In relation to dashboards, it’s desirable that data always seems to available when needed so that the dashboards render immediately rather then the need to reload the data, ie that it wasn’t in the cache.
Issues which can asset this are that when application service was restarting due to running out of memory, this was addressed by two fixes.
1. The Default Azure setting is restart to the Web service if memory usage is able 80% for more than 30 seconds. This was a recent enough change that they made and the mode is easily switched off. See here
2. A new mode in the scheduler which when data. marting data would release the in memory cache so that the amount of memory used in reduced. Just make sure that the dataset has no views which are cached and this behavior will be the default.
Unlike the Redis deployments, when the web service restarts it loses all it cache, so a restart will mean that dashboard data is not in memory, so these two settings force the service to release data just for dashboards on start-up.
Recently added was the facility to view what was in the cache. Unfortunately essentially operated like a quantum observation, as it needed to stream the dataset object to find out how much memory was used and therefore negatively affected the performance. This is now been changes so that to generates an estimate on memory usage.
You can examine the content of the cache to identify memory issues using this feature
The scheduler allows 10 concurrent jobs to be processed at any one time. Long running queries might mean that a larger number of concurrent jobs might be more applicable. This is controlled by this setting