| Bug #36941 | Performance problem in ha_print_info (SHOW INNODB STATUS) | ||
|---|---|---|---|
| Submitted: | 25 May 2008 13:58 | Modified: | 16 Sep 2008 6:51 |
| Reporter: | Domas Mituzas | ||
| Status: | Closed | ||
| Category: | Server: InnoDB | Severity: | S3 (Non-critical) |
| Version: | 5.0-bk, 5.1-bk | OS: | Any |
| Assigned to: | Timothy Smith | Target Version: | 5.0+ |
| Tags: | regression, performance, innodb | ||
| Triage: | D3 (Medium) | ||
[25 May 2008 14:28]
Domas Mituzas
Opened an offspring at Bug#36942
[25 Jun 2008 15:26]
Heikki Tuuri
Vasil,
you could this inside #ifdef UNIV_DEBUG:
ha_print_info() in 5.1:
"
for (i = 0; i < hash_get_n_cells(table); i++) {
cell = hash_get_nth_cell(table, i);
if (cell->node) {
cells++;
}
}
fprintf(file,
"Hash table size %lu, used cells %lu",
(ulong) hash_get_n_cells(table), (ulong) cells);
"
in versions 6.0 and greater. I think 5.0 as the 'old stable version' should be left as
is.
Regards,
Heikki
[26 Jun 2008 9:14]
Vasil Dimov
Patch for 5.1 created, waiting for approval. Will not be fixed in 5.0.
[26 Jun 2008 16:20]
Vasil Dimov
The fix removes ", used cells %lu" from SHOW INNODB STATUS
[21 Aug 2008 0:23]
Bugs System
A patch for this bug has been committed. After review, it may be pushed to the relevant source trees for release in the next version. You can access the patch from: http://lists.mysql.com/commits/52096 2715 Timothy Smith 2008-08-20 Cherry-pick some changes from innodb-5.1-ss2545 snapshot. Includes fixes for Bug#37531, Bug#36941, Bug#36941, Bug#36942, Bug#38185. Also include test case from Bug 34300 which was left out from earlier snapshot (5.1-ss2387). Also include fix for Bug #29507, "TRUNCATE shows to many rows effected", since the fix for Bug 37531 depends on it.
[21 Aug 2008 19:59]
Bugs System
Pushed into 5.1.28 (revid:timothy.smith@sun.com-20080820221833-brmru9b77ddt3tfx) (version source revid:azundris@mysql.com-20080821081500-f2d61fh4u61owz3p) (pib:3)
[22 Aug 2008 1:40]
Bugs System
A patch for this bug has been committed. After review, it may be pushed to the relevant source trees for release in the next version. You can access the patch from: http://lists.mysql.com/commits/52267 2794 Timothy Smith 2008-08-21 [merge] Merge up from 5.1-bugteam. Among other things, this applies some changes from the innodb-5.1-ss2545 snapshot into 6.0, fixing Bug#37531, Bug#36941, Bug#36942, and Bug#38185.
[22 Aug 2008 1:42]
Bugs System
A patch for this bug has been committed. After review, it may be pushed to the relevant source trees for release in the next version. You can access the patch from: http://lists.mysql.com/commits/52268 2794 Timothy Smith 2008-08-21 [merge] Merge up from 5.1-bugteam. Among other things, this applies some changes from the innodb-5.1-ss2545 snapshot into 6.0, fixing Bug#37531, Bug#36941, Bug#36942, and Bug#38185.
[26 Aug 2008 21:49]
Paul DuBois
Noted in 5.1.28 changelog. Some performance problems of SHOW ENGINE INNODB STATUS were reduced by removing the "used cells" and "Total number of lock structs in row lock hash table" clauses from the output. These are now present only if UNIV_DEBUG is defined at MySQL build time. Setting report to NDI pending push into 6.0.x.
[14 Sep 2008 1:42]
Bugs System
Pushed into 6.0.7-alpha (revid:timothy.smith@sun.com-20080820221833-brmru9b77ddt3tfx) (version source revid:john.embretsen@sun.com-20080808091208-ht48kyzsk7rim74g) (pib:3)
[15 Sep 2008 23:10]
James Day
I've filed bug #39462 about the removal of this output without following the deprecation process, suggesting using zero values instead of complete removal.
[16 Sep 2008 6:51]
Paul DuBois
Noted in 6.0.7 changelog.
Description: There's performance hotspot in 'SHOW INNODB STATUS', ha_print_info() has this code: for (i = 0; i < hash_get_n_cells(table); i++) { cell = hash_get_nth_cell(table, i); if (cell->node) { cells++; } } It is used to show how many nodes are free/used in such line: Hash table size 3112859, used cells 1050451, node heap has 1890 buffer(s) With freshly started server and >10GB buffer pool this code takes 0.3s, and makes 'SHOW INNODB STATUS' unusable in high load environments. How to repeat: run 'show innodb status' on modern database machine Suggested fix: either maintain aggregate value, or strip this from status output. I chose latter.