Delayed rate limit application or throttling on writing to GCE disk?

[ryotahashimoto]

We use GCE and Zonal Standard persistent disk for etcd backend and found a strange behavior on the write IOPS performance transition. 

GCE once allowed burst access but it was shaped within minutes and the IOPS got reduced significantly. 

Neither of the peak or shaped IOPS value matches the IOPS performance spec table.  Is there any documented or undocumented rate limit or throttling mechanism for GCE zonal disks?  I could imagine region-free disks may be affected by network performance, but not sure of zonal ones.  

https://cloud.google.com/compute/docs/disks/performance#performance_limits

[siegfredv]

There are many factors that determine the performance of the block storage volumes that you attach to your virtual machine (VM) instances such the Machine type, the numbers of vCpu's or the Disk type.

To increase disk performance, you need to either resize your persistent disk or change to a higher machine. type.

You can check this documentation for other factors that affect the persistent disk's performance.

[ryotahashimoto]

@siegfredvthank you for the reply.  Unfortunately, as I also pasted the link, I've read the article and knows the upper limit described, but such an odd rate limit behavior is not described, so wanted to know if it is the spec/expected behavior or not, as we can consume more IOPS than described in the article, but TBH don't want it but need more stable access without such an unpredictable rate limit.