Skip to content

message_data_load_latency

Type: Histogram
Buckets: 0.005, 0.01, 0.025, 0.05, 0.1, 0.25, 0.5, 1.0, 2.5, 5.0, 10.0
How many seconds it takes to load message data from spool.

High values indicate IO pressure which may be caused by policy that operates on the message body post-reception. We recommend avoiding logging header values as that is the most common cause of this metric spiking and has the biggest impact in resolving it.

IO pressure may also be alleviated by tuning other constraints and/or RocksDB Parameters

Histogram

This metric is a histogram which means that it is exported as three underlying metrics:

  • message_data_load_latency_count - a counter tracking how many events have been accumulated into the histogram
  • message_data_load_latency_sum - a counter tracking the total value of all of the events have been accumulated into the histogram
  • message_data_load_latency_bucket - a counter tracking the number of events that fall within the various buckets shown above. This counter has an additional le label that indicates the bucket threshold. For example, the first bucket for this histogram will generate a label le="0.005" which will keep track of the number of events whose value was less-or-equal (le) that value.

The recommended visualization for a histogram is a heatmap based on message_data_load_latency_bucket.

While it is possible to calculate a mean average for message_data_load_latency by computing message_data_load_latency_sum / message_data_load_latency_count, it can be difficult to reason about what that value means if the traffic patterns are not uniform since the launch of the process. We strongly recommend using a heatmap visualization instead of computing an average value.