It can be useful to determine bottlenecks in Zabbix server by implementing per process commands prof_enable and prof_disable that will enable profiling of mutexes such as how long mutex was held and how long waited to acquire it as well as profiling of how much time was spent for history sync in total and how much time out of it was spent on inserting history into database, thus can determine how long waited on locks and how long worked with database, profiling of other places can be added in future a well.

Example use case to determine that system information was holding locks for too long and did not allow history syncers to process history:
ZBX-21537

Patch:
ZBX-19307-5.0-profiling-6.diff

If profiling is enabled for process then it will write results every 30 seconds to log file until disabled, profiling will be as lightweight as possible by using pointer to constant string in a vector.