You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@falcon.apache.org by Venkat Ranganathan <vr...@hortonworks.com> on 2015/10/24 17:50:06 UTC
Oozie uber mode and other modes of running the software
FALCON-1551 was interesting for two reasons - that we are still caught by the FS caching issue (this should have been dealt by most of the projects one time or another), but the use of Oozie uber mode - I am wondering if we have done uber mode testing generally - Is this the mode that is used in your environments – It makes sense when thousands of Oozie jobs are run – saving one container. I would like to understand more about your experience with uber mode if you have – may be in the next Falcon sync up.
I think we have create a list of modes that we expect the product to run before we introduce changes generally – Given that there are different sets of execution modes that we want to test. Please add more on the scenarios that I am most likely missing. Let us discuss this also if we have time.
1. Security (Kerberos and Wire encryption)
2. NN HA/RM HA
3. Oozie HA
4. Oozie ubermode for launcher job
5. Multi cluster with prism
6. Multi cluster managed with Oozie (with single Falcon instance)
7. Hadoop/Hive/Oozie versions and the implications (S3 and Azure blob storage went out of hadoop-common in 2.6.0+ and the implications).
Thanks
Venkat
Re: Oozie uber mode and other modes of running the software
Posted by Venkat Ranganathan <vr...@hortonworks.com>.
Thanks Pragya. Srikanth, good point about the cluster version requirements.
It is definitely something to keep in mind. Thanks for the heads up.
Venkat
On 10/26/15, 3:07 AM, "pragya mittal" <mi...@gmail.com> wrote:
>Typo :
>Oozie version is 4.2 and RM HA testing is still in progress.
>
>On Sun, Oct 25, 2015 at 9:25 AM, pragya mittal <mi...@gmail.com>
>wrote:
>
>> Hi Venkat,
>>
>> From past three months, Inmobi has been using Oozie in uber mode.
>> Currently we are using falcon-0.7 with oozie-3.2 (along with several
>> patches specific to our use case). Following use cases have been tested and
>> are active in our production environment.
>>
>> 1. NN HA/RM HA
>> 2. Oozie ubermode for launcher job
>> 3. Multi cluster with prism
>> 4. Sanity of falcon features with oozie(in uber mode).
>>
>> After falcon-0.8 release we plan to go forward with Oozie HA feature.
>>
>> Please let me know if you have any more concerns.
>>
>> Regards,
>> Pragya Mittal
>>
>> On Sun, Oct 25, 2015 at 8:15 AM, Srikanth Sundarrajan <sriksun@hotmail.com
>> > wrote:
>>
>>> Am guessing venkat you meant to refer FALCON-1555. Good check list to go
>>> through for each release.
>>>
>>> Specifically on
>>> >> 6. Multi cluster managed with Oozie (with single Falcon instance)
>>>
>>> Since we haven't particularly done anything by isolating the classes and
>>> their dependencies for each cluster with independent class loader, I think
>>> this would work only as long as all the clusters are compatible. If things
>>> were to change say with hadoop-3 for ex, this may no longer work.
>>>
>>> Regards
>>> Srikanth Sundarrajan
>>>
>>> > From: ajay.yadav@inmobi.com
>>> > Date: Sun, 25 Oct 2015 05:29:20 +0530
>>> > Subject: Re: Oozie uber mode and other modes of running the software
>>> > To: dev@falcon.apache.org
>>> >
>>> > Venkat,
>>> >
>>> > Sounds good, let's discuss it in the next sync up. For the benefit of
>>> > everyone on the mailing list, I will summarise our use cases. Yes, we
>>> > (Inmobi) use Oozie's uber mode in our production environment(IIRC we are
>>> > using it since Apache Falcon 0.7). We do have several thousand jobs
>>> running
>>> > per day(several at minutely frequencies) and a heavily utilised
>>> cluster, so
>>> > this makes sense for us. We do test all Oozie releases, though sometimes
>>> > our Oozie releases are not exactly aligned with Oozie's general release
>>> -
>>> > it might contain some extra patches, might be from trunk etc. and our
>>> > testing might be specific to features being used at Inmobi. In fact last
>>> > time Pragya Mittal did the testing for both Apache Falcon 0.7 and Oozie.
>>> > She can add more on this.
>>> >
>>> > List of scenarios looks comprehensive. Just curious, what storage
>>> backends
>>> > are other users using for titan?
>>> >
>>> >
>>> > Cheers
>>> > Ajay Yadava
>>> >
>>> > On Sat, Oct 24, 2015 at 9:20 PM, Venkat Ranganathan <
>>> > vranganathan@hortonworks.com> wrote:
>>> >
>>> > > FALCON-1551 was interesting for two reasons - that we are still
>>> caught by
>>> > > the FS caching issue (this should have been dealt by most of the
>>> projects
>>> > > one time or another), but the use of Oozie uber mode - I am wondering
>>> if we
>>> > > have done uber mode testing generally - Is this the mode that is used
>>> in
>>> > > your environments – It makes sense when thousands of Oozie jobs are
>>> run –
>>> > > saving one container. I would like to understand more about your
>>> > > experience with uber mode if you have – may be in the next Falcon
>>> sync up.
>>> > >
>>> > > I think we have create a list of modes that we expect the product to
>>> run
>>> > > before we introduce changes generally – Given that there are
>>> different sets
>>> > > of execution modes that we want to test. Please add more on the
>>> scenarios
>>> > > that I am most likely missing. Let us discuss this also if we have
>>> time.
>>> > >
>>> > > 1. Security (Kerberos and Wire encryption)
>>> > > 2. NN HA/RM HA
>>> > > 3. Oozie HA
>>> > > 4. Oozie ubermode for launcher job
>>> > > 5. Multi cluster with prism
>>> > > 6. Multi cluster managed with Oozie (with single Falcon instance)
>>> > > 7. Hadoop/Hive/Oozie versions and the implications (S3 and Azure
>>> blob
>>> > > storage went out of hadoop-common in 2.6.0+ and the implications).
>>> > >
>>> > >
>>> > > Thanks
>>> > >
>>> > > Venkat
>>> > >
>>> > >
>>> >
>>> > --
>>> > _____________________________________________________________
>>> > The information contained in this communication is intended solely for
>>> the
>>> > use of the individual or entity to whom it is addressed and others
>>> > authorized to receive it. It may contain confidential or legally
>>> privileged
>>> > information. If you are not the intended recipient you are hereby
>>> notified
>>> > that any disclosure, copying, distribution or taking any action in
>>> reliance
>>> > on the contents of this information is strictly prohibited and may be
>>> > unlawful. If you have received this communication in error, please
>>> notify
>>> > us immediately by responding to this email and then delete it from your
>>> > system. The firm is neither liable for the proper and complete
>>> transmission
>>> > of the information contained in this communication nor for any delay in
>>> its
>>> > receipt.
>>>
>>>
>>
>>
Re: Oozie uber mode and other modes of running the software
Posted by pragya mittal <mi...@gmail.com>.
Typo :
Oozie version is 4.2 and RM HA testing is still in progress.
On Sun, Oct 25, 2015 at 9:25 AM, pragya mittal <mi...@gmail.com>
wrote:
> Hi Venkat,
>
> From past three months, Inmobi has been using Oozie in uber mode.
> Currently we are using falcon-0.7 with oozie-3.2 (along with several
> patches specific to our use case). Following use cases have been tested and
> are active in our production environment.
>
> 1. NN HA/RM HA
> 2. Oozie ubermode for launcher job
> 3. Multi cluster with prism
> 4. Sanity of falcon features with oozie(in uber mode).
>
> After falcon-0.8 release we plan to go forward with Oozie HA feature.
>
> Please let me know if you have any more concerns.
>
> Regards,
> Pragya Mittal
>
> On Sun, Oct 25, 2015 at 8:15 AM, Srikanth Sundarrajan <sriksun@hotmail.com
> > wrote:
>
>> Am guessing venkat you meant to refer FALCON-1555. Good check list to go
>> through for each release.
>>
>> Specifically on
>> >> 6. Multi cluster managed with Oozie (with single Falcon instance)
>>
>> Since we haven't particularly done anything by isolating the classes and
>> their dependencies for each cluster with independent class loader, I think
>> this would work only as long as all the clusters are compatible. If things
>> were to change say with hadoop-3 for ex, this may no longer work.
>>
>> Regards
>> Srikanth Sundarrajan
>>
>> > From: ajay.yadav@inmobi.com
>> > Date: Sun, 25 Oct 2015 05:29:20 +0530
>> > Subject: Re: Oozie uber mode and other modes of running the software
>> > To: dev@falcon.apache.org
>> >
>> > Venkat,
>> >
>> > Sounds good, let's discuss it in the next sync up. For the benefit of
>> > everyone on the mailing list, I will summarise our use cases. Yes, we
>> > (Inmobi) use Oozie's uber mode in our production environment(IIRC we are
>> > using it since Apache Falcon 0.7). We do have several thousand jobs
>> running
>> > per day(several at minutely frequencies) and a heavily utilised
>> cluster, so
>> > this makes sense for us. We do test all Oozie releases, though sometimes
>> > our Oozie releases are not exactly aligned with Oozie's general release
>> -
>> > it might contain some extra patches, might be from trunk etc. and our
>> > testing might be specific to features being used at Inmobi. In fact last
>> > time Pragya Mittal did the testing for both Apache Falcon 0.7 and Oozie.
>> > She can add more on this.
>> >
>> > List of scenarios looks comprehensive. Just curious, what storage
>> backends
>> > are other users using for titan?
>> >
>> >
>> > Cheers
>> > Ajay Yadava
>> >
>> > On Sat, Oct 24, 2015 at 9:20 PM, Venkat Ranganathan <
>> > vranganathan@hortonworks.com> wrote:
>> >
>> > > FALCON-1551 was interesting for two reasons - that we are still
>> caught by
>> > > the FS caching issue (this should have been dealt by most of the
>> projects
>> > > one time or another), but the use of Oozie uber mode - I am wondering
>> if we
>> > > have done uber mode testing generally - Is this the mode that is used
>> in
>> > > your environments – It makes sense when thousands of Oozie jobs are
>> run –
>> > > saving one container. I would like to understand more about your
>> > > experience with uber mode if you have – may be in the next Falcon
>> sync up.
>> > >
>> > > I think we have create a list of modes that we expect the product to
>> run
>> > > before we introduce changes generally – Given that there are
>> different sets
>> > > of execution modes that we want to test. Please add more on the
>> scenarios
>> > > that I am most likely missing. Let us discuss this also if we have
>> time.
>> > >
>> > > 1. Security (Kerberos and Wire encryption)
>> > > 2. NN HA/RM HA
>> > > 3. Oozie HA
>> > > 4. Oozie ubermode for launcher job
>> > > 5. Multi cluster with prism
>> > > 6. Multi cluster managed with Oozie (with single Falcon instance)
>> > > 7. Hadoop/Hive/Oozie versions and the implications (S3 and Azure
>> blob
>> > > storage went out of hadoop-common in 2.6.0+ and the implications).
>> > >
>> > >
>> > > Thanks
>> > >
>> > > Venkat
>> > >
>> > >
>> >
>> > --
>> > _____________________________________________________________
>> > The information contained in this communication is intended solely for
>> the
>> > use of the individual or entity to whom it is addressed and others
>> > authorized to receive it. It may contain confidential or legally
>> privileged
>> > information. If you are not the intended recipient you are hereby
>> notified
>> > that any disclosure, copying, distribution or taking any action in
>> reliance
>> > on the contents of this information is strictly prohibited and may be
>> > unlawful. If you have received this communication in error, please
>> notify
>> > us immediately by responding to this email and then delete it from your
>> > system. The firm is neither liable for the proper and complete
>> transmission
>> > of the information contained in this communication nor for any delay in
>> its
>> > receipt.
>>
>>
>
>
Re: Oozie uber mode and other modes of running the software
Posted by pragya mittal <mi...@gmail.com>.
Hi Venkat,
>From past three months, Inmobi has been using Oozie in uber mode. Currently
we are using falcon-0.7 with oozie-3.2 (along with several patches specific
to our use case). Following use cases have been tested and are active in
our production environment.
1. NN HA/RM HA
2. Oozie ubermode for launcher job
3. Multi cluster with prism
4. Sanity of falcon features with oozie(in uber mode).
After falcon-0.8 release we plan to go forward with Oozie HA feature.
Please let me know if you have any more concerns.
Regards,
Pragya Mittal
On Sun, Oct 25, 2015 at 8:15 AM, Srikanth Sundarrajan <sr...@hotmail.com>
wrote:
> Am guessing venkat you meant to refer FALCON-1555. Good check list to go
> through for each release.
>
> Specifically on
> >> 6. Multi cluster managed with Oozie (with single Falcon instance)
>
> Since we haven't particularly done anything by isolating the classes and
> their dependencies for each cluster with independent class loader, I think
> this would work only as long as all the clusters are compatible. If things
> were to change say with hadoop-3 for ex, this may no longer work.
>
> Regards
> Srikanth Sundarrajan
>
> > From: ajay.yadav@inmobi.com
> > Date: Sun, 25 Oct 2015 05:29:20 +0530
> > Subject: Re: Oozie uber mode and other modes of running the software
> > To: dev@falcon.apache.org
> >
> > Venkat,
> >
> > Sounds good, let's discuss it in the next sync up. For the benefit of
> > everyone on the mailing list, I will summarise our use cases. Yes, we
> > (Inmobi) use Oozie's uber mode in our production environment(IIRC we are
> > using it since Apache Falcon 0.7). We do have several thousand jobs
> running
> > per day(several at minutely frequencies) and a heavily utilised cluster,
> so
> > this makes sense for us. We do test all Oozie releases, though sometimes
> > our Oozie releases are not exactly aligned with Oozie's general release -
> > it might contain some extra patches, might be from trunk etc. and our
> > testing might be specific to features being used at Inmobi. In fact last
> > time Pragya Mittal did the testing for both Apache Falcon 0.7 and Oozie.
> > She can add more on this.
> >
> > List of scenarios looks comprehensive. Just curious, what storage
> backends
> > are other users using for titan?
> >
> >
> > Cheers
> > Ajay Yadava
> >
> > On Sat, Oct 24, 2015 at 9:20 PM, Venkat Ranganathan <
> > vranganathan@hortonworks.com> wrote:
> >
> > > FALCON-1551 was interesting for two reasons - that we are still caught
> by
> > > the FS caching issue (this should have been dealt by most of the
> projects
> > > one time or another), but the use of Oozie uber mode - I am wondering
> if we
> > > have done uber mode testing generally - Is this the mode that is used
> in
> > > your environments – It makes sense when thousands of Oozie jobs are
> run –
> > > saving one container. I would like to understand more about your
> > > experience with uber mode if you have – may be in the next Falcon sync
> up.
> > >
> > > I think we have create a list of modes that we expect the product to
> run
> > > before we introduce changes generally – Given that there are different
> sets
> > > of execution modes that we want to test. Please add more on the
> scenarios
> > > that I am most likely missing. Let us discuss this also if we have
> time.
> > >
> > > 1. Security (Kerberos and Wire encryption)
> > > 2. NN HA/RM HA
> > > 3. Oozie HA
> > > 4. Oozie ubermode for launcher job
> > > 5. Multi cluster with prism
> > > 6. Multi cluster managed with Oozie (with single Falcon instance)
> > > 7. Hadoop/Hive/Oozie versions and the implications (S3 and Azure
> blob
> > > storage went out of hadoop-common in 2.6.0+ and the implications).
> > >
> > >
> > > Thanks
> > >
> > > Venkat
> > >
> > >
> >
> > --
> > _____________________________________________________________
> > The information contained in this communication is intended solely for
> the
> > use of the individual or entity to whom it is addressed and others
> > authorized to receive it. It may contain confidential or legally
> privileged
> > information. If you are not the intended recipient you are hereby
> notified
> > that any disclosure, copying, distribution or taking any action in
> reliance
> > on the contents of this information is strictly prohibited and may be
> > unlawful. If you have received this communication in error, please notify
> > us immediately by responding to this email and then delete it from your
> > system. The firm is neither liable for the proper and complete
> transmission
> > of the information contained in this communication nor for any delay in
> its
> > receipt.
>
>
RE: Oozie uber mode and other modes of running the software
Posted by Srikanth Sundarrajan <sr...@hotmail.com>.
Am guessing venkat you meant to refer FALCON-1555. Good check list to go through for each release.
Specifically on
>> 6. Multi cluster managed with Oozie (with single Falcon instance)
Since we haven't particularly done anything by isolating the classes and their dependencies for each cluster with independent class loader, I think this would work only as long as all the clusters are compatible. If things were to change say with hadoop-3 for ex, this may no longer work.
Regards
Srikanth Sundarrajan
> From: ajay.yadav@inmobi.com
> Date: Sun, 25 Oct 2015 05:29:20 +0530
> Subject: Re: Oozie uber mode and other modes of running the software
> To: dev@falcon.apache.org
>
> Venkat,
>
> Sounds good, let's discuss it in the next sync up. For the benefit of
> everyone on the mailing list, I will summarise our use cases. Yes, we
> (Inmobi) use Oozie's uber mode in our production environment(IIRC we are
> using it since Apache Falcon 0.7). We do have several thousand jobs running
> per day(several at minutely frequencies) and a heavily utilised cluster, so
> this makes sense for us. We do test all Oozie releases, though sometimes
> our Oozie releases are not exactly aligned with Oozie's general release -
> it might contain some extra patches, might be from trunk etc. and our
> testing might be specific to features being used at Inmobi. In fact last
> time Pragya Mittal did the testing for both Apache Falcon 0.7 and Oozie.
> She can add more on this.
>
> List of scenarios looks comprehensive. Just curious, what storage backends
> are other users using for titan?
>
>
> Cheers
> Ajay Yadava
>
> On Sat, Oct 24, 2015 at 9:20 PM, Venkat Ranganathan <
> vranganathan@hortonworks.com> wrote:
>
> > FALCON-1551 was interesting for two reasons - that we are still caught by
> > the FS caching issue (this should have been dealt by most of the projects
> > one time or another), but the use of Oozie uber mode - I am wondering if we
> > have done uber mode testing generally - Is this the mode that is used in
> > your environments – It makes sense when thousands of Oozie jobs are run –
> > saving one container. I would like to understand more about your
> > experience with uber mode if you have – may be in the next Falcon sync up.
> >
> > I think we have create a list of modes that we expect the product to run
> > before we introduce changes generally – Given that there are different sets
> > of execution modes that we want to test. Please add more on the scenarios
> > that I am most likely missing. Let us discuss this also if we have time.
> >
> > 1. Security (Kerberos and Wire encryption)
> > 2. NN HA/RM HA
> > 3. Oozie HA
> > 4. Oozie ubermode for launcher job
> > 5. Multi cluster with prism
> > 6. Multi cluster managed with Oozie (with single Falcon instance)
> > 7. Hadoop/Hive/Oozie versions and the implications (S3 and Azure blob
> > storage went out of hadoop-common in 2.6.0+ and the implications).
> >
> >
> > Thanks
> >
> > Venkat
> >
> >
>
> --
> _____________________________________________________________
> The information contained in this communication is intended solely for the
> use of the individual or entity to whom it is addressed and others
> authorized to receive it. It may contain confidential or legally privileged
> information. If you are not the intended recipient you are hereby notified
> that any disclosure, copying, distribution or taking any action in reliance
> on the contents of this information is strictly prohibited and may be
> unlawful. If you have received this communication in error, please notify
> us immediately by responding to this email and then delete it from your
> system. The firm is neither liable for the proper and complete transmission
> of the information contained in this communication nor for any delay in its
> receipt.
Re: Oozie uber mode and other modes of running the software
Posted by Ajay Yadav <aj...@inmobi.com>.
Venkat,
Sounds good, let's discuss it in the next sync up. For the benefit of
everyone on the mailing list, I will summarise our use cases. Yes, we
(Inmobi) use Oozie's uber mode in our production environment(IIRC we are
using it since Apache Falcon 0.7). We do have several thousand jobs running
per day(several at minutely frequencies) and a heavily utilised cluster, so
this makes sense for us. We do test all Oozie releases, though sometimes
our Oozie releases are not exactly aligned with Oozie's general release -
it might contain some extra patches, might be from trunk etc. and our
testing might be specific to features being used at Inmobi. In fact last
time Pragya Mittal did the testing for both Apache Falcon 0.7 and Oozie.
She can add more on this.
List of scenarios looks comprehensive. Just curious, what storage backends
are other users using for titan?
Cheers
Ajay Yadava
On Sat, Oct 24, 2015 at 9:20 PM, Venkat Ranganathan <
vranganathan@hortonworks.com> wrote:
> FALCON-1551 was interesting for two reasons - that we are still caught by
> the FS caching issue (this should have been dealt by most of the projects
> one time or another), but the use of Oozie uber mode - I am wondering if we
> have done uber mode testing generally - Is this the mode that is used in
> your environments – It makes sense when thousands of Oozie jobs are run –
> saving one container. I would like to understand more about your
> experience with uber mode if you have – may be in the next Falcon sync up.
>
> I think we have create a list of modes that we expect the product to run
> before we introduce changes generally – Given that there are different sets
> of execution modes that we want to test. Please add more on the scenarios
> that I am most likely missing. Let us discuss this also if we have time.
>
> 1. Security (Kerberos and Wire encryption)
> 2. NN HA/RM HA
> 3. Oozie HA
> 4. Oozie ubermode for launcher job
> 5. Multi cluster with prism
> 6. Multi cluster managed with Oozie (with single Falcon instance)
> 7. Hadoop/Hive/Oozie versions and the implications (S3 and Azure blob
> storage went out of hadoop-common in 2.6.0+ and the implications).
>
>
> Thanks
>
> Venkat
>
>
--
_____________________________________________________________
The information contained in this communication is intended solely for the
use of the individual or entity to whom it is addressed and others
authorized to receive it. It may contain confidential or legally privileged
information. If you are not the intended recipient you are hereby notified
that any disclosure, copying, distribution or taking any action in reliance
on the contents of this information is strictly prohibited and may be
unlawful. If you have received this communication in error, please notify
us immediately by responding to this email and then delete it from your
system. The firm is neither liable for the proper and complete transmission
of the information contained in this communication nor for any delay in its
receipt.