broker icon indicating copy to clipboard operation
broker copied to clipboard

Add additional metrics

Open Neverlord opened this issue 3 years ago • 0 comments

To allow users to better understand what Broker is doing and to understand the health of the system, we need additional metrics. The list we've came up with so far:

  • Unsent messages, i.e., number of buffered messages at the core actor
  • Unsent bytes, i.e., accumulated sizes of buffered messages
  • Input and output channels of stores (a master will have one channel per clone)
  • Number of entries in the data store
  • Number of buffered commands, i.e., command messages that wait for one or more ACKs

Ideas for metrics that would need some extra clarification or are not straightforward to implement:

  • High-water metrics, e.g., largest message yet / highest amount of buffered messages?
  • Latencies: this one is tricky, because ideally this time would be measured from creating an event (in Zeek) to writing the message to a socket; however, messages currently have no timestamps

Neverlord avatar Jul 09 '22 07:07 Neverlord