Autoscaling

Cortex autoscales your web services on a per-API basis based on your configuration.

Autoscaling replicas

min_replicas: The lower bound on how many replicas can be running for an API.

max_replicas: The upper bound on how many replicas can be running for an API.

target_replica_concurrency (default: processes_per_replica * threads_per_process): This is the desired number of in-flight requests per replica, and is the metric which the autoscaler uses to make scaling decisions.

Replica concurrency is simply how many requests have been sent to a replica and have not yet been responded to (also referred to as in-flight requests). Therefore, it includes requests which are currently being processed and requests which are waiting in the replica's queue.

The autoscaler uses this formula to determine the number of desired replicas:

desired replicas = sum(in-flight requests accross all replicas) / target_replica_concurrency

For example, setting target_replica_concurrency to processes_per_replica * threads_per_process (the default) causes the cluster to adjust the number of replicas so that on average, requests are immediately processed without waiting in a queue, and processes/threads are never idle.

max_replica_concurrency (default: 1024): This is the maximum number of in-flight requests per replica before requests are rejected with HTTP error code 503. max_replica_concurrency includes requests that are currently being processed as well as requests that are waiting in the replica's queue (a replica can actively process processes_per_replica * threads_per_process requests concurrently, and will hold any additional requests in a local queue). Decreasing max_replica_concurrency and configuring the client to retry when it receives 503 responses will improve queue fairness accross replicas by preventing requests from sitting in long queues.

window (default: 60s): The time over which to average the API wide in-flight requests (which is the sum of in-flight requests in each replica). The longer the window, the slower the autoscaler will react to changes in API wide in-flight requests, since it is averaged over the window. API wide in-flight requests is calculated every 10 seconds, so window must be a multiple of 10 seconds.

downscale_stabilization_period (default: 5m): The API will not scale below the highest recommendation made during this period. Every 10 seconds, the autoscaler makes a recommendation based on all of the other configuration parameters described here. It will then take the max of the current recommendation and all recommendations made during the downscale_stabilization_period, and use that to determine the final number of replicas to scale to. Increasing this value will cause the cluster to react more slowly to decreased traffic, and will reduce thrashing.

upscale_stabilization_period (default: 1m): The API will not scale above the lowest recommendation made during this period. Every 10 seconds, the autoscaler makes a recommendation based on all of the other configuration parameters described here. It will then take the min of the current recommendation and all recommendations made during the upscale_stabilization_period, and use that to determine the final number of replicas to scale to. Increasing this value will cause the cluster to react more slowly to increased traffic, and will reduce thrashing.

max_downscale_factor (default: 0.75): The maximum factor by which to scale down the API on a single scaling event. For example, if max_downscale_factor is 0.5 and there are 10 running replicas, the autoscaler will not recommend fewer than 5 replicas. Increasing this number will allow the cluster to shrink more quickly in response to dramatic dips in traffic.

max_upscale_factor (default: 1.5): The maximum factor by which to scale up the API on a single scaling event. For example, if max_upscale_factor is 10 and there are 5 running replicas, the autoscaler will not recommend more than 50 replicas. Increasing this number will allow the cluster to grow more quickly in response to dramatic spikes in traffic.

downscale_tolerance (default: 0.05): Any recommendation falling within this factor below the current number of replicas will not trigger a scale down event. For example, if downscale_tolerance is 0.1 and there are 20 running replicas, a recommendation of 18 or 19 replicas will not be acted on, and the API will remain at 20 replicas. Increasing this value will prevent thrashing, but setting it too high will prevent the cluster from maintaining it's optimal size.

upscale_tolerance (default: 0.05): Any recommendation falling within this factor above the current number of replicas will not trigger a scale up event. For example, if upscale_tolerance is 0.1 and there are 20 running replicas, a recommendation of 21 or 22 replicas will not be acted on, and the API will remain at 20 replicas. Increasing this value will prevent thrashing, but setting it too high will prevent the cluster from maintaining it's optimal size.

Autoscaling instances

Cortex spins up and down instances based on the aggregate resource requests of all APIs. The number of instances will be at least min_instances and no more than max_instances (configured during installation and modifiable via cortex cluster configure).

Overprovisioning

The default value for target_replica_concurrency is processes_per_replica * threads_per_process, which behaves well in many situations (see above for an explanation of how target_replica_concurrency affects autoscaling). However, if your application is sensitive to spikes in traffic or if creating new replicas takes too long (see below), you may find it helpful to maintain extra capacity to handle the increased traffic while new replicas are being created. This can be accomplished by setting target_replica_concurrency to a lower value relative to the expected replica's concurrency. The smaller target_replica_concurrency is, the more unused capacity your API will have, and the more room it will have to handle sudden increased load. The increased request rate will still trigger the autoscaler, and your API will stabilize again (maintaining the overprovisioned capacity).

For example, if you've determined that each replica in your API can handle 2 requests, you would set target_replica_concurrency to 2. In a scenario where your API is receiving 8 concurrent requests on average, the autoscaler would maintain 4 live replicas (8/2 = 4). If you wanted to overprovision by 25%, you can set target_replica_concurrency to 1.6 causing the autoscaler maintain 5 live replicas (8/1.6 = 5).

Autoscaling responsiveness

Assuming that window and upscale_stabilization_period are set to their default values (1 minute), it could take up to 2 minutes of increased traffic before an extra replica is requested. As soon as the additional replica is requested, the replica request will be visible in the output of cortex get, but the replica won't yet be running. If an extra instance is required to schedule the newly requested replica, it could take a few minutes for AWS to provision the instance (depending on the instance type), plus a few minutes for the newly provisioned instance to download your api image and for the api to initialize (via its __init__() method).

Keep these delays in mind when considering overprovisioning (see above) and when determining appropriate values for window and upscale_stabilization_period. If you want the autoscaler to react as quickly as possible, set upscale_stabilization_period and window to their minimum values (0s and 10s respectively).

If it takes a long time to initialize your API replica (i.e. install dependencies and run your predictor's __init__() function), consider building your own API image to use instead of the default image. With this approach, you can pre-download/build/install any custom dependencies and bake them into the image.