flink-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Maximilian Michels <...@apache.org>
Subject Re: Running continuously on yarn with kerberos
Date Tue, 27 Oct 2015 17:12:44 GMT
Hi Niels,

You're welcome. Some more information on how this would be configured:

In the kdc.conf, there are two variables:

        max_life = 2h 0m 0s
        max_renewable_life = 7d 0h 0m 0s

max_life is the maximum life of the current ticket. However, it may be
renewed up to a time span of max_renewable_life from the first ticket issue
on. This means that from the first ticket issue, new tickets may be
requested for one week. Each renewed ticket has a life time of max_life (2
hours in this case).

Please let us know about any difficulties with long-running streaming
application and Kerberos.

Best regards,

On Tue, Oct 27, 2015 at 2:46 PM, Niels Basjes <Niels@basjes.nl> wrote:

> Hi,
> Thanks for your feedback.
> So I guess I'll have to talk to the security guys about having special
> kerberos ticket expiry times for these types of jobs.
> Niels Basjes
> On Fri, Oct 23, 2015 at 11:45 AM, Maximilian Michels <mxm@apache.org>
> wrote:
>> Hi Niels,
>> Thank you for your question. Flink relies entirely on the Kerberos
>> support of Hadoop. So your question could also be rephrased to "Does
>> Hadoop support long-term authentication using Kerberos?". And the
>> answer is: Yes!
>> While Hadoop uses Kerberos tickets to authenticate users with services
>> initially, the authentication process continues differently
>> afterwards. Instead of saving the ticket to authenticate on a later
>> access, Hadoop creates its own security tockens (DelegationToken) that
>> it passes around. These are authenticated to Kerberos periodically. To
>> my knowledge, the tokens have a life span identical to the Kerberos
>> ticket maximum life span. So be sure to set the maximum life span very
>> high for long streaming jobs. The renewal time, on the other hand, is
>> not important because Hadoop abstracts this away using its own
>> security tockens.
>> I'm afraid there is not Kerberos how-to yet. If you are on Yarn, then
>> it is sufficient to authenticate the client with Kerberos. On a Flink
>> standalone cluster you need to ensure that, initially, all nodes are
>> authenticated with Kerberos using the kinit tool.
>> Feel free to ask if you have more questions and let us know about any
>> difficulties.
>> Best regards,
>> Max
>> On Thu, Oct 22, 2015 at 2:06 PM, Niels Basjes <Niels@basjes.nl> wrote:
>> > Hi,
>> >
>> > I want to write a long running (i.e. never stop it) streaming flink
>> > application on a kerberos secured Hadoop/Yarn cluster. My application
>> needs
>> > to do things with files on HDFS and HBase tables on that cluster so
>> having
>> > the correct kerberos tickets is very important. The stream is to be
>> ingested
>> > from Kafka.
>> >
>> > One of the things with Kerberos is that the tickets expire after a
>> > predetermined time. My knowledge about kerberos is very limited so I
>> hope
>> > you guys can help me.
>> >
>> > My question is actually quite simple: Is there an howto somewhere on
>> how to
>> > correctly run a long running flink application with kerberos that
>> includes a
>> > solution for the kerberos ticket timeout  ?
>> >
>> > Thanks
>> >
>> > Niels Basjes
> --
> Best regards / Met vriendelijke groeten,
> Niels Basjes

View raw message