Scale Apps in Azure App Service Flashcards
1
Q
What is autoscaling?
A
- can be triggered according to a schedule or by assessing whether the system is running short on resources
- adjust available resources based on current demand with scaling IN and OUT not UP and DOWN
- efficient autoscaling ensures sufficient resources are available to handle large volumes of requests at peak times, while managing costs when demand drops
2
Q
How does autoscaling monitor app services?
A
- Detects situations where other resources are required to handle an increasing workload and ensures those resources are available before the system becomes overloaded
- Adds or removes web servers and balances load between them
- Doesnt have any effect on power of web servers, just the number of servers
3
Q
What are auto scaling rules?
A
- Specify the threshold for a metric and triggers an auto scale event when this threshold is crossed
- Need to define rules carefully, we don’t want to scale up to handle a DDoS attack
- Instead we would implement detecting and filtering of requests that occur during such an attack before they reach the service
4
Q
What are the + of autoscaling?
A
- elasticity; enables you to scale out during business hours and in during offline hours
- improved availability and fault tolerance; ensures client requests to a service wont be denied because of instance timeout or crashing
5
Q
What are the - of autoscaling?
A
- Does not increasing power of resources so any resource intensive process might need to be manually scaled up otherwise it could cause crashing
- long term growth; a web page that starts off with a few users but then grows will cause issues as autoscaling has an overhead associated with monitoring resources and determining whether to trigger a scaling event
6
Q
How do App Service Plans handle auto scaling?
A
- When the web app scales out azure starts new instances of the hardware defined by the app service plan to the app
- to prevent runaway autoscaling an app service plan has an instance limit
- more expensive plans have a higher limit
- not all pricing tiers support auto scaling
7
Q
What are the two ways to scale?
A
- Based on metric; number of HTTP requests awaiting processing for example
- Based on schedule; certain days of the week or times
- you can combine them so an instance scales out if it goes above a metric on certain days of the week
8
Q
What are the metrics you can monitor for web apps?
A
- CPU %
- Memory %
- Disk queue length
- Http queue length
- data in; bytes recieved across all instances
- data out; bytes sent across all instances
9
Q
What is the first step of autoscaling?
A
- An auto-scale rule aggregates the values retrieved for a metric for all instances across a period of time known as the time grain
- each metric has its own time grain
- this aggregated value is the time aggregation
10
Q
What is the second step of autoscaling?
A
- Performs further aggregation of the value calculated by the first step over a longer, user-specified period known as the duration
- min value for this is 5 mins
- for example; if its set to 10 mins and time grain of metrics is 1 min then it will aggregate the 10 values collected
11
Q
What is the third step of autoscaling?
A
- calculated value for the duration can be different from the time grain
- e.g. if the time aggregation is Average and the stats gathered is CPU % across 1 min time grain, each min the avg CPU % use across all instances for that min is calculated
- if the time grain is set to Max, and the duration of the rule is set to 10 mins, the max of the 10 avg values for the CPU % usage is used to determine whether the rule threshold has been crossed
12
Q
Describe an autoscale action?
A
- can be sale out or scale in
- uses operators such as >,< or =
- has a cooldown period specified in mins
- during the cooldown the rule wont be triggered, allowing system to stabilise
13
Q
Why should autoscale rules be defined in pairs?
A
- one is used for how the system scale up when it exceeds the threshold and one for vise versa
14
Q
Can an autoscale condition have more than one autoscale rule?
A
- yes
- they dont have to be related
- one could be for HTTP queue length and the other for CPU %
- on scale out autoscale runs if any rule is met, on scale in all rules must be met
15
Q
Where can auto scaling settings be found?
A
- App Service Plan -> Scale Out pane
- by default only implements manual scaling
- has a run history chart to show how the number of instances vary over time and which conditions caused each change
