How to Read and Interpret Data from Azure Metric Logs

Question

There are some spikes in the IO request count from SQL Managed Instance metric logs that are confusing to interpret.

For instance, in the attached screenshot, the graph uses a 1-minute time granularity and average aggregation (Avg, Min, and Max provide the same result). At 6:13 PM, the data shows 21.74k.

Does this indicate that the total IO operations during the 1-minute interval are 21.74k? If so, when changing to a 5-minute time granularity, the spike value is 8.47k, which does not seem to represent a total of 8.47k operations within 5 minutes.

Alternatively, does this represent the average IO per second (IOPS) being 21.74k? If that's the case, the SQL Managed Instance, which has a 4 vCore limit of 16k IOPS, would need clarification on how it could reach 21.74k.

Answer 1

Hi @Ben Liu
Welcome to the Microsoft Q&A Platform! Thank you for asking your question here. 

The 21.74k value you see at 6:13 PM represents the IOPS per second for that 1 minute, not the total number of transactions. So, that means the system processed an average of about 21,740 jobs per second during those minutes. Switching to a 5 minute view and seeing 8.47k shows the IOPS per second for that 5 minute. 

Those spikes can be common, especially in short bursts of activity, and shouldn't be a concern unless they occur more frequently or take longer than expected However, if you see consistently high IOPS it's probably worth it you check to optimize queries or check that your current no service level is handling the load. 

If you frequently exceed IOPS limits, you may want to consider upgrading or upgrading to more efficient storage. Setting up alerts for high IOPS can help you catch any issues quickly.
If you have any further queries, do let us know  

---------------------------------------------------------------------------------------------------------  
If the answer is helpful, please click "Accept Answer" and "Upvote it"