You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Today, we have a per-tablet timeout of 2.5 seconds. When we batch load data in SQL, we hit this timeout frequently. We should look into increasing this limit for batch loads. One option could be to use the client timeout (or some percentage of the client timeout) as the per-tablet limit.
The text was updated successfully, but these errors were encountered:
…f the fixed 60 secs and 2.5 seconds hardcoded in our code.
Summary:
Currently we have a 60 second limit for the top level queries and each
rpc uses the 2.5 second timeout.
We also have the STATEMENT_TIMEOUT for the SQL statements.
Instead of using the 2.5 second timeout for the rpcs we should ideally
use rpc timeout to be the min{STATEMENT_TIMEOUT,
FLAGS_client_read_write_timeout_ms}
This involves 2 changes:
1. Flowing the user defined timeout down to the rpc layer.
2. Using the deadline calculated in the rpc layer for all the requests.
Test Plan:
Manually checked that the rpc layer has the default timeout of 60 seconds.
When the timeout is changed with "set statement_timeout=30", the rpc layer used
the new timeout.
Added a new test TestPgTimeout to test the fact that the timeouts are adhered to.
Reviewers: mihnea, neha, bogdan, sergei
Reviewed By: sergei
Subscribers: yql
Differential Revision: https://phabricator.dev.yugabyte.com/D8374
Today, we have a per-tablet timeout of 2.5 seconds. When we batch load data in SQL, we hit this timeout frequently. We should look into increasing this limit for batch loads. One option could be to use the client timeout (or some percentage of the client timeout) as the per-tablet limit.
The text was updated successfully, but these errors were encountered: