Hi Igniters, I started IGNITE-8580 <https://issues.apache.org/jira/browse/IGNITE-8580> ticket related to print page read/write metrics and did some investigation what other databases provide for the similar purposes.
Based on the investigation I want to propose my raw vision of how to IGNITE can be more transparent from performance perspective. Need to collect statistics for logical (from memory) and physical (from storage) page reads/writes. All these metrics should be separated by next dimensions: 1) index/cache 2) query level 3) node/cluster ... Seems the statistics should be limited by time. If we will have such statistics we could realize such things as: 1) Get IO statistics per SQL query, global or/and splitted by indexes/caches. 2) Have ability to understand why performance goes down in case it related to IO. For example on concrete node or cache. 3) Evaluate effectiveness of use indexes. Find unused indexes. 4) Keep TOP queries with aggressive physical reads .... Such statistics could be available least at JMX and SQL interfaces. Let's discuss. In case it will be interested for you I can dig deeper into the area and prepare IEP based on our discussion. Igniters, what do you think? -- Живи с улыбкой! :D
