Exchange receiver decode optimization to do squashing work at the same time#6202
Conversation
Signed-off-by: yibin <huyibin@pingcap.com>
Signed-off-by: yibin <huyibin@pingcap.com>
Signed-off-by: yibin <huyibin@pingcap.com>
|
[REVIEW NOTIFICATION] This pull request has been approved by:
To complete the pull request process, please ask the reviewers in the list to review by filling The full list of commands accepted by this bot can be found here. DetailsReviewer can indicate their review by submitting an approval review. |
Signed-off-by: yibin <huyibin@pingcap.com>
Signed-off-by: yibin <huyibin@pingcap.com>
|
Fine-grained partitioning will exacerbate the chunk fragmentation in exchange and would benefit from this pr much more from normal cases. |
Yeah, and for large-scale clusters, it would benefit also. |
| codec.readColumnMeta(i, istr, column); | ||
|
|
||
| /// Data | ||
| MutableColumnPtr read_column = column.type->createColumn(); |
There was a problem hiding this comment.
Looks like read_column is not used?
|
|
||
| #include <Flash/Coprocessor/StreamingDAGResponseWriter.cpp> | ||
| #include <Flash/Mpp/BroadcastOrPassThroughWriter.cpp> | ||
| #include <Flash/Mpp/ExchangeReceiver.cpp> |
Signed-off-by: yibin <huyibin@pingcap.com>
|
/run-unit-tests |
Signed-off-by: yibin <huyibin@pingcap.com>
Signed-off-by: yibin <huyibin@pingcap.com>
|
/run-unit-tests |
|
/run-integration-test |
| /// Data | ||
| MutableColumnPtr read_column = column.type->createColumn(); | ||
| if (reserve_size > 0) | ||
| read_column->reserve(reserve_size); |
There was a problem hiding this comment.
reserve rows if reserve_size <= 0 and if reserve_size >0, reserve std::max(rows, reserve_size) ?
Signed-off-by: yibin <huyibin@pingcap.com>
Signed-off-by: yibin <huyibin@pingcap.com>
Signed-off-by: yibin <huyibin@pingcap.com>
Signed-off-by: yibin <huyibin@pingcap.com>
Signed-off-by: yibin <huyibin@pingcap.com>
|
/merge |
|
@yibin87: It seems you want to merge this PR, I will help you trigger all the tests: /run-all-tests You only need to trigger If you have any questions about the PR merge process, please refer to pr process. DetailsInstructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the ti-community-infra/tichi repository. |
|
This pull request has been accepted and is ready to merge. DetailsCommit hash: e8594cf |
|
/type ehencement |
|
@JasonWu0506: The label(s) DetailsIn response to this:
Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the ti-community-infra/tichi repository. |
|
/type enhencement |
|
@JasonWu0506: The label(s) DetailsIn response to this:
Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the ti-community-infra/tichi repository. |
What problem does this PR solve?
Issue Number: close #6157
Problem Summary:
What is changed and how it works?
Previously, TiFlash adds a Squash transform after ExchangeReceiver, because the block output by ExchangeReceiver might be too small and not efficient to handle. However, the two-stage solution would introduce too many memory operations, like column allocations and de-allocations. So in this PR, the one-stage solution is provided.
For a simple local test, with 10 integer columns, 1024 rows per block, performance improves from: 140ms => 60ms.
Check List
Tests
Side effects
Documentation
Release note