HPA Portal sections Common HPA portal components Cache usage

Calypso cache metrics

The metrics are recorded directly from Calypso CacheMetrics object for cache created using CacheUtil.makeCache.

Multiple statistics exists on the Calypso caches, here is an extract :

  • population : number of elements in the cache
  • maxSize : the max cache size, this is the limit configured in database or using the dataserver GUI.
  • readMetric : requests and _hits for read actions
  • writeMetric : requests and _hits for put actions
  • removeMetric : requests and _hits for remove actions

Example of Hpa cache metrics

Here is an example of the recorded metrics for 1 cache :

Hpa|Cache|AccountSQL._accountHash:maxSize
Hpa|Cache|AccountSQL._accountHash:population
Hpa|Cache|AccountSQL._accountHash:ReadHitRatio %
Hpa|Cache|AccountSQL._accountHash:ReadRequests
Hpa|Cache|AccountSQL._accountHash:RemoveRequests

Details for HPA cache metrics

Cache name

In dataserver webadmin, the usual display of cache is like this :

Cache usage

If we focus on Book, there is only 1 line :

  • Limit name, here Book
  • Hit Ratio, which are the attributes : hit / requests
  • Current size, which is population
  • Max size, which is maxSize

This is a direct read of the object describe previously.

What HPA displayed is like this :

Cache usage

For a given limit, here Book, there is in fact 3 cache objects :

  • _books
  • _booksPerAttribute
  • _nameTold

Future releases of HPA will displayed the limit name used for the technical cache.

In the dataserver view, only _books statistics are displayed.

Metrics

Every 5 minutes, the caches metrics are read and recorded.

HPA agents iterate over the caches and record :

  • population : value is directly recorded at each interval
  • maxSize : value is directly recorded at each interval

Absolute or per interval

The following values are recorded per interval, meaning that only the differences between the beginning and the end of the interval is recorded. This is easier to see the evolution of the value.

For example, let’s consider the hits attribute for the ReadMetric which count number a time a cache is read and contains the value for the given key.

Time Calypso value Hpa recorded value
T0 0 0
T0 + 5mn 100 100
T0 + 10mn 200 100
T0 + 15mn 200 (no activity) 0

It’s easier to follow the evolution of the metrics and setup an alert if hitRatio is recorded per interval; In this example T0 + 15mn is a period where the cache hit ratio was 0 so the alert is easy to setup.

Here are the metrics recorded per interval by the agent :

  • ReadHitRatio % : from readMetric, this is the value hit / requests for the given interval
  • ReadRequests : number of time the cache is interrogated for the given interval
  • RemoveRequests : number of removes of the cache for the given interval

They are mapped like this on the GUI :

  • Cache name
  • Hit Ratio = the average ReadHitRatio % on the selected period
  • Fill rate = compute only on GUI side - population / maxSize
  • Population = the average value of population for the selected period
  • Max. = the maximum value of maxSize for the selected period
  • Reads = sum of ReadRequests, it is the total number of read on the selected period
  • Eviction = sum of RemoveRequests, it is the total number of eviction on the selected period

Cache usage

HPA colors meaning

If the population is over 98% of the maxSize, and the cache name is orange

If an eviction happened on the period, the cache name is red.

Advice on cache management

If the hit ratio is low, consider the benefits of using cache or be sure to correctly fill it.

If the hit ratio is 100% and the population equals the max size (i.e. the fill rate is 100%), consider increasing the cache size.

If evictions are high, consider increasing the cache size.

Always ensure that the code have to use the cache and not a fresh value.

As a limit can be used for different technical caches, ensure that modifying a limit correct the right cache.