Java app is using <1% cpu after upgrade from Ubuntu 18.04 to 20.04

Viewed 144

I'm working on a Java / MySQL app and test runs of batch processing were taking 3-4seconds on Ubuntu 18.04.

I made the upgrade to Ubuntu 20.04 last night, and kept the same mysql conf file, and now the same app is taking 1min and 45-47 seconds!!!

6.27user 0.41system 1:45.53elapsed 6%CPU (0avgtext+0avgdata 168724maxresident)k
0inputs+384outputs (0major+42056minor)pagefaults 0swaps

I noticed with lscpu that the server is now running in min freq on average. I tried adding "acpi=ht" to /etc/default/grub, and restarted the machine, to turn off power management but the processors is still running at low frequency and presumably frequency boost isn't working.

On-line CPU(s) list:             0-15
Thread(s) per core:              2
Core(s) per socket:              4
Socket(s):                       2
NUMA node(s):                    2
Vendor ID:                       GenuineIntel
CPU family:                      6
Model:                           44
Model name:                      Intel(R) Xeon(R) CPU           E5620  @ 2.40GHz
Stepping:                        2
Frequency boost:                 enabled
CPU MHz:                         1599.592
CPU max MHz:                     2401.0000
CPU min MHz:                     1600.0000
BogoMIPS:                        4799.75
Virtualization:                  VT-x
L1d cache:                       256 KiB
L1i cache:                       256 KiB
L2 cache:                        2 MiB
L3 cache:                        24 MiB

Using mysqltuner to check the server's InnoDB metrics things look OK, so it seems like this is related to the CPU settings / OS throttling or a Java configuration / code issue.

-------- InnoDB Metrics ----------------------------------------------------------------------------
[--] InnoDB is enabled.
[--] InnoDB Thread Concurrency: 8
[OK] InnoDB File per table is activated
[OK] InnoDB buffer pool / data size: 9.0G/2.6G
[OK] Ratio InnoDB log file size / InnoDB Buffer pool size: 1.0G * 2/9.0G should be equal 25%
[OK] InnoDB buffer pool instances: 9
[--] Number of InnoDB Buffer Pool Chunk : 72 for 9 Buffer Pool Instance(s)
[OK] Innodb_buffer_pool_size aligned with Innodb_buffer_pool_chunk_size & Innodb_buffer_pool_instances
[OK] InnoDB Read buffer efficiency: 99.90% (2901594 hits/ 2904524 total)
[!!] InnoDB Write Log efficiency: 75.47% (19033 hits/ 25219 total)
[OK] InnoDB log waits: 0.00% (0 waits / 6186 writes)

Running top while the app is running shows that Java is only using <1%cpu usage after startup. For a batch process ideally it'd use all available processing power.

   2113 mysql     20   0   25.7g   2.7g  36264 S   4.3  11.6   0:12.35 mysqld
    353 root      20   0       0      0      0 S   2.7   0.0   0:02.99 md0_raid5
    244 root       0 -20       0      0      0 I   0.7   0.0   0:00.98 kworker/10:1H-kblockd
    248 root       0 -20       0      0      0 I   0.7   0.0   0:00.07 kworker/8:1H-kblockd
    399 root      20   0       0      0      0 D   0.7   0.0   0:00.61 jbd2/md0-8
   2372 bias      20   0 5597164 180592  27784 S   0.7   0.7   0:07.54 java

Running stress I get full cpu usage

   4654 user      20   0    3856    100      0 R 100.0   0.0   0:13.88 stress
   4643 user      20   0    3856    100      0 R 100.0   0.0   0:13.82 stress
   4645 user      20   0  134932   3272    272 R 100.0   0.0   0:13.84 stress
   4646 user      20   0    3856    100      0 R 100.0   0.0   0:13.80 stress

The maven pom.xml file is using the same mysql connector version that the OS has installed for the mysql server 8.0.21

 39     <dependency>
 40        <groupId>mysql</groupId>
 41        <artifactId>mysql-connector-java</artifactId>
 42        <version>8.0.21</version>
 43     </dependency>

Is the java app being throttled by the OS or by the JVM?

Are there known issues with 20.04 and mysql or java?

1 Answers

The MySQL JDBC driver default behavior apparently changed when updating the OS.

The previous driver turned off transaction auto commits for Java batch updates - manually adding/running setAutoCommit(false) on the data connection dropped the run time back to normal (apparently it was running atomic transactions for 1000 some inserts).

This is a good example of why people should defensively code and add intentional code even if the default behavior/configuration is what's used in code - the default can and will change!

https://coderanch.com/t/299833/databases/Batch-update-setAutoCommit-false

Related