amazon web services - AWS autoscale ELB status checks grace period -


i'm running servers in aws auto scale group. running servers behind load balancer. i'm using elb mange auto scaling groups healthchecks. when servers been started , join auto scale group join load balancer.

how time (i.e. healthcheck grace period) need wait until let them join load balancer?

should after servers in state of running?

should after servers passed system , instance status checks?

there 2 types of health check available auto scaling groups:

  • ec2 health check: uses ec2 status check determine whether instance healthy. operates @ hypervisor level , cannot see health of application running on instance.
  • elastic load balancer (elb) health check: causes auto scaling group delegate health check elastic load balancer, capable of checking specific http(s) url. means can check application correctly running on instance.

given system using elb health check, auto scaling trust results of elb health check when determining health of each ec2 instance. can dangerous because, if instance takes while start, health check incorrectly mark instance unhealthy. this, in turn, cause auto scaling terminate instance , launch replacement.

to avoid situation, there health check grace period setting (in seconds) in auto scaling group configuration. indicates how long auto scaling should wait until starts using elb health check (which, in turn, has settings how check , how many checks required mark instance healthy/unhealthy).

so, if application takes 3 minutes start, set health check grace period minimum of 180 seconds (3 minutes). documentation not state whether timing starts moment instance marked "running" or whether when status checks complete, perform timing tests avoid "bounce" situations.

in fact, i recommend setting health check grace period higher value (eg double amount of time required). not impact operation of system since healthy instance start serving traffic elb health check satisfied, sooner auto scaling grace period. worst case genuinely unhealthy instance terminated few minutes later, should rare occurrence.


Comments

Popular posts from this blog

python - mat is not a numerical tuple : openCV error -

c# - MSAA finds controls UI Automation doesn't -

wordpress - .htaccess: RewriteRule: bad flag delimiters -