kudu-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Weber, Richard" <riwe...@akamai.com>
Subject Re: Feature request for Kudu 1.3.0
Date Fri, 10 Feb 2017 21:23:59 GMT
Comments below



-- Rick Weber



From: Todd Lipcon <todd@cloudera.com>
Reply-To: "user@kudu.apache.org" <user@kudu.apache.org>
Date: Friday, February 10, 2017 at 4:12 PM
To: "user@kudu.apache.org" <user@kudu.apache.org>
Subject: Re: Feature request for Kudu 1.3.0


On Fri, Feb 10, 2017 at 10:32 AM, Weber, Richard <riweber@akamai.com> wrote:

I definitely would push for prioritization on this.


Our main use case is less about multiple racks and failure, and more about functionality during
the install process.  Our clusters are installed in logical regions, and we install 1/3 of
a region at a time.  That means 1/3 of the cluster can be down for the SW install, reboot,
or something else.  Allowing rack locality to be logically defined will allow the data to
still be available during normal maintenance operations.


That's an interesting use case. How long is the 1/3rd of the cluster typically down for? I'd
be afraid that, if it's down for more than a couple minutes, there's a decent chance of losing
one server in the other 2/3 region, which would leave a tablet at 1/3 replication and unavailable
for writes or consistent reads. Is that acceptable for your target use cases?


Nodes would be down typically for 5-15 minutes or so.  Are you saying that if 1 node goes
down, there's an increased chance of one of the other 2 going down as well?  That doesn't
sound good if losing a node increases the instability of the system.  Additionally, wouldn't
the tablets start re-replicating the data if 2/3 of the nodes detect the node is down for
too long?  


How does the system typically handle a node failing?  Is re-replication of data not automatic? 
(I haven't experimented with this enough)


Our install process is along the line of:

1)      copy software to target machine

2)      shut down services on machine

3)      expand software to final location

4)      reboot (if new kernel)

5)      restart services.



There are certain things we could consider doing to allow a replica to fall to 1/3 while still
remaining online, but don't think we've considered doing them any time particularly soon.
Would be good to get data on how important that is.


I don't think it's too critical to have a replica fail down to 1/3 and still be available,
or at least accept writes.  It'd be great to have it service read's at least.  But in our
use case, we're looking at (relatively) quick bounces for installs and potential reboots










-- Rick Weber



From: Todd Lipcon <todd@cloudera.com>
Reply-To: "user@kudu.apache.org" <user@kudu.apache.org>
Date: Friday, February 10, 2017 at 12:45 PM
To: "user@kudu.apache.org" <user@kudu.apache.org>
Subject: Re: Feature request for Kudu 1.3.0


Hi Jeff, 


Thanks for the input on prioritization.


I'm curious: do you have more than two racks in your cluster? With Kudu's replication strategy,
we need at least three racks to be able to survive a full rack outage. (with just two racks
it's impossible to distinguish a loss of a rack with a partition between the two racks).




On Fri, Feb 10, 2017 at 7:27 AM, Jeff Dasch <jeff_dasch@yahoo.com> wrote:

Any chance we can get a fix for KUDU-1535 "Add rack awareness" added to the 1.3.0 release?


While I appreciate the need for Kerberos and TLS for some production systems, for my use-case
data availability really takes priority.


I looked at your scoping document, and for what it's worth, I'm fine with a shell script that
is similar to what Hadoop uses.









Todd Lipcon
Software Engineer, Cloudera



Todd Lipcon
Software Engineer, Cloudera

View raw message