Accumulo upgrade rollback...

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

Accumulo upgrade rollback...

Robert Rosekrans

Accumulo team,

Our team is upgrading from Accumulo 1.5.4 to 1.6.6 then to 1.7.3.  We are testing the upgrade in our integration environment. Before we started the upgrade, we copied our HDFS /accumulo directory to /accumulo_backup. My hope was that I could just swap out the directories to rollback. However when I did, the metadata table didn't load. I looked at the tables and the metadata table was listed as "metadata" versus "!METADATA" indicating it was using the new format of the tables.  My guess is that zookeeper is maintaining the previous upgraded table names.

My question is what is the easiest way to rollback the data.

Thanks
Bob


Sent from my Verizon, Samsung Galaxy smartphone
Reply | Threaded
Open this post in threaded view
|

Re: Accumulo upgrade rollback...

Michael Wall
Hi Bob,

Unfortunately, rolling back an upgraded system is not an automated task.
Moving from 1.5.4 to 1.6.6 as you saw !METADATA changed to
accumulo.metadata and accumulo.root.  The upgrade also changes the data
version of the rfiles, so any new files written from mutations or
compactions will not be readable by prior versions.  Good for you that you
saved off the /accumulo directory in HDFS.

What you are going to need to do is reinitialize Accumulo, create the
tables with whatever splits and configs you need, then bulk import the
rfiles from your /accumulo_backup directory in the new tables.  That will
make the metadata and zookeeper happy.  You can follow the directions for a
Zookeeper failure at
http://accumulo.apache.org/1.8/accumulo_user_manual.html#zookeeper_failure.

Interested in what you are seeing that made you want to rollback.  If you
are testing, you may want to try upgrading to the latest 1.8, which is
currently 1.8.1.

Let us know how it goes.

Mike



On Tue, Jun 6, 2017 at 2:25 AM Robert Rosekrans <
[hidden email]> wrote:

>
> Accumulo team,
>
> Our team is upgrading from Accumulo 1.5.4 to 1.6.6 then to 1.7.3.  We are
> testing the upgrade in our integration environment. Before we started the
> upgrade, we copied our HDFS /accumulo directory to /accumulo_backup. My
> hope was that I could just swap out the directories to rollback. However
> when I did, the metadata table didn't load. I looked at the tables and the
> metadata table was listed as "metadata" versus "!METADATA" indicating it
> was using the new format of the tables.  My guess is that zookeeper is
> maintaining the previous upgraded table names.
>
> My question is what is the easiest way to rollback the data.
>
> Thanks
> Bob
>
>
> Sent from my Verizon, Samsung Galaxy smartphone
>
Reply | Threaded
Open this post in threaded view
|

RE: EXTERNAL: Re: Accumulo upgrade rollback...

Rosekrans, Robert

Mike,

Thanks for the quick response.  It was very helpful and what I expected.

Thanks for the link to the zookeeper failure docs.

Long story short, we decided to go to 1.7.3 for now and then look at 1.8.1 in a couple months after it has a little more soak time.  The only reason for the rollback test was in case our production upgrade has any issues.

Again thanks for the quick reply.

Bob

Robert Rosekrans
iNovex Information Systems
Analytic Portfolio Tech Lead
(C) 410-303-9144
(W) 410-379-5311

From: Michael Wall [mailto:[hidden email]]
Sent: Tuesday, June 06, 2017 8:42 AM
To: [hidden email]; [hidden email]
Cc: Rosekrans, Robert (US N-INOVEX INFORMATION SYSTEMS INC.) <[hidden email]>
Subject: EXTERNAL: Re: Accumulo upgrade rollback...

Hi Bob,

Unfortunately, rolling back an upgraded system is not an automated task.  Moving from 1.5.4 to 1.6.6 as you saw !METADATA changed to accumulo.metadata and accumulo.root.  The upgrade also changes the data version of the rfiles, so any new files written from mutations or compactions will not be readable by prior versions.  Good for you that you saved off the /accumulo directory in HDFS.

What you are going to need to do is reinitialize Accumulo, create the tables with whatever splits and configs you need, then bulk import the rfiles from your /accumulo_backup directory in the new tables.  That will make the metadata and zookeeper happy.  You can follow the directions for a Zookeeper failure at http://accumulo.apache.org/1.8/accumulo_user_manual.html#zookeeper_failure.

Interested in what you are seeing that made you want to rollback.  If you are testing, you may want to try upgrading to the latest 1.8, which is currently 1.8.1.

Let us know how it goes.

Mike



On Tue, Jun 6, 2017 at 2:25 AM Robert Rosekrans <[hidden email]<mailto:[hidden email]>> wrote:

Accumulo team,

Our team is upgrading from Accumulo 1.5.4 to 1.6.6 then to 1.7.3.  We are testing the upgrade in our integration environment. Before we started the upgrade, we copied our HDFS /accumulo directory to /accumulo_backup. My hope was that I could just swap out the directories to rollback. However when I did, the metadata table didn't load. I looked at the tables and the metadata table was listed as "metadata" versus "!METADATA" indicating it was using the new format of the tables.  My guess is that zookeeper is maintaining the previous upgraded table names.

My question is what is the easiest way to rollback the data.

Thanks
Bob


Sent from my Verizon, Samsung Galaxy smartphone