Repository navigation
Document the new timeout and cancellation behavior #47
Description
- Soft (server-side) vs. hard (client-side) cancellation
- Cancellation timeout (in the connection string params page as well)
- Timeouts are per-operation, the cancellation token isn't stored internally.
- This means you must call Read/NextResult with the token to actually trigger the cancellation (see NpgsqlDataReader.Dispose takes a very long time when stopping early (reading the rest of the data?) npgsql#4542)
- But just one Read/NextResult might not work, since if it has the data buffered, it ignores the token entirely.
- So the guidance should always be to consume the reader until the end.
- And are not aggregated (Make timeouts aggregated npgsql#1041)
- Nothing (yet) supported in multiplexing
- Describe statement_timeout and where it's problematic, and also how to configure it on the connection string (NpgsqlRetryingExecutionStrategy with CommandTimeout induced TimeoutExceptions efcore.pg#2486 (comment))
Metadata
Metadata
Assignees
Labels
No labels
Activity
baal2000 commentedon Sep 17, 2025
@roji we've ran into the the same misunderstanding as in npgsql/npgsql#4542 (comment) of whether a "cancelled" NpgSqlReader ignores the remaining data in the pipe when disposed or reads to the end.
What stops you from updating the documentation?
roji commentedon Sep 18, 2025
Nothing, except time and priority constraints.
baal2000 commentedon Sep 18, 2025
Understand.
Just as a suggestion: when this happens, please expand on NpgsqlDataReader.Close and NpgsqlDataReader.CloseAsync by posting a remark similar to SqlDataReader.Close:
and