker node failed to send message to another worker
node
Key: IGNITE-23458
URL: https://issues.apache.org/jira/browse/IGNITE-23458
Project: Ignite
Issue Type: Improvement
Components: sql
Affects Versions: 3.0.0-beta1
Reporter: Evgeny Stanilovsky
Case: _Execution of multistage query involves communication between nodes in
order to exchange control messages and batches of intermediate results. In case
one node is unable to send a message to another node, it must inform the
coordinator in order to let the latter handle the problem. The recovery is
possible only if no rows have been returned to the client yet. To recover from
such a situation, current execution must be dropped, query must be restarted
from mapping stage. In case of RW transaction, tx itself must be restarted if
it was implicit, otherwise recovery is not possible._
>From the other side, we have assumptions:
1. Message is delivered through messaging service or logical topology event is
raised.
2. Messages need to be delivered also during temporary recipient unavailability.
Thus if message can`t be sent, appropriate error message will be send into
coordinator node and some kind of error related handling need to be implemented:
1. In case of runtime error - all execution need to be cancelled with
appropriate informative exception on the client side.
2. In case on messaging service error - coordinator need to reschedule query
execution according to logic above.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
