A colleague of mine found some hint under “Avro type” [2] in the State
evolution schema page:
“Example: RocksDB state backend relies on binary objects identity, rather than
hashCode method implementation. Any changes to the keys object structure could
lead to non deterministic behaviour.”
I gu
Hi Dylan,
thanks for the investigation. I can now also reproduce it my code. Yes,
this is a bug. I opened
https://issues.apache.org/jira/browse/FLINK-21070
and will try to fix this asap.
Thanks,
Timo
On 20.01.21 17:52, Dylan Forciea wrote:
Timo,
I converted what I had to Java, and ended u
@Yang: I think this would be valuable to document. I think it's a natural
question to ask whether you can have standby JMs with Kubernetes. What do you
think? If you agree, we could create a JIRA ticket and work on the "official"
docs for this.
On Thu, Jan 21, 2021, at 5:05 AM, Yang Wang wrote:
> Before the changes, as I am writing this reply, the local time here is
> 2021-01-21 12:03:35 (Beijing time, UTC+8).
> And I tried these 5 functions in sql client, and got:
>
> Flink SQL> select now(), PROCTIME(), CURRENT_TIMESTAMP, CURRENT_DATE,
> CURRENT_TIME;
> +-+-
Hey David,
this is a good catch! I've filed a JIRA ticket to address this in the docs
more prominently: https://issues.apache.org/jira/browse/FLINK-21073
Thanks a lot for reporting this issue!
On Thu, Jan 21, 2021 at 9:24 AM David Haglund
wrote:
> A colleague of mine found some hint under “Avr
I opened a PR. Feel free to try it out.
https://github.com/apache/flink/pull/14720
Btw:
>> env.createTemporarySystemFunction("LatestNonNullLong",
>> classOf[LatestNonNull[Long]])
>>
>> env.createTemporarySystemFunction("LatestNonNullString",
>> classOf[LatestNonNull[String]])
don't make a diff
Thanks a lot for your message.
Why is there a difference of 5 minutes between the timestamp of the job
submission from the client to the timestamp on the JobManager where the
submission is received?
Is there any service / custom logic involved in the job submission? (e.g. a
proxy in between, that
Hello everybody,
I was trying to get rid of the deprecation warnings about
using BatchTableEnvironment.registerTableSink() but I don't know how to
proceed.
My current code does the following:
BatchTableEnvironment benv = BatchTableEnvironment.create(env);
benv.registerTableSink("outUsers", getFie
Timo,
Will do! I have been patching in a change locally that I have a PR [1] out for,
so if this will end up in the next 1.12 patch release, I may add this in with
it once it has been approved and merged.
On a side note, that PR has been out since the end of October (looks like I
need to do a
Since our CI system is able to build Flink, I believe it's a local issue.
Are you sure that the build is failing when you build Flink from the root
directory (not calling maven from within a maven module?)
On Tue, Jan 19, 2021 at 11:19 AM Smile@LETTers
wrote:
> Hi,
> I got an error when tried t
Hello, everyone
I want to start the discussion of FLIP-162: Consistent Flink SQL time function
behavior[1].
We’ve some initial discussion of several problematic functions in dev mail
list[2], and I think it's the right time to resolve them by a FLIP.
Currently some time function behaviors
Hi Dylan,
I can help with a review for your PR tomorrow. In general, I would
recommend to just ping people a couple of times that have been worked on
the component before (see git blame) to get a review. We are all busy
and need a bit of pushing from time to time ;-)
Thanks,
Timo
On 21.01.2
Hi Robert,
I sent you an email with instructions to create an account to view the logs
through our secure repository. I’ve included the JobManager and client
application logs there.
We have a thread pool that we use to submit multiple jobs in parallel, but in
there there’s no retry logic – if
It works now. Job manager is able to start. But now, I have run into
another issue.
It seems job manager is trying to create configmap in default namespace and
namespace/service account where I run job manager does not have access to
configmap
GET at:
https://X.X.X.X/api/v1/namespaces/default/con
Currently, my jobs that require JDBC initialize a connection in the open
method directly via JDBC driver.
1. What are the established best practices for this?
2. Is it better to use a connection pool that can validate the connection
and reconnect?
3. Would each operator require its own connection
Is this comment in the file
flink-streaming-java/src/main/java/org/apache/flink/streaming/api/datastream/CoGroupedStreams.java
accurate?
" * Note: Right now, the groups are being built in memory so you need to
ensure that they don't
* get too big. Otherwise the JVM might crash."
Looking at the s
Is this comment in the file
flink-streaming-java/src/main/java/org/apache/flink/streaming/api/datastream/CoGroupedStreams.java
accurate?
" * Note: Right now, the groups are being built in memory so you need to
ensure that they don't
* get too big. Otherwise the JVM might crash."
Looking at the s
I wanted to report that I tried out your PR, and it does solve my issue. I am
able to create a generic LatestNonNull and it appears to do what is expected.
Thanks,
Dylan Forciea
On 1/21/21, 8:50 AM, "Timo Walther" wrote:
I opened a PR. Feel free to try it out.
https://github.com/apac
I have a Streaming process where new directories are added daily in S3.
s3://foo/bar/2021-01-18/data.gz
s3://foo/bar/2021-01-19/data.gz
s3://foo/bar/2021-01-20/data.gz
If I start the process, it won't pick up anything other than the
directories visible when the process was started.
The textInput
Hi,
We recently migrated from 1.9.1 to flink 1.11 and notice the new job cannot
consume from savepoint taken in 1.9.1. Here is the list of operator id and
max parallelism of savepoints taken in both versions. The only code change
is version upgrade.
savepoint 1.9.1:
```
Id: 8a74550ce6afad759d5f1d
I sent this a little prematurely. Will the streaming process find new
directories under the parent?
If the input path is
s3://foo.bar/
and directories are added daily, should I expect that the newly added
directories+files will get processed?
Thanks!
Wayne
On 2021/01/21 23:20:41, Billy Bain
Hi,
I think you could try something like this
firstStream
.coGroup(secondStream)
.where(_.id)
.equalTo(_.id)
.window(TumblingEventTimeWindows.of(Time.seconds(1)))
.with(new MyCogroupFunction())
.uid("myCoGroup")
Best,
Guowei
On Fri, Jan 22, 2021 at 4:33 AM Sudh
Yes. I agree with you that it is valuable to document how to start multiple
JobManagers in HA mode for K8s deployment.
I have a ticket to track this documentation improvement[1].
[1]. https://issues.apache.org/jira/browse/FLINK-21082
Best,
Yang
Ufuk Celebi 于2021年1月21日周四 下午9:08写道:
> @Yang: I th
You could set config option "kubernetes.namespace" to your flink-conf
ConfigMap. And then
KubernetesHAService will use it to create/watch the ConfigMap. Please note
the default service
account has enough permission.
Of course, you could also set the config option
"kubernetes.service-account" to an
Hi David,
Thanks for your enthusiasm to figure out the root cause. The key difference is
that RocksDB holds binary objects which are only defined by the serialized
bytes while Fs/MemoryStateBackend holds objects in pojo format which are
defined by the hashCode and equals. If you want to achieve
Hey Andreas,
thanks a lot for providing me with the full logs.
The JobManager actually received 2 job submissions.
There are 2 relevant log messages.
1. "Received JobGraph submission xxx"
2. "Submitting job"
1 is logged right after the dispatcher receives the JobGraph, before the
duplicate submiss
26 matches
Mail list logo