MoanasDaddyXu opened a new issue, #65947:
URL: https://github.com/apache/doris/issues/65947

   ### Tracking
   
   - Tracking issue: #65265
   - Internal reference: DORIS-27332
   
   ### Search before asking
   
   - [x] I searched the existing issues and found no issue reporting the same 
binlog import-performance regression.
   
   ### Version
   
   Reproduced on Doris commit `87d57a84729`.
   
   ### What's Wrong?
   
   Enabling binlog on a table causes a significant import-performance 
regression under the same workload.
   
   | Binlog | Import throughput metric | Elapsed-time metric |
   | --- | ---: | ---: |
   | Enabled | 1,068,208.99 | 2,340.366 |
   | Disabled | 3,085,916.86 | 810.132 |
   
   With binlog enabled:
   
   - Import throughput drops by approximately **65.4%**, to 34.6% of the 
binlog-disabled result.
   - Elapsed time increases to approximately **2.89x** the binlog-disabled 
result.
   
   The comparison used the same Doris build and import workload, with binlog 
enablement as the relevant variable.
   
   ### What You Expected?
   
   Enabling binlog has unavoidable write amplification, but it should not 
reduce import throughput by approximately 65% under the same workload. The 
binlog write path should be profiled and optimized, or the expected overhead 
and limiting factors should be documented.
   
   ### How to Reproduce?
   
   1. Prepare a table and a repeatable bulk-import workload.
   2. Run the workload with binlog disabled and record throughput and elapsed 
time.
   3. Recreate the table with the same schema and settings, enabling binlog as 
the relevant change.
   4. Run the identical workload and compare throughput and elapsed time.
   5. Observe that the binlog-enabled run reaches only about one third of the 
throughput and takes nearly three times as long.
   
   ### Impact
   
   The regression substantially reduces sustained ingestion capacity for 
workloads that require binlog, including downstream incremental-consumption 
scenarios.
   
   ### Are you willing to submit PR?
   
   - [ ] Yes, I am willing to submit a PR!
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to