You are viewing a plain text version of this content. The canonical link for it is here.
Posted to solr-user@lucene.apache.org by Thomas Heigl <th...@umschalt.com> on 2012/12/06 11:46:49 UTC

Minimum HA Setup with SolrCloud

Hey all,

I'm in the process of migrating a single Solr 4.0 instace to a SolrCloud
setup for availability reasons.

After studying the wiki page for SolrCloud I'm not sure what the absolute
minimum setup is that would allow for one machine to go down.

Would it be enough to have one shard with one leader and one replica? Could
this setup tolerate one of the instances going down? Or do I need three
instance because Zookeeper needs a quorum of instances?

Cheers,

Thomas

Re: Minimum HA Setup with SolrCloud

Posted by Thomas Heigl <th...@umschalt.com>.
Thanks a lot guys!

On Thu, Dec 6, 2012 at 4:22 PM, Markus Jelsma <ma...@openindex.io>wrote:

>
> -----Original message-----
> > From:Yonik Seeley <yo...@lucidworks.com>
> > Sent: Thu 06-Dec-2012 16:01
> > To: solr-user@lucene.apache.org
> > Subject: Re: Minimum HA Setup with SolrCloud
> >
> > On Thu, Dec 6, 2012 at 9:56 AM, Markus Jelsma
> > <ma...@openindex.io> wrote:
> > > The quorum is the minimun, so it depends on how many you have running
> in the ensemble. If it's three or four, then two is the quorum
> >
> > I think that for 4 ZK servers, then 3 would be the quorum?
>
> Yes, my wrong. I meant to say that four nodes does not allow an extra ZK
> to go down because of the quorum.
>
> >
> > -Yonik
> > http://lucidworks.com
> >
>

RE: Minimum HA Setup with SolrCloud

Posted by Markus Jelsma <ma...@openindex.io>.
-----Original message-----
> From:Yonik Seeley <yo...@lucidworks.com>
> Sent: Thu 06-Dec-2012 16:01
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
> 
> On Thu, Dec 6, 2012 at 9:56 AM, Markus Jelsma
> <ma...@openindex.io> wrote:
> > The quorum is the minimun, so it depends on how many you have running in the ensemble. If it's three or four, then two is the quorum
> 
> I think that for 4 ZK servers, then 3 would be the quorum?

Yes, my wrong. I meant to say that four nodes does not allow an extra ZK to go down because of the quorum.

> 
> -Yonik
> http://lucidworks.com
> 

Re: Minimum HA Setup with SolrCloud

Posted by Mark Miller <ma...@gmail.com>.
On Dec 6, 2012, at 6:54 AM, Yonik Seeley <yo...@lucidworks.com> wrote:

> On Thu, Dec 6, 2012 at 9:56 AM, Markus Jelsma
> <ma...@openindex.io> wrote:
>> The quorum is the minimun, so it depends on how many you have running in the ensemble. If it's three or four, then two is the quorum
> 
> I think that for 4 ZK servers, then 3 would be the quorum?
> 

Yup - 4 requires 3 as a quorum, same as 5. So 4 allows one to go down, 5 allows 2 to go down.

- Mark


Re: Minimum HA Setup with SolrCloud

Posted by Yonik Seeley <yo...@lucidworks.com>.
On Thu, Dec 6, 2012 at 9:56 AM, Markus Jelsma
<ma...@openindex.io> wrote:
> The quorum is the minimun, so it depends on how many you have running in the ensemble. If it's three or four, then two is the quorum

I think that for 4 ZK servers, then 3 would be the quorum?

-Yonik
http://lucidworks.com

RE: Minimum HA Setup with SolrCloud

Posted by Markus Jelsma <ma...@openindex.io>.
The quorum is the minimun, so it depends on how many you have running in the ensemble. If it's three or four, then two is the quorum and therefore the minumum. Three is regarded as a minumum in the ensemble because two makes no sense.
 
-----Original message-----
> From:Jack Krupansky <ja...@basetechnology.com>
> Sent: Thu 06-Dec-2012 15:53
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
> 
> "but if those same machines are running zookeeper, you need 3."
> 
> And one of those 3 can go down? I thought 3 was the minimum number of 
> zookeepers.
> 
> -- Jack Krupansky
> 
> -----Original Message----- 
> From: Mark Miller
> Sent: Thursday, December 06, 2012 9:30 AM
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
> 
> It depends on if you are running embedded zk or an external zk ensemble.
> 
> One leader and a replica is all you need for Solr to allow on machine to go 
> down - but if those same machines are running zookeeper, you need 3.
> 
> You could also run zookeeper on one external machine and then it would be 
> fine if you lost one solr node - but if you the one external zk node went 
> down you would lose the ability to do updates until it was brought back up.
> 
> - Mark
> 
> On Dec 6, 2012, at 2:46 AM, Thomas Heigl <th...@umschalt.com> wrote:
> 
> > Hey all,
> >
> > I'm in the process of migrating a single Solr 4.0 instace to a SolrCloud
> > setup for availability reasons.
> >
> > After studying the wiki page for SolrCloud I'm not sure what the absolute
> > minimum setup is that would allow for one machine to go down.
> >
> > Would it be enough to have one shard with one leader and one replica? 
> > Could
> > this setup tolerate one of the instances going down? Or do I need three
> > instance because Zookeeper needs a quorum of instances?
> >
> > Cheers,
> >
> > Thomas 
> 
> 

Re: Minimum HA Setup with SolrCloud

Posted by Yonik Seeley <yo...@lucidworks.com>.
On Thu, Dec 6, 2012 at 5:21 PM, Jack Krupansky <ja...@basetechnology.com> wrote:
> If 1 is the minimum, what is the 3 "minimum" all about?

The minimum for running an ensemble (a cluster) and having any sort of
fault tolerance?

> The zk web page does say "Three ZooKeeper servers is the minimum recommended
> size for an ensemble, and we also recommend that they run on separate
> machines" - but it does say "recommended".
>
> But back to the original question - it sounds as if 4 nodes would the
> recommended minimum number of nodes if you want to tolerate one machine
> going down and maintaining that 3 zookeeper recommended minimum ensemble.

Nope - the "ensemble" are not the number of servers that are currently
up, but the number in the cluster.  This is important to prevent split
brain.
3 ZK servers means that a majority of the servers is 2, hence you can lose 1.
5 ZK servers means that a majority of the servers is 3, hence you can lose 2.

-Yonik
http://lucidworks.com

Re: Minimum HA Setup with SolrCloud

Posted by Walter Underwood <wu...@wunderwood.org>.
And that link includes this sentence: "For example, with four machines ZooKeeper can only handle the failure of a single machine; if two machines fail, the remaining two machines do not constitute a majority."

wunder

On Dec 6, 2012, at 2:25 PM, Jack Krupansky wrote:

> Slightly more recent link:
> http://zookeeper.apache.org/doc/r3.4.5/zookeeperAdmin.html
> 
> -- Jack Krupansky
> 
> -----Original Message----- From: Jack Krupansky
> Sent: Thursday, December 06, 2012 5:21 PM
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
> 
> Rewind.
> 
> If 1 is the minimum, what is the 3 "minimum" all about?
> 
> The zk web page does say "Three ZooKeeper servers is the minimum recommended
> size for an ensemble, and we also recommend that they run on separate
> machines" - but it does say "recommended".
> 
> But back to the original question - it sounds as if 4 nodes would the
> recommended minimum number of nodes if you want to tolerate one machine
> going down and maintaining that 3 zookeeper recommended minimum ensemble.
> 
> From the zookeeper web page:
> 
> "For reliable ZooKeeper service, you should deploy ZooKeeper in a cluster
> known as an ensemble. As long as a majority of the ensemble are up, the
> service will be available. Because Zookeeper requires a majority, it is best
> to use an odd number of machines. For example, with four machines ZooKeeper
> can only handle the failure of a single machine; if two machines fail, the
> remaining two machines do not constitute a majority. However, with five
> machines ZooKeeper can handle the failure of two machines."
> 
> See:
> http://zookeeper.apache.org/doc/r3.1.2/zookeeperAdmin.html
> 
> -- Jack Krupansky
> 
> -----Original Message----- From: Otis Gospodnetic
> Sent: Thursday, December 06, 2012 4:39 PM
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
> 
> 1 is the minimum :)
> 2 makes no sense.
> 3 must be the most common number in the zoo.
> 
> Otis
> --
> Performance Monitoring - http://sematext.com/spm/index.html
> Search Analytics - http://sematext.com/search-analytics/index.html
> 
> 
> 
> 
> On Thu, Dec 6, 2012 at 9:46 AM, Jack Krupansky
> <ja...@basetechnology.com>wrote:
> 
>> "but if those same machines are running zookeeper, you need 3."
>> 
>> And one of those 3 can go down? I thought 3 was the minimum number of
>> zookeepers.
>> 
>> -- Jack Krupansky
>> 
>> -----Original Message----- From: Mark Miller
>> Sent: Thursday, December 06, 2012 9:30 AM
>> To: solr-user@lucene.apache.org
>> Subject: Re: Minimum HA Setup with SolrCloud
>> 
>> 
>> It depends on if you are running embedded zk or an external zk ensemble.
>> 
>> One leader and a replica is all you need for Solr to allow on machine to
>> go down - but if those same machines are running zookeeper, you need 3.
>> 
>> You could also run zookeeper on one external machine and then it would be
>> fine if you lost one solr node - but if you the one external zk node went
>> down you would lose the ability to do updates until it was brought back up.
>> 
>> - Mark
>> 
>> On Dec 6, 2012, at 2:46 AM, Thomas Heigl <th...@umschalt.com> wrote:
>> 
>> Hey all,
>>> 
>>> I'm in the process of migrating a single Solr 4.0 instace to a SolrCloud
>>> setup for availability reasons.
>>> 
>>> After studying the wiki page for SolrCloud I'm not sure what the absolute
>>> minimum setup is that would allow for one machine to go down.
>>> 
>>> Would it be enough to have one shard with one leader and one replica?
>>> Could
>>> this setup tolerate one of the instances going down? Or do I need three
>>> instance because Zookeeper needs a quorum of instances?
>>> 
>>> Cheers,
>>> 
>>> Thomas
>>> 
>> 

--
Walter Underwood
wunder@wunderwood.org




Re: Minimum HA Setup with SolrCloud

Posted by Mark Miller <ma...@gmail.com>.
On Dec 6, 2012, at 2:32 PM, Michael Della Bitta <mi...@appinions.com> wrote:

> I just rethought what I wrote and it doesn't make any sense. :)
> 
> If you have two remaining nodes left when you have a three node ensemble,
> how are ties broken? Or does Zookeeper not resolve ties since it doesn't
> tolerate partitions?
> 
> Michael

Tie breaking happens when you lose nodes or have partitions.

So if you have 3 nodes and its partitioned into 2 and 1, the 2 nodes "wins" and the 1 node is 'dead'.

You can't then survive a another partition of the 2 nodes - both 1 node ensembles would be 'dead'.

- Mark

Re: Minimum HA Setup with SolrCloud

Posted by Michael Della Bitta <mi...@appinions.com>.
I just rethought what I wrote and it doesn't make any sense. :)

If you have two remaining nodes left when you have a three node ensemble,
how are ties broken? Or does Zookeeper not resolve ties since it doesn't
tolerate partitions?

Michael


Michael Della Bitta

------------------------------------------------
Appinions
18 East 41st Street, 2nd Floor
New York, NY 10017-6271

www.appinions.com

Where Influence Isn’t a Game


On Thu, Dec 6, 2012 at 5:30 PM, Michael Della Bitta <
michael.della.bitta@appinions.com> wrote:

> Jack,
>
> The recommended ensemble configured size takes into consideration that you
> might have a node failure. You can still run with two while you replace the
> third, so it's sort of like RAID-5.
>
> If you run with four configured nodes, you're still running with
> RAID-5-like failure survival characteristics. Two configured can't be
> the minimum because the minimum requires an odd number of machines to break
> ties, so your minimum is three when you configure for four.
>
> Configuring for five nodes lets you survive two failures because you now
> have a minimum of three
>
>
> Michael Della Bitta
>
> ------------------------------------------------
> Appinions
> 18 East 41st Street, 2nd Floor
> New York, NY 10017-6271
>
> www.appinions.com
>
> Where Influence Isn’t a Game
>
>
> On Thu, Dec 6, 2012 at 5:25 PM, Jack Krupansky <ja...@basetechnology.com>wrote:
>
>> Slightly more recent link:
>> http://zookeeper.apache.org/**doc/r3.4.5/zookeeperAdmin.html<http://zookeeper.apache.org/doc/r3.4.5/zookeeperAdmin.html>
>>
>> -- Jack Krupansky
>>
>> -----Original Message----- From: Jack Krupansky
>> Sent: Thursday, December 06, 2012 5:21 PM
>>
>> To: solr-user@lucene.apache.org
>> Subject: Re: Minimum HA Setup with SolrCloud
>>
>> Rewind.
>>
>> If 1 is the minimum, what is the 3 "minimum" all about?
>>
>> The zk web page does say "Three ZooKeeper servers is the minimum
>> recommended
>> size for an ensemble, and we also recommend that they run on separate
>> machines" - but it does say "recommended".
>>
>> But back to the original question - it sounds as if 4 nodes would the
>> recommended minimum number of nodes if you want to tolerate one machine
>> going down and maintaining that 3 zookeeper recommended minimum ensemble.
>>
>> From the zookeeper web page:
>>
>> "For reliable ZooKeeper service, you should deploy ZooKeeper in a cluster
>> known as an ensemble. As long as a majority of the ensemble are up, the
>> service will be available. Because Zookeeper requires a majority, it is
>> best
>> to use an odd number of machines. For example, with four machines
>> ZooKeeper
>> can only handle the failure of a single machine; if two machines fail, the
>> remaining two machines do not constitute a majority. However, with five
>> machines ZooKeeper can handle the failure of two machines."
>>
>> See:
>> http://zookeeper.apache.org/**doc/r3.1.2/zookeeperAdmin.html<http://zookeeper.apache.org/doc/r3.1.2/zookeeperAdmin.html>
>>
>> -- Jack Krupansky
>>
>> -----Original Message----- From: Otis Gospodnetic
>> Sent: Thursday, December 06, 2012 4:39 PM
>> To: solr-user@lucene.apache.org
>> Subject: Re: Minimum HA Setup with SolrCloud
>>
>> 1 is the minimum :)
>> 2 makes no sense.
>> 3 must be the most common number in the zoo.
>>
>> Otis
>> --
>> Performance Monitoring - http://sematext.com/spm/index.**html<http://sematext.com/spm/index.html>
>> Search Analytics - http://sematext.com/search-**analytics/index.html<http://sematext.com/search-analytics/index.html>
>>
>>
>>
>>
>> On Thu, Dec 6, 2012 at 9:46 AM, Jack Krupansky
>> <ja...@basetechnology.com>**wrote:
>>
>>  "but if those same machines are running zookeeper, you need 3."
>>>
>>> And one of those 3 can go down? I thought 3 was the minimum number of
>>> zookeepers.
>>>
>>> -- Jack Krupansky
>>>
>>> -----Original Message----- From: Mark Miller
>>> Sent: Thursday, December 06, 2012 9:30 AM
>>> To: solr-user@lucene.apache.org
>>> Subject: Re: Minimum HA Setup with SolrCloud
>>>
>>>
>>> It depends on if you are running embedded zk or an external zk ensemble.
>>>
>>> One leader and a replica is all you need for Solr to allow on machine to
>>> go down - but if those same machines are running zookeeper, you need 3.
>>>
>>> You could also run zookeeper on one external machine and then it would be
>>> fine if you lost one solr node - but if you the one external zk node went
>>> down you would lose the ability to do updates until it was brought back
>>> up.
>>>
>>> - Mark
>>>
>>> On Dec 6, 2012, at 2:46 AM, Thomas Heigl <th...@umschalt.com> wrote:
>>>
>>>  Hey all,
>>>
>>>>
>>>> I'm in the process of migrating a single Solr 4.0 instace to a SolrCloud
>>>> setup for availability reasons.
>>>>
>>>> After studying the wiki page for SolrCloud I'm not sure what the
>>>> absolute
>>>> minimum setup is that would allow for one machine to go down.
>>>>
>>>> Would it be enough to have one shard with one leader and one replica?
>>>> Could
>>>> this setup tolerate one of the instances going down? Or do I need three
>>>> instance because Zookeeper needs a quorum of instances?
>>>>
>>>> Cheers,
>>>>
>>>> Thomas
>>>>
>>>>
>>>
>>>
>

Re: Minimum HA Setup with SolrCloud

Posted by Michael Della Bitta <mi...@appinions.com>.
Jack,

The recommended ensemble configured size takes into consideration that you
might have a node failure. You can still run with two while you replace the
third, so it's sort of like RAID-5.

If you run with four configured nodes, you're still running with
RAID-5-like failure survival characteristics. Two configured can't be
the minimum because the minimum requires an odd number of machines to break
ties, so your minimum is three when you configure for four.

Configuring for five nodes lets you survive two failures because you now
have a minimum of three

Michael Della Bitta

------------------------------------------------
Appinions
18 East 41st Street, 2nd Floor
New York, NY 10017-6271

www.appinions.com

Where Influence Isn’t a Game


On Thu, Dec 6, 2012 at 5:25 PM, Jack Krupansky <ja...@basetechnology.com>wrote:

> Slightly more recent link:
> http://zookeeper.apache.org/**doc/r3.4.5/zookeeperAdmin.html<http://zookeeper.apache.org/doc/r3.4.5/zookeeperAdmin.html>
>
> -- Jack Krupansky
>
> -----Original Message----- From: Jack Krupansky
> Sent: Thursday, December 06, 2012 5:21 PM
>
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
>
> Rewind.
>
> If 1 is the minimum, what is the 3 "minimum" all about?
>
> The zk web page does say "Three ZooKeeper servers is the minimum
> recommended
> size for an ensemble, and we also recommend that they run on separate
> machines" - but it does say "recommended".
>
> But back to the original question - it sounds as if 4 nodes would the
> recommended minimum number of nodes if you want to tolerate one machine
> going down and maintaining that 3 zookeeper recommended minimum ensemble.
>
> From the zookeeper web page:
>
> "For reliable ZooKeeper service, you should deploy ZooKeeper in a cluster
> known as an ensemble. As long as a majority of the ensemble are up, the
> service will be available. Because Zookeeper requires a majority, it is
> best
> to use an odd number of machines. For example, with four machines ZooKeeper
> can only handle the failure of a single machine; if two machines fail, the
> remaining two machines do not constitute a majority. However, with five
> machines ZooKeeper can handle the failure of two machines."
>
> See:
> http://zookeeper.apache.org/**doc/r3.1.2/zookeeperAdmin.html<http://zookeeper.apache.org/doc/r3.1.2/zookeeperAdmin.html>
>
> -- Jack Krupansky
>
> -----Original Message----- From: Otis Gospodnetic
> Sent: Thursday, December 06, 2012 4:39 PM
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
>
> 1 is the minimum :)
> 2 makes no sense.
> 3 must be the most common number in the zoo.
>
> Otis
> --
> Performance Monitoring - http://sematext.com/spm/index.**html<http://sematext.com/spm/index.html>
> Search Analytics - http://sematext.com/search-**analytics/index.html<http://sematext.com/search-analytics/index.html>
>
>
>
>
> On Thu, Dec 6, 2012 at 9:46 AM, Jack Krupansky
> <ja...@basetechnology.com>**wrote:
>
>  "but if those same machines are running zookeeper, you need 3."
>>
>> And one of those 3 can go down? I thought 3 was the minimum number of
>> zookeepers.
>>
>> -- Jack Krupansky
>>
>> -----Original Message----- From: Mark Miller
>> Sent: Thursday, December 06, 2012 9:30 AM
>> To: solr-user@lucene.apache.org
>> Subject: Re: Minimum HA Setup with SolrCloud
>>
>>
>> It depends on if you are running embedded zk or an external zk ensemble.
>>
>> One leader and a replica is all you need for Solr to allow on machine to
>> go down - but if those same machines are running zookeeper, you need 3.
>>
>> You could also run zookeeper on one external machine and then it would be
>> fine if you lost one solr node - but if you the one external zk node went
>> down you would lose the ability to do updates until it was brought back
>> up.
>>
>> - Mark
>>
>> On Dec 6, 2012, at 2:46 AM, Thomas Heigl <th...@umschalt.com> wrote:
>>
>>  Hey all,
>>
>>>
>>> I'm in the process of migrating a single Solr 4.0 instace to a SolrCloud
>>> setup for availability reasons.
>>>
>>> After studying the wiki page for SolrCloud I'm not sure what the absolute
>>> minimum setup is that would allow for one machine to go down.
>>>
>>> Would it be enough to have one shard with one leader and one replica?
>>> Could
>>> this setup tolerate one of the instances going down? Or do I need three
>>> instance because Zookeeper needs a quorum of instances?
>>>
>>> Cheers,
>>>
>>> Thomas
>>>
>>>
>>
>>

Re: Minimum HA Setup with SolrCloud

Posted by Jack Krupansky <ja...@basetechnology.com>.
Slightly more recent link:
http://zookeeper.apache.org/doc/r3.4.5/zookeeperAdmin.html

-- Jack Krupansky

-----Original Message----- 
From: Jack Krupansky
Sent: Thursday, December 06, 2012 5:21 PM
To: solr-user@lucene.apache.org
Subject: Re: Minimum HA Setup with SolrCloud

Rewind.

If 1 is the minimum, what is the 3 "minimum" all about?

The zk web page does say "Three ZooKeeper servers is the minimum recommended
size for an ensemble, and we also recommend that they run on separate
machines" - but it does say "recommended".

But back to the original question - it sounds as if 4 nodes would the
recommended minimum number of nodes if you want to tolerate one machine
going down and maintaining that 3 zookeeper recommended minimum ensemble.

>From the zookeeper web page:

"For reliable ZooKeeper service, you should deploy ZooKeeper in a cluster
known as an ensemble. As long as a majority of the ensemble are up, the
service will be available. Because Zookeeper requires a majority, it is best
to use an odd number of machines. For example, with four machines ZooKeeper
can only handle the failure of a single machine; if two machines fail, the
remaining two machines do not constitute a majority. However, with five
machines ZooKeeper can handle the failure of two machines."

See:
http://zookeeper.apache.org/doc/r3.1.2/zookeeperAdmin.html

-- Jack Krupansky

-----Original Message----- 
From: Otis Gospodnetic
Sent: Thursday, December 06, 2012 4:39 PM
To: solr-user@lucene.apache.org
Subject: Re: Minimum HA Setup with SolrCloud

1 is the minimum :)
2 makes no sense.
3 must be the most common number in the zoo.

Otis
--
Performance Monitoring - http://sematext.com/spm/index.html
Search Analytics - http://sematext.com/search-analytics/index.html




On Thu, Dec 6, 2012 at 9:46 AM, Jack Krupansky
<ja...@basetechnology.com>wrote:

> "but if those same machines are running zookeeper, you need 3."
>
> And one of those 3 can go down? I thought 3 was the minimum number of
> zookeepers.
>
> -- Jack Krupansky
>
> -----Original Message----- From: Mark Miller
> Sent: Thursday, December 06, 2012 9:30 AM
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
>
>
> It depends on if you are running embedded zk or an external zk ensemble.
>
> One leader and a replica is all you need for Solr to allow on machine to
> go down - but if those same machines are running zookeeper, you need 3.
>
> You could also run zookeeper on one external machine and then it would be
> fine if you lost one solr node - but if you the one external zk node went
> down you would lose the ability to do updates until it was brought back 
> up.
>
> - Mark
>
> On Dec 6, 2012, at 2:46 AM, Thomas Heigl <th...@umschalt.com> wrote:
>
>  Hey all,
>>
>> I'm in the process of migrating a single Solr 4.0 instace to a SolrCloud
>> setup for availability reasons.
>>
>> After studying the wiki page for SolrCloud I'm not sure what the absolute
>> minimum setup is that would allow for one machine to go down.
>>
>> Would it be enough to have one shard with one leader and one replica?
>> Could
>> this setup tolerate one of the instances going down? Or do I need three
>> instance because Zookeeper needs a quorum of instances?
>>
>> Cheers,
>>
>> Thomas
>>
>
> 

Re: Minimum HA Setup with SolrCloud

Posted by Mark Miller <ma...@gmail.com>.
On Dec 6, 2012, at 2:55 PM, Jack Krupansky <ja...@basetechnology.com> wrote:

> I mean, there must be some hard-core downside, other than that you can't lose any more nodes.

Nope, not really. You just can't lose any more nodes.

Technically, you will also lose a bit of read performance - but write performance generally declines with more nodes, so you'd also gain a bit of write performance. But for Solr this should matter little to none.

- Mark

Re: Minimum HA Setup with SolrCloud

Posted by Jack Krupansky <ja...@basetechnology.com>.
And this is precisely why the mystery remains - because you're only 
describing half the picture! Describe the rest of the picture - including 
what exactly those two zks can and can't do, including resolution of ties 
and the concept of "constitu.

-- Jack Krupansky

-----Original Message----- 
From: Walter Underwood
Sent: Thursday, December 06, 2012 8:33 PM
To: solr-user@lucene.apache.org
Subject: Re: Minimum HA Setup with SolrCloud

Configure an ensemble of three. When one goes down, you still have an 
ensemble of three, but with one down. The ensemble size is not reset after 
failures.

wunder

On Dec 6, 2012, at 5:20 PM, Jack Krupansky wrote:

> The part I still find confusing is that if you start with 3 and lose 1, 
> your have 2, which means you can't always break a tie, right? How is this 
> explained? As opposed to saying that 4 is the minimum if you need to 
> tolerate a loss of 1.
>
> -- Jack Krupansky
>
> -----Original Message----- From: Walter Underwood
> Sent: Thursday, December 06, 2012 7:51 PM
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
>
> What is the mystery? Two is not more than half of four. Therefore, two 
> machines is not a quorum for a four machine Zookeeper ensemble.
>
> wunder
>
> On Dec 6, 2012, at 4:50 PM, Jack Krupansky wrote:
>
>> It's still an unresolved mystery, for now.
>>
>> -- Jack Krupansky
>>
>> -----Original Message----- From: Walter Underwood
>> Sent: Thursday, December 06, 2012 7:30 PM
>> To: solr-user@lucene.apache.org
>> Subject: Re: Minimum HA Setup with SolrCloud
>>
>> The Zookeeper ensemble knows the total size. It does not adjust it each 
>> time that a machine is partitioned or down.
>>
>> Two machines is not a quorum for a four machine ensemble.
>>
>> Why do you think that the documentation would get this wrong?
>>
>> wunder
>>
>> On Dec 6, 2012, at 4:14 PM, Jack Krupansky wrote:
>>
>>> But that is the context I was originally referring to - that with 4 zk 
>>> you can lose only one, that you can't lose two. So, if you want to 
>>> tolerate a loss on one, 4 zk would be the minimum... but then it was 
>>> claimed that you COULD start with 3 zk and loss of one would be fine. I 
>>> mean whether you start with 4 and lose 2 or start with 3 and lose 1 is 
>>> the same, right?
>>>
>>> -- Jack Krupansky
>>>
>>> -----Original Message----- From: Yonik Seeley
>>> Sent: Thursday, December 06, 2012 6:34 PM
>>> To: solr-user@lucene.apache.org
>>> Subject: Re: Minimum HA Setup with SolrCloud
>>>
>>> On Thu, Dec 6, 2012 at 5:55 PM, Jack Krupansky <ja...@basetechnology.com> 
>>> wrote:
>>>> I trust that you have the right answer, Mark, but maybe I'm just 
>>>> struggling
>>>> to parse this statement: "the remaining two machines do not constitute 
>>>> a
>>>> majority."
>>>>
>>>> If you start with 3 zk and lose one, you have an ensemble that does not
>>>> "constitute a majority".
>>>
>>> I think you took that out of context.  They were talking about losing
>>> 2 nodes in a 4 node cluster.
>>>
>>> "For example, with four machines ZooKeeper can only handle the failure
>>> of a single machine; if two machines fail, the remaining two machines
>>> do not constitute a majority."
>>>
>>> -Yonik
>>> http://lucidworks.com
>>
>> --
>> Walter Underwood
>> wunder@wunderwood.org
>>
>>
>>
>
> --
> Walter Underwood
> wunder@wunderwood.org
>
>
>

--
Walter Underwood
wunder@wunderwood.org




Re: Minimum HA Setup with SolrCloud

Posted by Walter Underwood <wu...@wunderwood.org>.
Configure an ensemble of three. When one goes down, you still have an ensemble of three, but with one down. The ensemble size is not reset after failures.

wunder

On Dec 6, 2012, at 5:20 PM, Jack Krupansky wrote:

> The part I still find confusing is that if you start with 3 and lose 1, your have 2, which means you can't always break a tie, right? How is this explained? As opposed to saying that 4 is the minimum if you need to tolerate a loss of 1.
> 
> -- Jack Krupansky
> 
> -----Original Message----- From: Walter Underwood
> Sent: Thursday, December 06, 2012 7:51 PM
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
> 
> What is the mystery? Two is not more than half of four. Therefore, two machines is not a quorum for a four machine Zookeeper ensemble.
> 
> wunder
> 
> On Dec 6, 2012, at 4:50 PM, Jack Krupansky wrote:
> 
>> It's still an unresolved mystery, for now.
>> 
>> -- Jack Krupansky
>> 
>> -----Original Message----- From: Walter Underwood
>> Sent: Thursday, December 06, 2012 7:30 PM
>> To: solr-user@lucene.apache.org
>> Subject: Re: Minimum HA Setup with SolrCloud
>> 
>> The Zookeeper ensemble knows the total size. It does not adjust it each time that a machine is partitioned or down.
>> 
>> Two machines is not a quorum for a four machine ensemble.
>> 
>> Why do you think that the documentation would get this wrong?
>> 
>> wunder
>> 
>> On Dec 6, 2012, at 4:14 PM, Jack Krupansky wrote:
>> 
>>> But that is the context I was originally referring to - that with 4 zk you can lose only one, that you can't lose two. So, if you want to tolerate a loss on one, 4 zk would be the minimum... but then it was claimed that you COULD start with 3 zk and loss of one would be fine. I mean whether you start with 4 and lose 2 or start with 3 and lose 1 is the same, right?
>>> 
>>> -- Jack Krupansky
>>> 
>>> -----Original Message----- From: Yonik Seeley
>>> Sent: Thursday, December 06, 2012 6:34 PM
>>> To: solr-user@lucene.apache.org
>>> Subject: Re: Minimum HA Setup with SolrCloud
>>> 
>>> On Thu, Dec 6, 2012 at 5:55 PM, Jack Krupansky <ja...@basetechnology.com> wrote:
>>>> I trust that you have the right answer, Mark, but maybe I'm just struggling
>>>> to parse this statement: "the remaining two machines do not constitute a
>>>> majority."
>>>> 
>>>> If you start with 3 zk and lose one, you have an ensemble that does not
>>>> "constitute a majority".
>>> 
>>> I think you took that out of context.  They were talking about losing
>>> 2 nodes in a 4 node cluster.
>>> 
>>> "For example, with four machines ZooKeeper can only handle the failure
>>> of a single machine; if two machines fail, the remaining two machines
>>> do not constitute a majority."
>>> 
>>> -Yonik
>>> http://lucidworks.com
>> 
>> --
>> Walter Underwood
>> wunder@wunderwood.org
>> 
>> 
>> 
> 
> --
> Walter Underwood
> wunder@wunderwood.org
> 
> 
> 

--
Walter Underwood
wunder@wunderwood.org




Re: Minimum HA Setup with SolrCloud

Posted by Jack Krupansky <ja...@basetechnology.com>.
The part I still find confusing is that if you start with 3 and lose 1, your 
have 2, which means you can't always break a tie, right? How is this 
explained? As opposed to saying that 4 is the minimum if you need to 
tolerate a loss of 1.

-- Jack Krupansky

-----Original Message----- 
From: Walter Underwood
Sent: Thursday, December 06, 2012 7:51 PM
To: solr-user@lucene.apache.org
Subject: Re: Minimum HA Setup with SolrCloud

What is the mystery? Two is not more than half of four. Therefore, two 
machines is not a quorum for a four machine Zookeeper ensemble.

wunder

On Dec 6, 2012, at 4:50 PM, Jack Krupansky wrote:

> It's still an unresolved mystery, for now.
>
> -- Jack Krupansky
>
> -----Original Message----- From: Walter Underwood
> Sent: Thursday, December 06, 2012 7:30 PM
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
>
> The Zookeeper ensemble knows the total size. It does not adjust it each 
> time that a machine is partitioned or down.
>
> Two machines is not a quorum for a four machine ensemble.
>
> Why do you think that the documentation would get this wrong?
>
> wunder
>
> On Dec 6, 2012, at 4:14 PM, Jack Krupansky wrote:
>
>> But that is the context I was originally referring to - that with 4 zk 
>> you can lose only one, that you can't lose two. So, if you want to 
>> tolerate a loss on one, 4 zk would be the minimum... but then it was 
>> claimed that you COULD start with 3 zk and loss of one would be fine. I 
>> mean whether you start with 4 and lose 2 or start with 3 and lose 1 is 
>> the same, right?
>>
>> -- Jack Krupansky
>>
>> -----Original Message----- From: Yonik Seeley
>> Sent: Thursday, December 06, 2012 6:34 PM
>> To: solr-user@lucene.apache.org
>> Subject: Re: Minimum HA Setup with SolrCloud
>>
>> On Thu, Dec 6, 2012 at 5:55 PM, Jack Krupansky <ja...@basetechnology.com> 
>> wrote:
>>> I trust that you have the right answer, Mark, but maybe I'm just 
>>> struggling
>>> to parse this statement: "the remaining two machines do not constitute a
>>> majority."
>>>
>>> If you start with 3 zk and lose one, you have an ensemble that does not
>>> "constitute a majority".
>>
>> I think you took that out of context.  They were talking about losing
>> 2 nodes in a 4 node cluster.
>>
>> "For example, with four machines ZooKeeper can only handle the failure
>> of a single machine; if two machines fail, the remaining two machines
>> do not constitute a majority."
>>
>> -Yonik
>> http://lucidworks.com
>
> --
> Walter Underwood
> wunder@wunderwood.org
>
>
>

--
Walter Underwood
wunder@wunderwood.org




Re: Minimum HA Setup with SolrCloud

Posted by Walter Underwood <wu...@wunderwood.org>.
What is the mystery? Two is not more than half of four. Therefore, two machines is not a quorum for a four machine Zookeeper ensemble.

wunder

On Dec 6, 2012, at 4:50 PM, Jack Krupansky wrote:

> It's still an unresolved mystery, for now.
> 
> -- Jack Krupansky
> 
> -----Original Message----- From: Walter Underwood
> Sent: Thursday, December 06, 2012 7:30 PM
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
> 
> The Zookeeper ensemble knows the total size. It does not adjust it each time that a machine is partitioned or down.
> 
> Two machines is not a quorum for a four machine ensemble.
> 
> Why do you think that the documentation would get this wrong?
> 
> wunder
> 
> On Dec 6, 2012, at 4:14 PM, Jack Krupansky wrote:
> 
>> But that is the context I was originally referring to - that with 4 zk you can lose only one, that you can't lose two. So, if you want to tolerate a loss on one, 4 zk would be the minimum... but then it was claimed that you COULD start with 3 zk and loss of one would be fine. I mean whether you start with 4 and lose 2 or start with 3 and lose 1 is the same, right?
>> 
>> -- Jack Krupansky
>> 
>> -----Original Message----- From: Yonik Seeley
>> Sent: Thursday, December 06, 2012 6:34 PM
>> To: solr-user@lucene.apache.org
>> Subject: Re: Minimum HA Setup with SolrCloud
>> 
>> On Thu, Dec 6, 2012 at 5:55 PM, Jack Krupansky <ja...@basetechnology.com> wrote:
>>> I trust that you have the right answer, Mark, but maybe I'm just struggling
>>> to parse this statement: "the remaining two machines do not constitute a
>>> majority."
>>> 
>>> If you start with 3 zk and lose one, you have an ensemble that does not
>>> "constitute a majority".
>> 
>> I think you took that out of context.  They were talking about losing
>> 2 nodes in a 4 node cluster.
>> 
>> "For example, with four machines ZooKeeper can only handle the failure
>> of a single machine; if two machines fail, the remaining two machines
>> do not constitute a majority."
>> 
>> -Yonik
>> http://lucidworks.com
> 
> --
> Walter Underwood
> wunder@wunderwood.org
> 
> 
> 

--
Walter Underwood
wunder@wunderwood.org




Re: Minimum HA Setup with SolrCloud

Posted by Jack Krupansky <ja...@basetechnology.com>.
It's still an unresolved mystery, for now.

-- Jack Krupansky

-----Original Message----- 
From: Walter Underwood
Sent: Thursday, December 06, 2012 7:30 PM
To: solr-user@lucene.apache.org
Subject: Re: Minimum HA Setup with SolrCloud

The Zookeeper ensemble knows the total size. It does not adjust it each time 
that a machine is partitioned or down.

Two machines is not a quorum for a four machine ensemble.

Why do you think that the documentation would get this wrong?

wunder

On Dec 6, 2012, at 4:14 PM, Jack Krupansky wrote:

> But that is the context I was originally referring to - that with 4 zk you 
> can lose only one, that you can't lose two. So, if you want to tolerate a 
> loss on one, 4 zk would be the minimum... but then it was claimed that you 
> COULD start with 3 zk and loss of one would be fine. I mean whether you 
> start with 4 and lose 2 or start with 3 and lose 1 is the same, right?
>
> -- Jack Krupansky
>
> -----Original Message----- From: Yonik Seeley
> Sent: Thursday, December 06, 2012 6:34 PM
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
>
> On Thu, Dec 6, 2012 at 5:55 PM, Jack Krupansky <ja...@basetechnology.com> 
> wrote:
>> I trust that you have the right answer, Mark, but maybe I'm just 
>> struggling
>> to parse this statement: "the remaining two machines do not constitute a
>> majority."
>>
>> If you start with 3 zk and lose one, you have an ensemble that does not
>> "constitute a majority".
>
> I think you took that out of context.  They were talking about losing
> 2 nodes in a 4 node cluster.
>
> "For example, with four machines ZooKeeper can only handle the failure
> of a single machine; if two machines fail, the remaining two machines
> do not constitute a majority."
>
> -Yonik
> http://lucidworks.com

--
Walter Underwood
wunder@wunderwood.org




Re: Minimum HA Setup with SolrCloud

Posted by Walter Underwood <wu...@wunderwood.org>.
The Zookeeper ensemble knows the total size. It does not adjust it each time that a machine is partitioned or down.

Two machines is not a quorum for a four machine ensemble.

Why do you think that the documentation would get this wrong?

wunder

On Dec 6, 2012, at 4:14 PM, Jack Krupansky wrote:

> But that is the context I was originally referring to - that with 4 zk you can lose only one, that you can't lose two. So, if you want to tolerate a loss on one, 4 zk would be the minimum... but then it was claimed that you COULD start with 3 zk and loss of one would be fine. I mean whether you start with 4 and lose 2 or start with 3 and lose 1 is the same, right?
> 
> -- Jack Krupansky
> 
> -----Original Message----- From: Yonik Seeley
> Sent: Thursday, December 06, 2012 6:34 PM
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
> 
> On Thu, Dec 6, 2012 at 5:55 PM, Jack Krupansky <ja...@basetechnology.com> wrote:
>> I trust that you have the right answer, Mark, but maybe I'm just struggling
>> to parse this statement: "the remaining two machines do not constitute a
>> majority."
>> 
>> If you start with 3 zk and lose one, you have an ensemble that does not
>> "constitute a majority".
> 
> I think you took that out of context.  They were talking about losing
> 2 nodes in a 4 node cluster.
> 
> "For example, with four machines ZooKeeper can only handle the failure
> of a single machine; if two machines fail, the remaining two machines
> do not constitute a majority."
> 
> -Yonik
> http://lucidworks.com 

--
Walter Underwood
wunder@wunderwood.org




Re: Minimum HA Setup with SolrCloud

Posted by Jack Krupansky <ja...@basetechnology.com>.
But that is the context I was originally referring to - that with 4 zk you 
can lose only one, that you can't lose two. So, if you want to tolerate a 
loss on one, 4 zk would be the minimum... but then it was claimed that you 
COULD start with 3 zk and loss of one would be fine. I mean whether you 
start with 4 and lose 2 or start with 3 and lose 1 is the same, right?

-- Jack Krupansky

-----Original Message----- 
From: Yonik Seeley
Sent: Thursday, December 06, 2012 6:34 PM
To: solr-user@lucene.apache.org
Subject: Re: Minimum HA Setup with SolrCloud

On Thu, Dec 6, 2012 at 5:55 PM, Jack Krupansky <ja...@basetechnology.com> 
wrote:
> I trust that you have the right answer, Mark, but maybe I'm just 
> struggling
> to parse this statement: "the remaining two machines do not constitute a
> majority."
>
> If you start with 3 zk and lose one, you have an ensemble that does not
> "constitute a majority".

I think you took that out of context.  They were talking about losing
2 nodes in a 4 node cluster.

"For example, with four machines ZooKeeper can only handle the failure
of a single machine; if two machines fail, the remaining two machines
do not constitute a majority."

-Yonik
http://lucidworks.com 


Re: Minimum HA Setup with SolrCloud

Posted by Yonik Seeley <yo...@lucidworks.com>.
On Thu, Dec 6, 2012 at 5:55 PM, Jack Krupansky <ja...@basetechnology.com> wrote:
> I trust that you have the right answer, Mark, but maybe I'm just struggling
> to parse this statement: "the remaining two machines do not constitute a
> majority."
>
> If you start with 3 zk and lose one, you have an ensemble that does not
> "constitute a majority".

I think you took that out of context.  They were talking about losing
2 nodes in a 4 node cluster.

"For example, with four machines ZooKeeper can only handle the failure
of a single machine; if two machines fail, the remaining two machines
do not constitute a majority."

-Yonik
http://lucidworks.com

Re: Minimum HA Setup with SolrCloud

Posted by Jack Krupansky <ja...@basetechnology.com>.
I trust that you have the right answer, Mark, but maybe I'm just struggling 
to parse this statement: "the remaining two machines do not constitute a 
majority."

If you start with 3 zk and lose one, you have an ensemble that does not 
"constitute a majority".

So, the question is what can or can't you do once your ensemble no longer 
has a majority? Why does "majority" matter if you can run just fine? I mean, 
there must be some hard-core downside, other than that you can't lose any 
more nodes.

-- Jack Krupansky

-----Original Message----- 
From: Mark Miller
Sent: Thursday, December 06, 2012 5:49 PM
To: solr-user@lucene.apache.org
Subject: Re: Minimum HA Setup with SolrCloud

3 is the minimum if you want to allow a node to go down.

1 is the minimum if you want the thing to work at all - but if the 1 goes 
down, ZooKeeper may stop working…

- Mark

On Dec 6, 2012, at 2:21 PM, Jack Krupansky <ja...@basetechnology.com> wrote:

> Rewind.
>
> If 1 is the minimum, what is the 3 "minimum" all about?
>
> The zk web page does say "Three ZooKeeper servers is the minimum 
> recommended size for an ensemble, and we also recommend that they run on 
> separate machines" - but it does say "recommended".
>
> But back to the original question - it sounds as if 4 nodes would the 
> recommended minimum number of nodes if you want to tolerate one machine 
> going down and maintaining that 3 zookeeper recommended minimum ensemble.
>
> From the zookeeper web page:
>
> "For reliable ZooKeeper service, you should deploy ZooKeeper in a cluster 
> known as an ensemble. As long as a majority of the ensemble are up, the 
> service will be available. Because Zookeeper requires a majority, it is 
> best to use an odd number of machines. For example, with four machines 
> ZooKeeper can only handle the failure of a single machine; if two machines 
> fail, the remaining two machines do not constitute a majority. However, 
> with five machines ZooKeeper can handle the failure of two machines."
>
> See:
> http://zookeeper.apache.org/doc/r3.1.2/zookeeperAdmin.html
>
> -- Jack Krupansky
>
> -----Original Message----- From: Otis Gospodnetic
> Sent: Thursday, December 06, 2012 4:39 PM
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
>
> 1 is the minimum :)
> 2 makes no sense.
> 3 must be the most common number in the zoo.
>
> Otis
> --
> Performance Monitoring - http://sematext.com/spm/index.html
> Search Analytics - http://sematext.com/search-analytics/index.html
>
>
>
>
> On Thu, Dec 6, 2012 at 9:46 AM, Jack Krupansky 
> <ja...@basetechnology.com>wrote:
>
>> "but if those same machines are running zookeeper, you need 3."
>>
>> And one of those 3 can go down? I thought 3 was the minimum number of
>> zookeepers.
>>
>> -- Jack Krupansky
>>
>> -----Original Message----- From: Mark Miller
>> Sent: Thursday, December 06, 2012 9:30 AM
>> To: solr-user@lucene.apache.org
>> Subject: Re: Minimum HA Setup with SolrCloud
>>
>>
>> It depends on if you are running embedded zk or an external zk ensemble.
>>
>> One leader and a replica is all you need for Solr to allow on machine to
>> go down - but if those same machines are running zookeeper, you need 3.
>>
>> You could also run zookeeper on one external machine and then it would be
>> fine if you lost one solr node - but if you the one external zk node went
>> down you would lose the ability to do updates until it was brought back 
>> up.
>>
>> - Mark
>>
>> On Dec 6, 2012, at 2:46 AM, Thomas Heigl <th...@umschalt.com> wrote:
>>
>> Hey all,
>>>
>>> I'm in the process of migrating a single Solr 4.0 instace to a SolrCloud
>>> setup for availability reasons.
>>>
>>> After studying the wiki page for SolrCloud I'm not sure what the 
>>> absolute
>>> minimum setup is that would allow for one machine to go down.
>>>
>>> Would it be enough to have one shard with one leader and one replica?
>>> Could
>>> this setup tolerate one of the instances going down? Or do I need three
>>> instance because Zookeeper needs a quorum of instances?
>>>
>>> Cheers,
>>>
>>> Thomas
>>>
>>
> 

Re: Minimum HA Setup with SolrCloud

Posted by Mark Miller <ma...@gmail.com>.
3 is the minimum if you want to allow a node to go down.

1 is the minimum if you want the thing to work at all - but if the 1 goes down, ZooKeeper may stop working…

- Mark

On Dec 6, 2012, at 2:21 PM, Jack Krupansky <ja...@basetechnology.com> wrote:

> Rewind.
> 
> If 1 is the minimum, what is the 3 "minimum" all about?
> 
> The zk web page does say "Three ZooKeeper servers is the minimum recommended size for an ensemble, and we also recommend that they run on separate machines" - but it does say "recommended".
> 
> But back to the original question - it sounds as if 4 nodes would the recommended minimum number of nodes if you want to tolerate one machine going down and maintaining that 3 zookeeper recommended minimum ensemble.
> 
> From the zookeeper web page:
> 
> "For reliable ZooKeeper service, you should deploy ZooKeeper in a cluster known as an ensemble. As long as a majority of the ensemble are up, the service will be available. Because Zookeeper requires a majority, it is best to use an odd number of machines. For example, with four machines ZooKeeper can only handle the failure of a single machine; if two machines fail, the remaining two machines do not constitute a majority. However, with five machines ZooKeeper can handle the failure of two machines."
> 
> See:
> http://zookeeper.apache.org/doc/r3.1.2/zookeeperAdmin.html
> 
> -- Jack Krupansky
> 
> -----Original Message----- From: Otis Gospodnetic
> Sent: Thursday, December 06, 2012 4:39 PM
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
> 
> 1 is the minimum :)
> 2 makes no sense.
> 3 must be the most common number in the zoo.
> 
> Otis
> --
> Performance Monitoring - http://sematext.com/spm/index.html
> Search Analytics - http://sematext.com/search-analytics/index.html
> 
> 
> 
> 
> On Thu, Dec 6, 2012 at 9:46 AM, Jack Krupansky <ja...@basetechnology.com>wrote:
> 
>> "but if those same machines are running zookeeper, you need 3."
>> 
>> And one of those 3 can go down? I thought 3 was the minimum number of
>> zookeepers.
>> 
>> -- Jack Krupansky
>> 
>> -----Original Message----- From: Mark Miller
>> Sent: Thursday, December 06, 2012 9:30 AM
>> To: solr-user@lucene.apache.org
>> Subject: Re: Minimum HA Setup with SolrCloud
>> 
>> 
>> It depends on if you are running embedded zk or an external zk ensemble.
>> 
>> One leader and a replica is all you need for Solr to allow on machine to
>> go down - but if those same machines are running zookeeper, you need 3.
>> 
>> You could also run zookeeper on one external machine and then it would be
>> fine if you lost one solr node - but if you the one external zk node went
>> down you would lose the ability to do updates until it was brought back up.
>> 
>> - Mark
>> 
>> On Dec 6, 2012, at 2:46 AM, Thomas Heigl <th...@umschalt.com> wrote:
>> 
>> Hey all,
>>> 
>>> I'm in the process of migrating a single Solr 4.0 instace to a SolrCloud
>>> setup for availability reasons.
>>> 
>>> After studying the wiki page for SolrCloud I'm not sure what the absolute
>>> minimum setup is that would allow for one machine to go down.
>>> 
>>> Would it be enough to have one shard with one leader and one replica?
>>> Could
>>> this setup tolerate one of the instances going down? Or do I need three
>>> instance because Zookeeper needs a quorum of instances?
>>> 
>>> Cheers,
>>> 
>>> Thomas
>>> 
>> 
> 


Re: Minimum HA Setup with SolrCloud

Posted by Jack Krupansky <ja...@basetechnology.com>.
Rewind.

If 1 is the minimum, what is the 3 "minimum" all about?

The zk web page does say "Three ZooKeeper servers is the minimum recommended 
size for an ensemble, and we also recommend that they run on separate 
machines" - but it does say "recommended".

But back to the original question - it sounds as if 4 nodes would the 
recommended minimum number of nodes if you want to tolerate one machine 
going down and maintaining that 3 zookeeper recommended minimum ensemble.

>From the zookeeper web page:

"For reliable ZooKeeper service, you should deploy ZooKeeper in a cluster 
known as an ensemble. As long as a majority of the ensemble are up, the 
service will be available. Because Zookeeper requires a majority, it is best 
to use an odd number of machines. For example, with four machines ZooKeeper 
can only handle the failure of a single machine; if two machines fail, the 
remaining two machines do not constitute a majority. However, with five 
machines ZooKeeper can handle the failure of two machines."

See:
http://zookeeper.apache.org/doc/r3.1.2/zookeeperAdmin.html

-- Jack Krupansky

-----Original Message----- 
From: Otis Gospodnetic
Sent: Thursday, December 06, 2012 4:39 PM
To: solr-user@lucene.apache.org
Subject: Re: Minimum HA Setup with SolrCloud

1 is the minimum :)
2 makes no sense.
3 must be the most common number in the zoo.

Otis
--
Performance Monitoring - http://sematext.com/spm/index.html
Search Analytics - http://sematext.com/search-analytics/index.html




On Thu, Dec 6, 2012 at 9:46 AM, Jack Krupansky 
<ja...@basetechnology.com>wrote:

> "but if those same machines are running zookeeper, you need 3."
>
> And one of those 3 can go down? I thought 3 was the minimum number of
> zookeepers.
>
> -- Jack Krupansky
>
> -----Original Message----- From: Mark Miller
> Sent: Thursday, December 06, 2012 9:30 AM
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
>
>
> It depends on if you are running embedded zk or an external zk ensemble.
>
> One leader and a replica is all you need for Solr to allow on machine to
> go down - but if those same machines are running zookeeper, you need 3.
>
> You could also run zookeeper on one external machine and then it would be
> fine if you lost one solr node - but if you the one external zk node went
> down you would lose the ability to do updates until it was brought back 
> up.
>
> - Mark
>
> On Dec 6, 2012, at 2:46 AM, Thomas Heigl <th...@umschalt.com> wrote:
>
>  Hey all,
>>
>> I'm in the process of migrating a single Solr 4.0 instace to a SolrCloud
>> setup for availability reasons.
>>
>> After studying the wiki page for SolrCloud I'm not sure what the absolute
>> minimum setup is that would allow for one machine to go down.
>>
>> Would it be enough to have one shard with one leader and one replica?
>> Could
>> this setup tolerate one of the instances going down? Or do I need three
>> instance because Zookeeper needs a quorum of instances?
>>
>> Cheers,
>>
>> Thomas
>>
>
> 


Re: Minimum HA Setup with SolrCloud

Posted by Michael Della Bitta <mi...@appinions.com>.
One is the loneliest number that you'll ever do,
Two can be as bad as one, it's the loneliest number since the single Zoo.


Michael Della Bitta

------------------------------------------------
Appinions
18 East 41st Street, 2nd Floor
New York, NY 10017-6271

www.appinions.com

Where Influence Isn’t a Game


On Thu, Dec 6, 2012 at 4:39 PM, Otis Gospodnetic <otis.gospodnetic@gmail.com
> wrote:

> 1 is the minimum :)
> 2 makes no sense.
> 3 must be the most common number in the zoo.
>
> Otis
> --
> Performance Monitoring - http://sematext.com/spm/index.html
> Search Analytics - http://sematext.com/search-analytics/index.html
>
>
>
>
> On Thu, Dec 6, 2012 at 9:46 AM, Jack Krupansky <jack@basetechnology.com
> >wrote:
>
> > "but if those same machines are running zookeeper, you need 3."
> >
> > And one of those 3 can go down? I thought 3 was the minimum number of
> > zookeepers.
> >
> > -- Jack Krupansky
> >
> > -----Original Message----- From: Mark Miller
> > Sent: Thursday, December 06, 2012 9:30 AM
> > To: solr-user@lucene.apache.org
> > Subject: Re: Minimum HA Setup with SolrCloud
> >
> >
> > It depends on if you are running embedded zk or an external zk ensemble.
> >
> > One leader and a replica is all you need for Solr to allow on machine to
> > go down - but if those same machines are running zookeeper, you need 3.
> >
> > You could also run zookeeper on one external machine and then it would be
> > fine if you lost one solr node - but if you the one external zk node went
> > down you would lose the ability to do updates until it was brought back
> up.
> >
> > - Mark
> >
> > On Dec 6, 2012, at 2:46 AM, Thomas Heigl <th...@umschalt.com> wrote:
> >
> >  Hey all,
> >>
> >> I'm in the process of migrating a single Solr 4.0 instace to a SolrCloud
> >> setup for availability reasons.
> >>
> >> After studying the wiki page for SolrCloud I'm not sure what the
> absolute
> >> minimum setup is that would allow for one machine to go down.
> >>
> >> Would it be enough to have one shard with one leader and one replica?
> >> Could
> >> this setup tolerate one of the instances going down? Or do I need three
> >> instance because Zookeeper needs a quorum of instances?
> >>
> >> Cheers,
> >>
> >> Thomas
> >>
> >
> >
>

Re: Minimum HA Setup with SolrCloud

Posted by Otis Gospodnetic <ot...@gmail.com>.
1 is the minimum :)
2 makes no sense.
3 must be the most common number in the zoo.

Otis
--
Performance Monitoring - http://sematext.com/spm/index.html
Search Analytics - http://sematext.com/search-analytics/index.html




On Thu, Dec 6, 2012 at 9:46 AM, Jack Krupansky <ja...@basetechnology.com>wrote:

> "but if those same machines are running zookeeper, you need 3."
>
> And one of those 3 can go down? I thought 3 was the minimum number of
> zookeepers.
>
> -- Jack Krupansky
>
> -----Original Message----- From: Mark Miller
> Sent: Thursday, December 06, 2012 9:30 AM
> To: solr-user@lucene.apache.org
> Subject: Re: Minimum HA Setup with SolrCloud
>
>
> It depends on if you are running embedded zk or an external zk ensemble.
>
> One leader and a replica is all you need for Solr to allow on machine to
> go down - but if those same machines are running zookeeper, you need 3.
>
> You could also run zookeeper on one external machine and then it would be
> fine if you lost one solr node - but if you the one external zk node went
> down you would lose the ability to do updates until it was brought back up.
>
> - Mark
>
> On Dec 6, 2012, at 2:46 AM, Thomas Heigl <th...@umschalt.com> wrote:
>
>  Hey all,
>>
>> I'm in the process of migrating a single Solr 4.0 instace to a SolrCloud
>> setup for availability reasons.
>>
>> After studying the wiki page for SolrCloud I'm not sure what the absolute
>> minimum setup is that would allow for one machine to go down.
>>
>> Would it be enough to have one shard with one leader and one replica?
>> Could
>> this setup tolerate one of the instances going down? Or do I need three
>> instance because Zookeeper needs a quorum of instances?
>>
>> Cheers,
>>
>> Thomas
>>
>
>

Re: Minimum HA Setup with SolrCloud

Posted by Jack Krupansky <ja...@basetechnology.com>.
"but if those same machines are running zookeeper, you need 3."

And one of those 3 can go down? I thought 3 was the minimum number of 
zookeepers.

-- Jack Krupansky

-----Original Message----- 
From: Mark Miller
Sent: Thursday, December 06, 2012 9:30 AM
To: solr-user@lucene.apache.org
Subject: Re: Minimum HA Setup with SolrCloud

It depends on if you are running embedded zk or an external zk ensemble.

One leader and a replica is all you need for Solr to allow on machine to go 
down - but if those same machines are running zookeeper, you need 3.

You could also run zookeeper on one external machine and then it would be 
fine if you lost one solr node - but if you the one external zk node went 
down you would lose the ability to do updates until it was brought back up.

- Mark

On Dec 6, 2012, at 2:46 AM, Thomas Heigl <th...@umschalt.com> wrote:

> Hey all,
>
> I'm in the process of migrating a single Solr 4.0 instace to a SolrCloud
> setup for availability reasons.
>
> After studying the wiki page for SolrCloud I'm not sure what the absolute
> minimum setup is that would allow for one machine to go down.
>
> Would it be enough to have one shard with one leader and one replica? 
> Could
> this setup tolerate one of the instances going down? Or do I need three
> instance because Zookeeper needs a quorum of instances?
>
> Cheers,
>
> Thomas 


Re: Minimum HA Setup with SolrCloud

Posted by Mark Miller <ma...@gmail.com>.
It depends on if you are running embedded zk or an external zk ensemble.

One leader and a replica is all you need for Solr to allow on machine to go down - but if those same machines are running zookeeper, you need 3.

You could also run zookeeper on one external machine and then it would be fine if you lost one solr node - but if you the one external zk node went down you would lose the ability to do updates until it was brought back up.

- Mark

On Dec 6, 2012, at 2:46 AM, Thomas Heigl <th...@umschalt.com> wrote:

> Hey all,
> 
> I'm in the process of migrating a single Solr 4.0 instace to a SolrCloud
> setup for availability reasons.
> 
> After studying the wiki page for SolrCloud I'm not sure what the absolute
> minimum setup is that would allow for one machine to go down.
> 
> Would it be enough to have one shard with one leader and one replica? Could
> this setup tolerate one of the instances going down? Or do I need three
> instance because Zookeeper needs a quorum of instances?
> 
> Cheers,
> 
> Thomas