If proxies are unlimited, why are trappers and syncers limited?

[docstupid]

This might seem to be a stupid question, but...

Given an architecture with few hosts at each autonomous systems, but with thousands of autonomous systems -- would it not make sense to allow more trapper threads on the central zabbix application server (given thousands of proxies) and syncer processes? At present, these seem limited to 1000K. Given a well configured and beefy configuration, this seems like an unnecessary choke point.

[kloczek]

This might seem to be a stupid question, but...

Given an architecture with few hosts at each autonomous systems, but with thousands of autonomous systems -- would it not make sense to allow more trapper threads on the central zabbix application server (given thousands of proxies) and syncer processes? At present, these seem limited to 1000K. Given a well configured and beefy configuration, this seems like an unnecessary choke point.

One sec .. do you want to say that in your monitoring stack you have spawned 1mln trapper processes to handle 1mln proxies???

Usually ratio between number of used proxes and number of trapper proceesses running on the server coud be at least 1:20 if not even 1:100 sometimes and it should be OK.

Other thing is that zabbix is not using threads but forked processes.

Yet another thing is that if you want to hold larger buffer of the data when server is down/unricheable you can tune agent to hold that data to up to 1h before it will start dropping oldes still not send data points (look on BufferSend and BufferSize agent configuration parameters).
Usually tune BufferSend and BufferSize is enough to not be forcet use proxies to have addional buffering to in case of some frequent disconnections.