Released ZBXNEXT-7049 in:

• pre-6.0.0alpha8 32d815b3fe, 2174299, 529ad71bf25

Released ZBXNEXT-7049 in:

• pre-6.0.0alpha8 1b9ef9b140

Updated documentation:

• What's new 6.0
• Upgrade notes 6.0
• Prometheus metrics

Updated API documentation:

• Item object
• Item prototype object
• Appendix 2. Changes from 5.4 to 6.0

@Ivo Kurzemnieks

I am a member of the Chinese document translation team.

About prometheus bulk processing documents, I don't understand how one preprocessing step output can improve multi-dependency performance?
or
How does the first Prometheus mode preprocessing step improve Prometheus inspection performance?

Is there a blog to explain?
I think it may be necessary to add external links or more sentences in the official documentation to state that this is an optimization of the underlying implementation.
And give some correct practices to apply this improve.

In my guess, the improve just in one special case.

• Multi dependent items' master item is same.
• These items apply same first prometheus pattern preprocessing.
  It also about how the backend logical about preprocess process processing. 

P.S.
I found this commit about etcd metics collect. It may can be a good example.
https://git.zabbix.com/projects/ZBX/repos/zabbix/commits/9eda77c8ef4d3fb8f93a5546ec895be760fa835a#templates/app/etcd_http/template_app_etcd_http.yaml

When master item value is processed preprocessing manager spawned 'jobs' for each dependent item. Those 'jobs' were sent to different preprocessing workers and processed separately. So there was overhead in data sending (and with prometheus data this easily can reach megabytes per dependent item) and prometheus data parsing.

There were two improvements made to improve this process:
1) dependent item bulk processing
Now the master item value is sent only to one preprocessing worker
2) prometheus caching
If prometheus preprocessing step is the first (which is the common setup) then worker will parse and index it only once. Then dependent item values can be extracted by using parsed data (and in most cases also the index).

Thank you for your explanation.