[YSQL] Batch writes (inserts / updates / deletes) #2934
Comments
d-uspenskiy
added a commit
that referenced
this issue
Feb 10, 2020
Summary: Enable buffering of write operation in each SQL statement. 1. Write operation buffering is started at the beginning of each statement (StartTransactionCommand function). Buffered operations are flushed at the successful end of statement (CommitTransactionCommand function). In case transaction is aborted nothing happens with buffered operations. They will be silently dropped at the beginning of next statement (on buffering start). 2. All buffered operations also will be flushed on each attempt to perform non-bufferable operation (operations which result required immediately). They are all read operations and write operations with `force_non_bufferable` flag set to true. 3. For now DocDB can't process multiple write operations related to same row id in single RPC. Second operation will not see the results of first one. As a result it is possible to insert multiple non-unique values into column with unique constraint. Now it is handled by flashing all buffered operations before processing new one in case it relates to already affected row id. 4. Buffered operation can produce `duplicate key value violates unique constraint` error. Description of this message required constraint name. For this purpose each operation stores postgres relation id. Also interface with `C` callbacks is introduced. `PgSession` object calls `FetchUniqueConstraintName` callback with relation id to get violated constraint name. 5. Buffered operations can produce ERROR description with irrelevant CONTEXT part. Separate issue #3528 is created for this problem. For now printing of CONTEXT part is suppressed. Test Plan: New test case is implemented ``` ./yb_build.sh --java-test org.yb.pgsql.TestOperationBatching ``` Reviewers: mikhail, neha, alex Reviewed By: neha, alex Subscribers: neha, alex, yql Differential Revision: https://phabricator.dev.yugabyte.com/D7782
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Today, when we have an update/delete statement like:
update foo set x=value where conditionor
delete from foo where x=valuewe first read all rows that satisfy the condition and then send 1 update/delete tserver write request per row. We should batch those write requests to make it more performant.
The text was updated successfully, but these errors were encountered: