nerdio 6.7 rolling drain mode issue

Avatar
Marek Sokół

Nerdio 6.7: rolling drain mode, Scale in aggressiveness was introduced.

My case: I would like to have Min active host capacity 0 in HOST POOL SIZING. The goal here is to save money over night or weekend. Still Start VM on connect having enabled.

Creating rolling drain window starting 6PM with scale in aggressiveness  set to high. The goal is to quicker limit number of active session hosts, but still providing AVD access.

Unexpected result I see as bug is:

When only one session host is active, users still are working, but RAM usage is bellow scale in trigger. Scaling plan kick in, (rolling drain mode aggressiveness is high), and last running session host is about to shuttled down, and users are logged off. No (other) active session host is waiting for users.

My request: when there is last session host active, and having active users, even Scale in aggressivenes is set to high, session host should keep working, and not to be shuted down.

Additional comment. I do not want to keep all weekend one session host active.

I would like efficiently reduce number of active session host (by using high aggressive mode)

I would like to allow users working over night/weekend if they need.

1

Comments (3 comments)

Sort by
1
Avatar
Dave Grain

Hi Merek

Thank you for your submission. Just to confirm what you are observing here and what you would like the behavior to be:

 No hosts running when there is no users require to use them. Start on connect ready to switch on a host as a user may need them, at any time. Scale in aggressively after 6pm, until the next working day.

What you are finding is the last user is working but the threshold on the RAM is meaning that Scale-in is kicking in and logging the active user out of the session.

As per the definition of the 3 scale-in aggressiveness levels:

  • High Aggressiveness: Scale in aggressiveness is set to High by default, which means it is guaranteed that after business hours, hosts that have active or disconnected sessions running on them are automatically deleted or powered off to reduce capacity. After business hours, the auto-scale logic first removes the hosts that have no sessions running on them. The remaining hosts are sorted based on the least number of sessions running on them. The users with active sessions are then consolidated and moved to a single host and the other hosts are removed by auto-scale. A warning message is sent to the active session users before removing the session hosts.
  • Medium Aggressiveness: When scale in aggressiveness is set to Medium, after business hours, the scaling logic only removes the hosts that have disconnected sessions running on them. The session hosts with active sessions running on them won't be removed. In this case, the host pool is scaled in to some extent.

So what is happening is the user isn’t using enough resource on the remaining host so Scale-in logic is saying they need to be logged off.

There are a couple of options to try which may help this situations for you:

  • Adjust the scale-in aggressiveness to medium, which will still remove unused hosts from the pool, but will not kick off a user who has an active session, no matter how much resource they are consuming.
  • Reduce the RAM trigger percentage for scaling in, so that any activity on the host will result in it not being triggered.
  • Add a second trigger for CPU, which will also need to be very low to trigger a scale-in action

What I will do on this side is talk to the developers here and see if there is way to have multiple sets of scale-in aggressiveness. So set it to High until a lower threshold of hosts are left and then set to Medium. It can then be set back to High when the number of hosts increases over a threshold.

 

 

 

 

 

0
Avatar
Marek Sokół

Hi Dave,

>Adjust the scale-in aggressiveness to medium, which will still remove unused hosts from the pool, but will not kick off a user who has an active session, no matter how much resource they are consuming.
I have already medium, but this leads I have few session host active over night with single, long lasting sessions, users

 

>Reduce the RAM trigger percentage for scaling in, so that any activity on the host will result in it not being triggered.

In full load I have 20 users per session host, I I reduce RAM trigger session hosts never will be scaled down - especially this average on hostpool, rather tackled by individual session host.

Best regards

Marek


>Add a second trigger for CPU, which will also need to be very low to trigger a scale-in action

already have - but issue is the same like about RAM trigger

Best regards

Marek

0
Avatar
Dave Grain

ok, understood thanks.

Please sign in to leave a comment.