[MDEV-17845] Extreme high open file limit used Created: 2018-11-27 Updated: 2020-08-25 Resolved: 2019-05-07 |
|
| Status: | Closed |
| Project: | MariaDB Server |
| Component/s: | Server |
| Affects Version/s: | 10.3.10, 10.3, 10.4 |
| Fix Version/s: | 10.3.15, 10.4.5 |
| Type: | Bug | Priority: | Major |
| Reporter: | Minh | Assignee: | Sergei Golubchik |
| Resolution: | Fixed | Votes: | 1 |
| Labels: | None | ||
| Environment: |
CentOs 7.4, xenforo 1.5.x |
||
| Description |
|
Recently i upgraded to MariaDB 10.3.10 using directadmin custombuild function. And i noticed the mysql performance not very good. and while checking mysqltuner.pl suggestion, i found the number open file limit used extremely high. -------- Storage Engine Statistics ----------------------------------------------------------------- -------- Analysis Performance Metrics -------------------------------------------------------------- -------- Security Recommendations ------------------------------------------------------------------ -------- CVE Security Recommendations -------------------------------------------------------------- -------- Performance Metrics ----------------------------------------------------------------------- -------- Performance schema ------------------------------------------------------------------------ -------- ThreadPool Metrics ------------------------------------------------------------------------ -------- MyISAM Metrics ---------------------------------------------------------------------------- -------- InnoDB Metrics ---------------------------------------------------------------------------- -------- AriaDB Metrics ---------------------------------------------------------------------------- -------- TokuDB Metrics ---------------------------------------------------------------------------- -------- XtraDB Metrics ---------------------------------------------------------------------------- -------- Galera Metrics ---------------------------------------------------------------------------- -------- Replication Metrics ----------------------------------------------------------------------- -------- Recommendations --------------------------------------------------------------------------- |
| Comments |
| Comment by Minh [ 2018-11-27 ] | |||||||||||||||||||||||||||||||||||||||
|
10.3.11 also similar issue. | |||||||||||||||||||||||||||||||||||||||
| Comment by Elena Stepanova [ 2018-11-28 ] | |||||||||||||||||||||||||||||||||||||||
|
When you connect to the server and run show global status like 'Open_files', what do you see? | |||||||||||||||||||||||||||||||||||||||
| Comment by Jarkko Uitti [ 2019-01-11 ] | |||||||||||||||||||||||||||||||||||||||
|
I am running 10.3.11 and have this issue too. show global status like 'open_files';
Rebooted db today and open_files was 40 minutes normal. Then again open_files jumped to extremely high. OS: Ubuntu 16.04.5 LTS | |||||||||||||||||||||||||||||||||||||||
| Comment by Elena Stepanova [ 2019-01-11 ] | |||||||||||||||||||||||||||||||||||||||
|
After it jumps up, does it ever change back to normal (without server restart), or does it stay this way? | |||||||||||||||||||||||||||||||||||||||
| Comment by Jarkko Uitti [ 2019-01-12 ] | |||||||||||||||||||||||||||||||||||||||
|
I am monitoring db status by nagios. Here is some statistics: Now it's
flush status doesn't have any effect. | |||||||||||||||||||||||||||||||||||||||
| Comment by Elena Stepanova [ 2019-01-12 ] | |||||||||||||||||||||||||||||||||||||||
|
Thank you. | |||||||||||||||||||||||||||||||||||||||
| Comment by Elena Stepanova [ 2019-01-14 ] | |||||||||||||||||||||||||||||||||||||||
|
ATTENTION: in the test case below, a wrong numeric comparison is used, due to a known bug in MTR: MDEV-18228.
Modify --run value at the beginning of the test as needed. It represents the number of queries which each thread will run. The problem appears as this:
Reproducible on 10.3 and 10.4. Couldn't reproduce on 10.2. Here is an alternative RQG test (modify the path to basedir as needed):
or, for an already running server,
(modify port if needed and make sure the server is running in shm, it's much easier reproducible there). With the RQG test, the problem represents like this:
| |||||||||||||||||||||||||||||||||||||||
| Comment by Sergei Golubchik [ 2019-05-04 ] | |||||||||||||||||||||||||||||||||||||||
|
please, review https://github.com/MariaDB/server/commit/a7548c58a57 | |||||||||||||||||||||||||||||||||||||||
| Comment by Sergey Vojtovich [ 2019-05-04 ] | |||||||||||||||||||||||||||||||||||||||
|
Ok to push. |