Github user lianhuiwang commented on the pull request:
https://github.com/apache/spark/pull/1589#issuecomment-62720066
I think RM will allocate more than one to spark's AM when a executor fails.
Here is a scenario:
1. spark requests 3 containers (AM & RM request total = 3)
2. RM allocates one and give it to spark (AM request total = 3 but RM
request total = 2)
3. spark task failed so AM will +1 to requests (AM request total = 4)
4. RM receive AM's request and allocates one to spark (RM request total =3
(4-1)), but actually is 2
5. but next RM will allocate three to spark.
and then in appmaster's log there will be the following logs:
Ignoring container %s at host %s, since we already have the required number
of containers for it.
in this scenario,when AM received one allocated container, AM donot remove
one request. so it causes RM allocate more containers to AM.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
