Hi!
In the above example the keySelector would run once before and once inside
the window operator. In that sense, the version below is a better way to do
it.
You can also create windows of 50 or max 100 ms by writing your own
trigger. Have a look at the count trigger. You can augment it by
Hi Stephan
Thanks for the explanation! We will stick to 1.0.3 to keep our code clean.
In the workaround case, how does key selector instantiated? One instance per
window operator?
By the way is there a way to create a hybrid window of count and time, like 50
items or max process time 100ms?
Hi Andrew!
Here is the reason for what is happening with your job:
You have used some sort of undocumented and unofficial corner case behavior
of Flink 1.0.0, namely, using parallel windowAll().
Initially, windowAll() was supposed to not be parallel, but the system did
not prevent to set a
Hi Aljoscha
We are not using state backend explicitly, recovery and state backend are
pointed to file path.
See attached json file
Thanks for the help.
Best regards
Andrew
> On 10 Aug 2016, at 11:38, Aljoscha Krettek wrote:
>
> Oh, are you by any chance specifying a
Oh, are you by any chance specifying a custom state backend for your job?
For example, RocksDBStateBackend.
Cheers,
Aljoscha
On Wed, 10 Aug 2016 at 11:17 Aljoscha Krettek wrote:
> Hi,
> could you maybe send us the output of "env.getExecutionPlan()". This would
> help us
Hi,
could you maybe send us the output of "env.getExecutionPlan()". This would
help us better understand which operators are used exactly. (You can of
course remove any security sensitive stuff.)
Cheers,
Aljoscha
On Tue, 9 Aug 2016 at 15:30 Andrew Ge Wu wrote:
> Oh
Oh sorry missed that part, no, Im not explicitly set that.
> On 09 Aug 2016, at 15:29, Aljoscha Krettek wrote:
>
> Hi,
> are you setting a StreamTimeCharacteristic, i.e.
> env.setStreamTimeCharacteristic?
>
> Cheers,
> Aljoscha
>
> On Tue, 9 Aug 2016 at 14:52 Andrew Ge
Hi,
are you setting a StreamTimeCharacteristic, i.e.
env.setStreamTimeCharacteristic?
Cheers,
Aljoscha
On Tue, 9 Aug 2016 at 14:52 Andrew Ge Wu wrote:
> Hi Aljoscha
>
>
> Plan attached, there are split streams and union operations around, but
> here is how windows are
Hi,
could you maybe post how exactly you specify the window? Also, did you set
a "stream time characteristic", for example EventTime?
That could help us pinpoint the problem.
Cheers,
Aljoscha
On Tue, 9 Aug 2016 at 12:42 Andrew Ge Wu wrote:
> I rolled back to 1.0.3
> If
I rolled back to 1.0.3
If I understand this correctly, the peak when topology starts is because it is
trying to fill all the buffers, but I can not see that in 1.1.0.
> On 09 Aug 2016, at 12:10, Robert Metzger wrote:
>
> Which source are you using?
>
> On Tue, Aug 9,
Which source are you using?
On Tue, Aug 9, 2016 at 11:50 AM, Andrew Ge Wu
wrote:
> Hi Robert
>
>
> Thanks for the quick reply, I guess I’m one of the early birds.
> Yes, it is much slower, I’m not sure why, I copied slaves, masters,
> log4j.properties and flink-conf.yaml
Hi Robert
Thanks for the quick reply, I guess I’m one of the early birds.
Yes, it is much slower, I’m not sure why, I copied slaves, masters,
log4j.properties and flink-conf.yaml directly from 1.0.3
I have parallelization 1 on my sources, I can increase that to achieve the same
speed, but I’m
Hi Andrew,
here is the release announcement, with a list of all changes:
http://flink.apache.org/news/2016/08/08/release-1.1.0.html,
http://flink.apache.org/blog/release_1.1.0-changelog.html
What does the chart say? Are the results different? is Flink faster or
slower now?
Regards,
Robert
On
13 matches
Mail list logo