I am in the process of migrating an existing Spring Boot app to GAE. The app is automatically scaled having a minimum of 0 instances. During my test runs I have noticed that initially the first instance is loaded and starts to handle requests. As auto-scaling kicks in, a second instance is launched. The traffic after the second instance is up and running, are all being sent to that second instance. Meantime, the first instance remains idle after having run the first handful of the request that prompted the auto scale.
I'm fairly new at tuning, I'm probably missing something.
app.yaml:
runtime: java17
instance_class: F2
automatic_scaling:
target_cpu_utilization: 0.8
min_instances: 0
max_instances: 4
min_pending_latency: 8s
max_pending_latency: 10s
max_concurrent_requests: 25
env_variables:
SPRING_PROFILES_ACTIVE: "prod"
This is the image of both instances after handling 100 requests in 5 minutes. the majority of the requests are being routed to the second instance that was autoscaled.
I have tried changing some of the autoscale values. I am expecting the GAE load balancer to distribute the load among all available instances. But I don't see an even distribution of traffic. Why is that?
Related
My app engine yaml file is somewhat like below
service: servicename
runtime: php74
automatic_scaling:
min_idle_instances: 2
max_pending_latency: 1s
env_variables:
CLOUD_SQL_CONNECTION_NAME: <MY-PROJECT>:<INSTANCE-REGION>:<MY-DATABASE>
DB_USER: my-db-user
DB_PASS: my-db-pass
DB_NAME: my-db
---
automatic scaling cause higher cost? what is the cheapest configuration I can set. it's not mandatory to have auto scaling at current stage of my application.
I think your cheapest configuration is just setting max_instances: 1 and commenting out the other options.
When you have traffic, the maximum number of instances that you will have will be 1. When there's no traffic, your instance goes down (effectively 0).
The downside with this approach (not having min_idle_instance as you currently do) is that brand new traffic to your site will take some time because of the time for your instance to be started.
Some weeks ago my app on App Engine just started to increase the number of idle instances to an unreasonable high amount, even when there is close to zero traffic. This of course impacts my bill which is skyrocketing.
My app is simple Node.js application serving a GraphQL API that connects to my CloudSQL database.
Why are all these idle instances being started?
My app.yaml:
runtime: nodejs12
service: default
handlers:
- url: /.*
script: auto
secure: always
redirect_http_response_code: 301
automatic_scaling:
max_idle_instances: 1
Screenshot of monitoring:
This is very strange behavior, as per the documentation it should only temporarily exceed the max_idle_instances.
Note: When settling back to normal levels after a load spike, the
number of idle instances can temporarily exceed your specified
maximum. However, you will not be charged for more instances than the
maximum number you've specified.
Some possible solutions:
Confirm in the console that the actual app.yaml configuration is the same as in the app engine console.
Set min_idle_instances to 1 and max_idle_instances to 2 (temporarily) and redeploy the application. It could be that there is just something wrong on the scaling side, and redeploying the application could solve this.
Check your logging (filter app engine) if there is any problem in shutting down the idle instances.
Finally, you could tweak settings like max_pending_latency. I have seen people build applications that take 2-3 seconds to start up, while the default is 30ms before another instance is being spun up.
This post suggests setting the following, which you could try:
instance_class: F1
automatic_scaling:
max_idle_instances: 1 # default value
min_pending_latency: automatic # default value
max_pending_latency: 30ms
Switch to basic_scaling, let Google determine the best scaling algorithm (last resort option). This would look something like this:
basic_scaling:
max_instances: 5
idle_timeout: 15m
The solution could of course also be a combination of 2 and 4.
Update after 24 hours:
I followed #Nebulastic suggestions, number 2 and 4, but it did not make any difference. So in frustration I disabled the entire Google App Engine (App Engine > Settings > Disable application) and left it off for 10 minutes and confirmed in the monitoring dashboard that everything was dead (sorry, users!).
After 10 minutes I enabled App Engine again and it booted only 1 instance. I've been monitoring it closely since and it seems (finally) to be good now. And now after the restart it also adheres to the "min" and "max" idle instances configuration - the suggestion from #Nebulastic. Thanks!
Screenshots:
Have you checked to make sure you dont have a bunch of old versions still running? https://console.cloud.google.com/appengine/versions
check for each service in the services dropdown
I am noticing something a little odd with Google App Engine. If my app has not been used and I go open it I notice that it takes some time to load, I also see in the GAE logs console that it is starting up a server during this time so that accounts for the wait (why not always have an instance running?)
After I open and close the app a couple of times I then notice in the versions tab of GAE that I have 7 running instances (all in the same version).
Im a little confused how GAE works, does it roll down your instances to 0 when there is no requests for a while and then on the flip side, does it spin up a new instance for every new client connecting ?
my app.yaml is looking like this:
runtime: nodejs10
env: standard
instance_class: F2
handlers:
- url: /.*
secure: always
redirect_http_response_code: 301
script: auto
You need to fine tune your App Engine scaling strategy, for example please check this app.yaml file
runtime: nodejs10
env: standard
instance_class: F2
handlers:
- url: /.*
secure: always
redirect_http_response_code: 301
script: auto
automatic_scaling:
min_instances: 1
max_instances: 4
min_idle_instances: 1
max_concurrent_requests: 25
target_throughput_utilization: 0.8
inbound_services:
- warmup
min_instances & min_idle_instances are set to 1 in order to have almost 1 instance ready for incoming requests and avoid cold start.
To avoid spin up new instances too fast, you can set max_concurrent_requests & target_throughput_utilization, in this example a new instance will be spin up until an instance reaches 20 concurrent requests (25 X 0.8)
As is mentioned in this document, it is necessary create a warmup endpoint in your application and add inbound_services in your app.yaml file, for example:
app.get('/_ah/warmup', (req, res) => {
// Handle your warmup logic. Initiate db connection, etc.
});
warmup calls carry the benefit of prepare your instances before an incoming request and reduce the latency of first request.
As you did not specify any scaling setting in your app.yaml, App Engine is using automatic scaling.
That means that the application has 0 minimum instances so when your app is not receiving any request at all it will scale down to 0. With that option you will sabve the costs that imply having an instance running all the time, but also cold starts will happen. A cold start happens each time a request reaches your application but there are no instances ready to serve it and a new one has to be created.
Regarding your application scaling up to 7 instances when the traffic load increases, it depends again on the workload that is receiving. You can control this behaviour as well by using the max_instances setting, although using a low value could affect your application's performance if more instances are needed.
App Engine will be spinning up new instances if the threshold value on target_cpu_utilization, target_throughput_utilization , max_concurrent_requests, max_pending_latency or min_pending_latency is reached. You can read about all of them here.
I have a Java Application in App Engine, and recently I started getting following error:
Unable to export to Monitering service because: GaxError RPC failed, caused by 3:One or more TimeSeries could not be written: Metrics cannot be written to gae_app. See https://cloud.google.com/monitoring/custom-metrics/creating-metrics#which-resource for a list of writable resource types.: timeSeries[0]
and this happens every time after health check log:
Health checks: instance=instanceName start=2020-01-14T14:28:07+00:00 end=2020-01-14T14:28:53+00:00 total=18 unhealthy=0 healthy=18
and after some time my instances would be restarted and the same thing starts to happen again.
app.yaml:
#https://cloud.google.com/appengine/docs/flexible/java/reference/app-yaml
#General settings
runtime: java
api_version: '1.0'
env: flex
runtime_config:
jdk: openjdk8
#service: service_name #Required if creating a service. Optional for the default service.
#https://cloud.google.com/compute/docs/machine-types
#Resource settings
resources:
cpu: 2
memory_gb: 6 #memory_gb = cpu * [0.9 - 6.5] - 0.4
# disk_size_gb: 10 #default
##Liveness checks - Liveness checks confirm that the VM and the Docker container are running. Instances that are deemed unhealthy are restarted.
liveness_check:
path: "/liveness_check"
timeout_sec: 20 #1-300 Timeout interval for each request, in seconds.
check_interval_sec: 30 #1-300 1-300Time interval between checks, in seconds.
failure_threshold: 6 #1-10 An instance is unhealthy after failing this number of consecutive checks.
success_threshold: 2 #1-10 An unhealthy instance becomes healthy again after successfully responding to this number of consecutive checks.
initial_delay_sec: 300 #0-3600 The delay, in seconds, after the instance starts during which health check responses are ignored. This setting can allow an instance more time at deployment to get up and running.
##Readiness checks - Readiness checks confirm that an instance can accept incoming requests. Instances that don't pass the readiness check are not added to the pool of available instances.
readiness_check:
path: "/readiness_check"
timeout_sec: 10 #1-300 Timeout interval for each request, in seconds.
check_interval_sec: 15 #1-300 Time interval between checks, in seconds.
failure_threshold: 4 #1-10 An instance is unhealthy after failing this number of consecutive checks.
success_threshold: 2 #1-10 An unhealthy instance becomes healthy after successfully responding to this number of consecutive checks.
app_start_timeout_sec: 300 #1-3600 The maximum time, in seconds, an instance has to become ready after the VM and other infrastructure are provisioned. After this period, the deployment fails and is rolled back. You might want to increase this setting if your application requires significant initialization tasks, such as downloading a large file, before it is ready to serve.
#Service scaling settings
automatic_scaling:
min_num_instances: 2
max_num_instances: 3
cpu_utilization:
target_utilization: 0.7
The error is caused by an upgrade of the stackdriver logging sidecar to 1.6.25 version, which starts to push FluentD metrics to Stackdriver monitoring via OpenCensus. However the integration with App Engine Flex doesn't work yet.
These errors should be logs only. It is not relative to the health check logs. It should not impact VM restart. If your VM instances are restarted frequently, there may be caused by some other reason. In Stackdriver logging UI, you can search Free disk space under vm.syslog stream and unhealthy sidecars under vm.events stream. If some logs show up, your instances restart may be caused by low free disk size or any unhealthy sidecar containers.
About once a week our flexible app engine node app goes offline and the following line appears in the logs: Restarting batch of VMs for version 20181008t134234 as part of rolling restart. We have our app set to automatic scaling with the following settings:
runtime: nodejs
env: flex
beta_settings:
cloud_sql_instances: tuzag-v2:us-east4:tuzag-db
automatic_scaling:
min_num_instances: 1
max_num_instances: 3
liveness_check:
path: "/"
check_interval_sec: 30
timeout_sec: 4
failure_threshold: 2
success_threshold: 2
readiness_check:
path: "/"
check_interval_sec: 15
timeout_sec: 4
failure_threshold: 2
success_threshold: 2
app_start_timeout_sec: 300
resources:
cpu: 1
memory_gb: 1
disk_size_gb: 10
I understand the rolling restarts of GCP/GAE, but am confused as to why Google isn't spinning up another VM before taking our primary one offline. Do we have to run with a min num of 2 instances to prevent this from happening? Is there a way I get configure my app.yaml to make sure another instance is spun up before it reboots the only running instance? After the reboot finishes, everything comes back online fine, but there's still 10 minutes of downtime, which isn't acceptable, especially considering we can't control when it reboots.
We know that it is expected behaviour that Flexible instances are restarted on a weekly basis. Provided that health checks are properly configured and are not the issue, the recommendation is, indeed, to set up a minimum of two instances.
There is no alternative functionality in App Engine Flex, of which I am aware of, that raises a new instance to avoid downtime as a result of a weekly restart. You could try to run directly on Google Compute Engine instead of App Engine and manage updates and maintenance by yourself, perhaps that would suit your purpose better.
Are you just guessing this based on that num instances graph in the app engine dashboard? Or is your app engine project actually unresponsive during that time?
You could use cron to hit it every 5 minutes to see if it's responsive.
Does this issue persist if you change cool_down_period_sec & target_utilization back to their defaults?
If your service is truly down during that time, maybe you should implement a request handler for liveliness checks:
https://cloud.google.com/appengine/docs/flexible/python/reference/app-yaml#updated_health_checks
Their default polling config would tell GAE to launch within a couple minutes
Another thing worth double checking is how long it takes your instance to start up.