Preprocessing and throttle related question

[yurtesen]

Hello,

I was looking at the new throttling preprocessing option.
https://blog.zabbix.com/zabbix-4-2-out-now/6791/

If I understand correctly the throttle option should be used on dependent items also? It is not enough that it is in main item? Is that correct?

Also, how does the discard option compare last value? Does it keep the last fetched raw data in memory or?

It is quite unclear to me in which circumstanses it is good to use it and in which circumstanses not.

For example:
* If we are using "Simple Change" on an item which is updated every 1m. Can discard with hearthbeat 1h be used after it? Because this depends on if Simple Change uses the previously fetched value or last value stored in database? The manual is not very clear?

Thanks,

[yurtesen]

splitek thank you for your response. Your examples make visualising easier. It looks like based on the metric type, the application location of discard can have different effects.

But I do not understand how the cache is used exactly. Because the preprocessor has to have cached data value from immediately previous step

1,1,2,5,3,2 - this is raw data
0,0,1,1,1,1 - true(1) if value > 1, or false(0)
0,d,1,d,d,d - discard unchanged (d - when it happens)

As you can see, it is not enough if input data is cached. The result of previous step must be cached. Similarly the final result can't be consulted because there can be steps after discard unchanged causing changes in the final value.

There is a ValueCache which is confusing and I do not understand what it does exactly https://www.zabbix.com/documentation...ms/value_cache
Documentation says
HistoryCacheSize : Size of history cache, in bytes. Shared memory size for storing history data.
ValueCacheSize - Size of history value cache, in bytes. Shared memory size for caching item history data requests.
Actually I am not sure what is the difference between history cache and history value cache..

So which cache is used exactly? and using discard unchanged is increasing memory usage per item?

[yurtesen]

splitek do you mean that value before discard is stored in ValueCache?

[yurtesen]

splitek if zabbix server has no knowledge of discarded value. How does it know if it should discard next received value or not? As I explained, in non-discard situations, the value received and the value written to database can be very different based on preprocessing steps. So even if the data was in database, it can't be relied upon.