New to Mongo Atlas. Is there a way to set the auto scaling (say M20 to M30) to happen in less than an hour? We have a transactional based system that uses Mongo, and an hour of CPU issues would be an issue for the system. Ideally I’d like to set something that if it spikes to 95% say 3 times in 5 minutes it scales up, then if it’s quiet for 2 hours is scales back down.
2 Likes
Hi Dennis! I’m a product manager on the Atlas team here at MongoDB. First of all, welcome to the platform - we’re glad to have you. For context, there is no way to configure how long it takes to auto-scale because Atlas handles all the resources on the backend and scales up as quickly as possible. Our system checks for 75%+ CPU and RAM utilization within an hour before scaling up, and 50%- CPU and RAM utilization within 24 hours before scaling down.
You can read more about the specifics of how auto-scaling works here: https://www.mongodb.com/docs/atlas/cluster-autoscaling/
However, I’d love to better understand your use case for your application as we’re always working on improving our auto-scaling functionality. If you’re open to it, send me an email at lori.berenberg@mongodb.com so we can learn more and ensure your feedback gets surfaced.
2 Likes
Hi Lori,
We are using Mongo for the back end of a OLTP system that is 24/7. If the system needs a resource increase, waiting an hour to get those resources via automation could mean the system is overwhelmed until it scales up, so having the ability to set a faster scale up time would be beneficial in our scenario.
1 Like
Thanks Dennis! Would you say that the time to trigger is more important than the actual time it takes to scale? And what is your tolerance for unnecessary scaling as a result of false bursts in activity? (i.e. a 5 minute burst that immediately goes back down within 10 mins)
1 Like
I’d think that for OLTP if it sees a sustained bust of X CPU or memory % (settable, but I’d think 85% would be what we would set) 2 minutes start the scaling and maintain it until there is an hour of activity that is reduced enough to trigger a scale down.
2 Likes
Hi. No only I couldn’t agree more with the request to scale up way faster than an hour but also I’m having an issue where “steal cpu” is compromising the stability of our system and Auto-Scaling functionality is not actually scaling up properly, so the cluster simply stays unresponsive.
What we need is the ability to configure custom rules based on alerts for auto-scaling. Just like in AWS auto-scaling.
1 Like
Hi Dennis and Tomas,
What we need is the ability to configure custom rules based on alerts for auto-scaling. Just like in AWS auto-scaling.
Firstly, thank you for providing your feedback regarding the current auto-scaling situations you are experiencing for your environments.
Although the ability to configure custom auto-scaling policies is not available, there is currently a feedback post regarding the configuration of the duration for how the auto-scaling is evaluated / monitored in which you can vote for but I do understand perhaps you would also like the ability to also change the actual CPU/Memory utilisation percentage as well.
If you would like, you could also create another feedback post specifically regarding customisable averages for the CPU/Memory percentages (rather than the duration) over the rolling period where others and yourself can vote for.
I have raised this feedback internally as well in hopes that there are some improvements in future for auto-scaling.
Regards,
Jason
2 Likes
This topic was automatically closed 5 days after the last reply. New replies are no longer allowed.