[
https://issues.apache.org/jira/browse/MYRIAD-131?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=14876146#comment-14876146
]
Santosh Marella commented on MYRIAD-131:
As part of adding constraints in the flex APIs, I'm modifying the behavior of
flexdown a bit. Previously, flexdown was killing NMs present in "active" state,
followed by killing NMs in "staging" and then NMs in "pending" state. The new
behavior would be:
1. Kill NMs in "pending" state, if the "profile" and "constraints" specified in
the API match those in "pending" state.
2. Kill NMs in "staging" state, if the "profile" and "constraints" specified in
the API match those in "staging" state.
3. Kill NMs in "active" state, if the "profile" and "constraints" specified in
the API match those in "active" state.
Example:
Setup:
- 3 nodes running mesos-slave: host131.example.com, host132.example.com,
host133.example.com
- host133.example.com has a mesos-slave attribute defined as "dfs:true"
a. flexup {"profile": "medium", "instances": 1, "constraints": ["dfs LIKE
true"]}
// launches nm.medium on host133, as host133 has attribute "dfs:true"
b. flexup {"profile": "zero", "instances": 2, "constraints": ["hostname LIKE
host13[0-9].example.com"]}
// launches nm.zero on host131 and host132
c. flexup {"profile": "zero", "instances": 3}
// (NO CONSTRAINTS) keeps 3 nm.zero instances in "pending"
d. flexdown {"profile": "zero", "instances": 1}
// kills a nm.zero in "pending" state. after this, 2 nm.zero instances
should be left in "pending" state and 2 nm.zero instances in "active", each on
host131 and host132.
e. flexdown {"profile": "zero", "instances": 1, "constraints": ["hostname
LIKE host131"]}
// kills "active" nm.zero running on host131. A "pending" nm.zero might
get launched on host131 (and become "active") if mesos offers resources from
host131.
> Timeout for tasks in Pending or Staging state
> -
>
> Key: MYRIAD-131
> URL: https://issues.apache.org/jira/browse/MYRIAD-131
> Project: Myriad
> Issue Type: Bug
> Components: Scheduler
>Reporter: Aashreya Ravi Shankar
>
> Currently tasks in the pending or staging state never times out. I think we
> need to set a timeout if a process is unable to get launched for any reason.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
