Since November 2018 we have performance problems with Zabbix 3.4.10. Since this time, we made tuning for tcp connections, tuning in the OS (CentOS 6.9), tuning in the MariaDB conf, tuning in the Zabbix Server conf and tuning in the Zabbix Proxies conf.

The following problem remains:

2 times a day (5:22 p.m. and 8:22 p.m., we have different hours in the past) History Syncer reaches 100% and proxies become "NoDataProcessing". While history syncer is at 100% several "NoData" alarms start to appear. To recover, we stop the 2 major proxies, restart the server and raise the 2 proxies with a time interval.

Number of hosts 6679 Number of items 732813 Number of triggers 358638 Required server performance, new values per second 2695.54

Hardwares (VM):

  • Server: CPU: 4 CPU Memory: 24GB

  • Database (Ver 15.1 Distrib 10.3.4-MariaDB, for Linux (x86_64) using readline 5.1): CPU: 8 CPUs Memory: 64GB Disk: SSD


1 Zabbix Server version 3.4.10 (lx-all-mon-zserver-01)
1 Database Ver 15.1 Distrib 10.3.4-MariaDB (lx-all-mon-zdb-03)
6 proxies for machines
1 proxy for Databases monitoring
1 proxy for webchecks

Logs, confs and graphs

migrated from superuser.com May 23 at 20:26

This question came from our site for computer enthusiasts and power users.

  • 100% of what? If CPU, then what queries were running at the time? To get to the bottom of the problem, we probably need to look at it from a MySQL/MariaDB point of view. Zabbix's output seems to be obfuscating the issue. Do you know how to connect to the database from phpmyadmin or mysql workbench or the commandline tool 'mysql'? – Rick James May 23 at 21:49
  • With the caveat that I can't access any of the files in that SharePoint link (and so probably others cannot either): Zabbix should be monitoring itself. Check its own graphs to see if a bottleneck is obvious. And post the data in the question itself, so that it remains accessible even if external links are not. – Michael Hampton May 24 at 0:16


  • My.cnf has been changed to the settings below:


[mysqld] datadir = /opt/timwe/data/mysql

innodb_log_group_home_dir = /opt/timwe/data/mysql/innodblog

log-bin = /opt/timwe/data/mysql/binlog

log_error = /opt/timwe/log/mysql_error.log

slow_query_log_file = /opt/timwe/log/slow_query.log


innodb_autoextend_increment = 256

innodb_buffer_pool_instances = 12

innodb_buffer_pool_size = 48G

innodb_concurrency_tickets = 5000

innodb_file_per_table = 1

innodb_flush_log_at_trx_commit = 2

innodb_flush_method = O_DIRECT

innodb_log_file_size = 512M

innodb_log_files_in_group = 8

innodb_old_blocks_time = 1000

innodb_open_files = 2048

innodb_stats_on_metadata = OFF

innodb_force_recovery = 0


bind-address =

default_storage_engine = innodb

event_scheduler = ON

binlog-row-event-max-size = 8192

binlog-format = MIXED

character_set_server = utf8

collation_server = utf8_bin

expire_logs_days = 1

join_buffer_size = 262144

max_allowed_packet = 32M

max_connect_errors = 10000

max_connections = 5000

max_heap_table_size = 134217728

port = 3306

query_cache_type = 1

query_cache_size = 268435456

slow-query-log = ON

table_open_cache = 2048

thread_cache_size = 512

tmp_table_size = 134217728

user = mysql

wait_timeout = 86400

[mysql_safe] log-error=/opt/timwe/log/mysqld_safe_error.log


  • Additional information request from your MariaDB 10.3.4 server. Post on pastebin.com and share the links. Text results of: B) SHOW GLOBAL STATUS; after minimum 24 hours UPTIME C) SHOW GLOBAL VARIABLES; D) SHOW FULL PROCESSLIST; E) complete MySQLTuner report F) SHOW ENGINE INNODB STATUS; G) SELECT name, count FROM information_schema.innodb_metrics ORDER BY name; for server workload tuning analysis. – Wilson Hauck Jun 27 at 0:06
  • What are the geographic locations of your involved MariaDB host and the servers you are monitoring? Country list (appx count of servers in each country) would be fine for basic info to get an idea of the scope of the effort. – Wilson Hauck Jun 27 at 0:54
  • Are you aware MariaDB 10.3.4 is a Beta version? Beta versions are not intended to be run in production. There have been many STABLE releases since.2018-01-18. Consider using 10.3.12 released 2019-01-07 as STABLE. It would have 6 months of 'traction' and you would not be considered to be an early adopter and need to find and report bugs. – Wilson Hauck Jun 27 at 16:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.