Currently major version upgrades seem to be impossible without some downtime.

The objective would be sustain the following during the upgrade:

• Data collection.
• Alert sending.
• Frontend accessibility and current data visibility.
• Having all the data collected during the upgrade window after upgrade is fully completed (values, alerts, IT service data etc).

There are currently workarounds that allow achieving some of the objectives (using proxies for data collection, using a cloned instance during the upgrade to sustain alerting), but none of the methods seems to allow achieving all objectives.

A complicated method might be having a replicated DB and another Zabbix server operating during the upgrade, and afterwards manually extracting data collected during the downtime, but this is not practical.
It would involve moving over history, trends, network discovery, IT service, active agent autoregistration, LLD, auditlog and much more - and if config automation is in place, also all config changes.