From: Nicolas C. <za...@nr...> - 2015-11-22 09:53:54
|
Hello, I'm using Zabbix 2.4.7 with MySQL Community Server 5.6.27 on a Debian Jessie virtual machine, we currently have 740 NVPS (roughly 60 millions rows per day). We use MySQL partitioning (https://www.zabbix.org/wiki/Docs/howto/mysql_partition) : hourly for "history" and "history_uint" and daily for the others tables. We tried upgrading MySQL to 5.7.9 and we experienced very poor performances with the database : - intensive reads on the database, even when nobody was on the frontend (didn't happen with MySQL 5.6) - Dashboard very slow when the filter is disabled (a few minutes to display the page) - Peaks of "history syncer processes", 100% during several minutes. With MySQL 5.6 the average is 1% busy with the default value StartDBSyncers=4 - Drops in "free history write cache". With MySQL 5.6 and HistoryCacheSize=128M we have 99.94% free history write cache on average. All this issues happened on a Saturday with no particular activity on the storage and the server. We switched back to 5.6 and everything was back to normal. Before upgrading on the production server, we tested on a Zabbix installation with less NVPS and MySQL 5.7 was working fine. For information, we tweaked your MySQL configuration with the percona guide : max_connections = 500 sync_binlog = 0 query_cache_size = 0 query_cache_type = 0 innodb_flush_method = O_DIRECT innodb_buffer_pool_instances = 8 innodb_file_per_table = 1 innodb_buffer_pool_size = 8G innodb_flush_log_at_trx_commit = 0 innodb_io_capacity = 800 Is anyone using MySQL 5.7 as database for Zabbix ? Any idea why it works normally on a test server with 40NVPS but not on a production server with 740NVPS ? Thanks, Nicolas |