You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@reef.apache.org by Tobin Baker <td...@cs.washington.edu> on 2015/11/12 21:30:31 UTC
log warnings on invalid ResourceManager state transitions
I'm constantly seeing warnings like this in the logs. Is this a bug?
Nov 12, 2015 12:27:09 PM
org.apache.reef.runtime.local.driver.ResourceManager sendRuntimeStatus
INFO: Allocated: 4, Outstanding requests: Optional:{0}
Nov 12, 2015 12:27:09 PM
org.apache.reef.runtime.common.driver.resourcemanager.ResourceManagerStatus
isLegalStateTransition
WARNING: Transition from RUNNING state to the same state. This shouldn't
happen.
Re: log warnings on invalid ResourceManager state transitions
Posted by Markus Weimer <ma...@weimo.de>.
On 2015-11-18 01:57, Dudoladov, Sergey wrote:
>> Remove the warning. There is no harm in setting the status to the
>> same value over and over.
>
> We can keep the log statement, but decrease its level to "finest".
> So we document in code the fact that repeatedly setting the status to the same state is OK.
+1
Markus
Re: log warnings on invalid ResourceManager state transitions
Posted by "Dudoladov, Sergey" <se...@tu-berlin.de>.
> Remove the warning. There is no harm in setting the status to the
> same value over and over.
We can keep the log statement, but decrease its level to "finest".
So we document in code the fact that repeatedly setting the status to the same state is OK.
Best,
Sergey
Re: log warnings on invalid ResourceManager state transitions
Posted by Markus Weimer <ma...@weimo.de>.
On 2015-11-13 07:59, Tobin Baker wrote:
> If it really shouldn't happen, shouldn't this be an assert?
Yes. However, I am no longer convinced that this should not happen. At
some level, we want messages from the resource manager that repeat an
already known state to be idempotent. The question is at which level we
want this to happen.
That warning stems from the idea that the resource manager plugin (the
runtime) doesn't send those messages if they just repeat the already
known status. However, that doesn't seem feasible, as the runtime status
also includes the the set of outstanding and allocated evaluators.
Hence, the message will be sent often with the same status.
Now, I see two possible responses:
(1) Remove the warning. There is no harm in setting the status to the
same value over and over.
(2) Add an `if` to `ResourceManagerStatus.setState()` to do nothing if
the state did not change.
Markus
Re: log warnings on invalid ResourceManager state transitions
Posted by Tobin Baker <td...@cs.washington.edu>.
You can look at the log I uploaded at
https://issues.apache.org/jira/secure/attachment/12771823/driver.stderr.
This was a test run for my Driver implementation (
https://github.com/uwescience/myria/blob/reef/src/edu/washington/escience/myria/daemon/MyriaDriver.java
).
If it really shouldn't happen, shouldn't this be an assert?
On Fri, Nov 13, 2015 at 3:10 AM, Dudoladov, Sergey <
sergey.dudoladov@tu-berlin.de> wrote:
>
> >org.apache.reef.runtime.common.driver.resourcemanager.ResourceManagerStatus
> > isLegalStateTransition WARNING: Transition from RUNNING state to the
> same state. This shouldn't
> > happen.
> > I'm constantly seeing warnings like this in the logs. Is this a bug?
>
> I added this warning in REEF-826; we thought these transitions shouldn't
> happen.
> Can you provide more details ?
>
Re: log warnings on invalid ResourceManager state transitions
Posted by "Dudoladov, Sergey" <se...@tu-berlin.de>.
>org.apache.reef.runtime.common.driver.resourcemanager.ResourceManagerStatus
> isLegalStateTransition WARNING: Transition from RUNNING state to the same state. This shouldn't
> happen.
> I'm constantly seeing warnings like this in the logs. Is this a bug?
I added this warning in REEF-826; we thought these transitions shouldn't happen.
Can you provide more details ?