pingcap/tidb

Imporve the performance of writing chunks to connection.

Open

#13,052 opened on Oct 31, 2019

View on GitHub
 (1 comment) (1 reaction) (0 assignees)Go (40,090 stars) (6,186 forks)batch import
component/serverhelp wantedtype/enhancementtype/performance

Description

Feature Request

Is your feature request related to a problem? Please describe:

In https://github.com/pingcap/tidb/blob/7bfa896/server/conn.go#L1269, we have a chunk to write to connection, but we dump a chunk to mysql protocol row by row.

Describe the feature you'd like:

Now we can vectorize dumpBinaryRow and dumpTextRow by converting a chunk to network buffer column by column.

Describe alternatives you've considered:

Teachability, Documentation, Adoption, Migration Strategy:

Contributor guide

Tech stack
gosql
Domain
backenddatabaseperformance
Issue type
feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
4
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
3-5 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
GoTiDB internalsMySQL protocolperformance optimization
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
20
Research direction
The issue asks to vectorize `dumpBinaryRow` and `dumpTextRow` in `server/conn.go` at line 1269. Currently, the chunk is written row by row. The solution involves converting a chunk to network buffer column by column. To implement this, first understand the current row by row dumping logic and the Chunk structure in TiDB. Then design a columnar conversion that writes all rows for one column before moving to the next. Benchmarks should be created to measure performance improvement. No existing PRs or assignees; the issue is from 2019, so re evaluate with current codebase and discuss with maintainers.