tgravescs commented on pull request #29413:
URL: https://github.com/apache/spark/pull/29413#issuecomment-675474599
From my experience:
1) For the dynamic allocation it definitely is, but in my opinion the right
solution there is to pull the dynamic allocation manager into the core
tgravescs commented on pull request #29413:
URL: https://github.com/apache/spark/pull/29413#issuecomment-675001276
>> Yes they can make but in next run , after they see some abrupt behaviour
(application hung/ Resource wasted).But what if due to this extra threshold
size there is no
tgravescs commented on pull request #29413:
URL: https://github.com/apache/spark/pull/29413#issuecomment-674877554
>> We are already setting some value in cluster defaults . But as I said
earlier also "There is no fixed size of the Queue which can be used in all the
Spark Jobs and even
tgravescs commented on pull request #29413:
URL: https://github.com/apache/spark/pull/29413#issuecomment-674069902
I was just saying, if you had one pool you took from you can't control it
per queue. Unless of course you put in some sort of minimum or something per
queue. In most of the
tgravescs commented on pull request #29413:
URL: https://github.com/apache/spark/pull/29413#issuecomment-673665565
Yeah so you have described the affects of it dropping the events, which I
know. The thing I want to know is why it dropped the events in your cases.
I'm not sure the
tgravescs commented on pull request #29413:
URL: https://github.com/apache/spark/pull/29413#issuecomment-673479231
Part of the problem with that is that in most cases some queues have higher
priority, which is why they were split apart. you really want the executor
management queue to
tgravescs commented on pull request #29413:
URL: https://github.com/apache/spark/pull/29413#issuecomment-673474159
> We can think of making it per queue basis any such approach based on the
user requirement like I have already done in this PR using the
conf(spark.set.optmized.event.queue)
tgravescs commented on pull request #29413:
URL: https://github.com/apache/spark/pull/29413#issuecomment-673097079
sorry I'm still not seeing any difference here then increasing the size of
the current queue? If both are not really allocating memory for the entire
amount until runtime
tgravescs commented on pull request #29413:
URL: https://github.com/apache/spark/pull/29413#issuecomment-672862588
so I definitely get the point here and it would definitely be nice to handle
this better somehow, but if you are setting it be queue size + some threshold
then the
tgravescs commented on pull request #29413:
URL: https://github.com/apache/spark/pull/29413#issuecomment-672856870
so one question, here, you are seeing events dropped that cause hangs in the
application code? What queue were they in?
10 matches
Mail list logo