| Bug #99948 | btr_attach_half_pages don't need to set prev_block or next_block | ||
|---|---|---|---|
| Submitted: | 21 Jun 2020 22:23 | Modified: | 22 Jun 2020 12:25 |
| Reporter: | Zongzhi Chen (OCA) | Email Updates: | |
| Status: | Verified | Impact on me: | |
| Category: | MySQL Server: InnoDB storage engine | Severity: | S3 (Non-critical) |
| Version: | 8.0 | OS: | Any |
| Assigned to: | CPU Architecture: | Any | |
[21 Jun 2020 22:41]
Zongzhi Chen
After runing many times of sysbech oltp_insert, there is about 2% performance optimization...
before the fix:
SQL statistics:
queries performed:
read: 0
write: 1417911
other: 0
total: 1417911
transactions: 1417911 (14172.55 per sec.)
queries: 1417911 (14172.55 per sec.)
ignored errors: 0 (0.00 per sec.)
reconnects: 0 (0.00 per sec.)
General statistics:
total time: 100.0443s
total number of events: 1417911
Latency (ms):
min: 0.13
avg: 18.06
max: 899.72
95th percentile: 108.68
sum: 25603694.01
Threads fairness:
events (avg/stddev): 5538.7148/174.79
execution time (avg/stddev): 100.0144/0.01
after the patch:
SQL statistics:
queries performed:
read: 0
write: 1442132
other: 0
total: 1442132
transactions: 1442132 (14415.49 per sec.)
queries: 1442132 (14415.49 per sec.)
ignored errors: 0 (0.00 per sec.)
reconnects: 0 (0.00 per sec.)
General statistics:
total time: 100.0384s
total number of events: 1442132
Latency (ms):
min: 0.13
avg: 17.75
max: 894.51
95th percentile: 108.68
sum: 25602354.60
Threads fairness:
events (avg/stddev): 5633.3281/183.16
execution time (avg/stddev): 100.0092/0.01
[22 Jun 2020 12:27]
MySQL Verification Team
Hi Mr. zongzhi, Thank you for your bug report. I have analysed your patch and I agree with you. Verified as reported.
Description: In btr_attach_half_pages function() It is in the middle step of InnoDB SMO. InnoDB try to link the page(p1), and new_page(p2) to page's prev_page(p3) and next_page(p4). However, if the direction is FSP_UP, it only need the set the p4's prev_page to new page(p2), and don't need to set the p3's next_page to p1, since it don't change p1. There is same waste operation when the direction is FSP_DOWN.. I suggest to remove the waste operation.. How to repeat: read the code.. Suggested fix: I have modify the code and running, even there is almost non performance optimization, at lease it running well.. @@ -2171,7 +2171,7 @@ static void btr_attach_half_pages( /* Update page links of the level */ - if (prev_block) { + if (prev_block && direction == FSP_DOWN) { #ifdef UNIV_BTR_DEBUG ut_a(page_is_comp(prev_block->frame) == page_is_comp(page)); ut_a(btr_page_get_next(prev_block->frame, mtr) == block->page.id.page_no()); @@ -2181,7 +2181,7 @@ static void btr_attach_half_pages( buf_block_get_page_zip(prev_block), lower_page_no, mtr); } - if (next_block) { + if (next_block && direction != FSP_DOWN) { #ifdef UNIV_BTR_DEBUG ut_a(page_is_comp(next_block->frame) == page_is_comp(page)); ut_a(btr_page_get_prev(next_block->frame, mtr) == page_get_page_no(page));