[
https://issues.apache.org/jira/browse/SCB-2742?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
liubao updated SCB-2742:
------------------------
Description:
Motivations:
# In error condition, the best choice for protection is fast fail, not ensure
the invocation success. In many situations, the way we try to make the
invocation success, will cause more problems.
# For example, if one instance error goes to 50% or 50% invocations timeout,
and we isolate the instance, this will migrate invocations to other instances.
In real applications, the migration will cause other instance fail.
(Microservice instances are stateless, and one fail , the other may fail. )
# We suggest users to use
instance-isolation-consumer,instance-bulkhead-consumer to make failure fail
fast, and not isolation the instance, so that when the window is match, this
instance can be available. And we do not migrate the invocations from this
instance to others, this will not cause too many invocations to other
instances.
# For extreme condition needs isolation instance, it's better to configure
alarms for circuit breaker and use service management to turn down the instance
manually.
was:
Motivations:
# In error condition, the best choice for protection is fast fail, not ensure
the invocation success. In many situations, the way we try to make the
invocation success, will cause more problems.
# For example, if one instance error goes to 50% or 50% invocations timeout,
and we isolate the instance, this will migrate invocations to other instances.
In real applications, the migration will cause other instance fail.
(Microservice instances are stateless, and one fail , the other may fail. )
# We suggest users to use
instance-isolation-consumer,instance-bulkhead-consumer to make failure fail
fast, and not isolation the instance, so that when the window is match, this
instance can be available. And we do not migrate the invocations from this
instance to others, this will not cause too many invocations to other
instances.
> change IsolationDiscoveryFilter enabled by false
> ------------------------------------------------
>
> Key: SCB-2742
> URL: https://issues.apache.org/jira/browse/SCB-2742
> Project: Apache ServiceComb
> Issue Type: Improvement
> Components: Java-Chassis
> Reporter: liubao
> Assignee: liubao
> Priority: Major
> Fix For: java-chassis-2.9.0
>
>
> Motivations:
>
> # In error condition, the best choice for protection is fast fail, not
> ensure the invocation success. In many situations, the way we try to make the
> invocation success, will cause more problems.
> # For example, if one instance error goes to 50% or 50% invocations timeout,
> and we isolate the instance, this will migrate invocations to other
> instances. In real applications, the migration will cause other instance
> fail. (Microservice instances are stateless, and one fail , the other may
> fail. )
> # We suggest users to use
> instance-isolation-consumer,instance-bulkhead-consumer to make failure fail
> fast, and not isolation the instance, so that when the window is match, this
> instance can be available. And we do not migrate the invocations from this
> instance to others, this will not cause too many invocations to other
> instances.
> # For extreme condition needs isolation instance, it's better to configure
> alarms for circuit breaker and use service management to turn down the
> instance manually.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
