Also, if you are doing 100 million per day, then with a bit of peak to
valley ratio, you are doing 5000 per second. That is pushing limits. If
your ZK has other loads or is on substandard hardware somehow.
If you are storing things in a database, don't you have an create or fail
operation (AKA
ks,
>
> Jeelani
>
> On 1/22/20, 12:57 PM, "Ted Dunning" wrote:
>
> Mohamed,
>
> Could you please answer the question about whether there will be an NDA
> required at the door?
>
>
>
> On Wed, Jan 22, 2020 at 12:51 PM Mohamed Jeelan
Mohamed,
Could you please answer the question about whether there will be an NDA
required at the door?
On Wed, Jan 22, 2020 at 12:51 PM Mohamed Jeelani wrote:
> ZooKeeper Meetup @ Facebook is just 1 week away! (January 29th 2020) -
> Please RSVP if you haven’t yet at -
WHat I suggested is almost exactly what Jordan suggested. I should have
read the rest of the thread before posting.
On Sat, Sep 21, 2019 at 9:54 AM Ted Dunning wrote:
>
> I would suggest that using an epoch number stored in ZK might be helpful.
> Every operation that the master ta
I would suggest that using an epoch number stored in ZK might be helpful.
Every operation that the master takes could be made conditional on the
epoch number using a multi-transaction.
Unfortunately, as you say, you have to have the update of the epoch be
atomic with becoming leader.
The natural
a fair lock
implementation, you are have better completion guarantees.
Does this do what you want?
On Wed, Aug 14, 2019 at 11:04 PM Zili Chen wrote:
> Thanks for your explanation Michael and Ted :-)
>
> Best,
> tison.
>
>
> Ted Dunning 于2019年8月15日周四 下午1:22写道：
>
>
nd where is the "hard"
> comes from?
>
> Best,
> tison.
>
>
> Ted Dunning 于2019年8月15日周四 上午9:40写道：
>
> > The multi op is atomic (all other operations will be before or after teh
> > multi), consistent (all viewers will see all the effects or none, and
The multi op is atomic (all other operations will be before or after teh
multi), consistent (all viewers will see all the effects or none, and
durable (because ZK is linearized anyway).
That leaves isolated which is kind of hard to talk about with ZK since all
operations are fast and sequential.
The multi operation *is* like a transaction. All the operations will
succeed or none.
On Wed, Aug 14, 2019 at 9:33 AM Andor Molnar wrote:
> "it's said that ZooKeeper has a transaction mechanism”
>
> I’m still confused with this. ZooKeeper doesn’t support transactions to my
> best knowledge.
Check here for theory
https://blog.acolyer.org/2015/03/10/zookeepers-atomic-broadcast-protocol-theory-and-practice/
And here
https://www.semanticscholar.org/paper/Zab%3A-High-performance-broadcast-for-primary-backup-Junqueira-Reed/b02c6b00bd5dbdbd951fddb00b906c82fa80f0b3
On Sun, Jul 15, 2018,
The most common error that causes this is to not set up the myid files
correctly.
On Thu, Jun 12, 2014 at 4:47 PM, Cameron McKenzie mckenzie@gmail.com
wrote:
This is not correct, 3 is a minimum for redundancy. If 1 goes down, the
other 2 can still form a quorum (as there are more than
mardi 10 juin 2014, Ted Dunning ted.dunn...@gmail.com a écrit :
On Tue, Jun 10, 2014 at 9:24 AM, Olivier Mallassi
olivier.malla...@gmail.com javascript:; wrote:
What are the guarantees zookeeper provide and a data grid does not
provide?
the lock? a sequential consistency?
You said
Also, if you have to create several state flags in ZK (perhaps you have to
choose several IP addresses, for instance) you can use the multi update
capability.
It is almost never necessary to use an actual lock recipe with ZK. Almost
always what you really want is some kind of first come, first
It is a problem if you expect subsequent watches to go out in milliseconds.
It isn't a problem if the resulting delays are OK with you. To me, it
sounds like it will be just fine. If the herd effect is too much, you can
always split the version flags into many pieces and update one version flag
On Sun, May 18, 2014 at 10:38 PM, Pramod Biligiri
pramodbilig...@gmail.comwrote:
Do you see any other problems with the approach I'm taking. If we see any
gains from it, we can look at the tricky issues next.
No. The idea of partitioning ZK has come up often and something roughly
like what
Pramod,
Have you looked at the multi command? That might cause you some serious
heartburn.
On Sun, May 18, 2014 at 8:25 PM, Pramod Biligiri
pramodbilig...@gmail.comwrote:
Hi,
[Let me know if you want this thread moved to the Dev list (or even to
JIRA). I was only seeing automated mails
On Sun, May 18, 2014 at 9:36 PM, Pramod Biligiri
pramodbilig...@gmail.comwrote:
I guess you mean that you can't parallellize the workload because a multi
command might require locking all the containers? Let me know if I'm
missing something.
Right. Getting that to work cleanly could be
Your email is a little ambiguous.
a) 5 instances with 3 as quorum could mean 5 instances configured and
running normally.
Or
b) it could mean 5 instances with 2 instances that are down.
In (a) restarting the leader instance *should* cause the cluster to do a
leader election again and form a
, but it should be hard to detect any outage.
Thank you
Best, Jae
On Wed, Apr 9, 2014 at 12:06 PM, Ted Dunning ted.dunn...@gmail.com
wrote:
Your email is a little ambiguous.
a) 5 instances with 3 as quorum could mean 5 instances configured and
running normally.
Or
b) it could
:
Interesting. Especially since at work we've been leaning towards using
bcache for performance reasons to be able to deal with the flood of input
we get - not so much for ZooKeeper but for Cassandra. Do you have any
opinions about bcache?
http://bcache.evilpiepirate.org/
On 21 March 2014 07:14, Ted
Test first without anything all that fancy gear. For all of the
applications that you mention except Kafka, the actual transaction rate is
near nil. A dedicated single spindle per ZK is likely to be be luxurious
accommodation. I doubt seriously that you need SSD's.
On Thu, Mar 20, 2014 at
SSD's also have the issue that it is common that recently written data is
not actually persisted. Worse, new data might be persisted while slightly
older data is not. These issues differ greatly across different hardware.
Disks with write caching disabled are vastly better understood.
On
The best way to mitigate this is to fix your platform problem. There is
very little that ZK can do to help.
IF it takes 10 seconds to flush to disk, this is going to make these
machines pretty intolerable for all kinds of applications.
To specifically answer your question, yes, you can increase
If you are going to try to put a web-ish interface on ZK, look at web
sockets so that you can get notifications.
On Wed, Jan 8, 2014 at 11:46 AM, Jordan Zimmerman
jor...@jordanzimmerman.com wrote:
My focus is langs that have mediocre or no ZK support (i.e. Ruby). I’m
thinking of adding a
@gmail.comwrote:
Excuse my ignorance (I'm relatively new to ZK), but how does the accuracy
of the clock affect this situation?
On Wed, Nov 27, 2013 at 11:53 AM, Ted Dunning ted.dunn...@gmail.com
wrote:
This is not necessarily true. The old master may not have an accurate
clock
This is not necessarily true. The old master may not have an accurate
clock.
The ascending id idea that Alex mentions is a very nice way to put more
guarantees on the process.
On Tue, Nov 26, 2013 at 2:58 PM, Alexander Shraer shra...@gmail.com wrote:
Cameron's solution basically relies on
For production settings with well-behaved applications, sharing is not a
bad idea. I would definitely isolate development efforts onto a dev
instance of ZK. And if you have trigger happy admins who think rebooting
fixes all ills, I would consider separating apps.
How is it that you wound up
those
types of failures is considered significant for some people. Are there
technical reasons that would prevent this idea from working?
On 09/10/2013 01:31 PM, Ted Dunning wrote:
I don't see the strong value here. A few failures would be detected more
quickly, but I am not convinced
Generally, ZK is much better as a coordination layer.
Starting with an expected transaction load well above the normal limits of
operation is not a grand idea.
Much better to do something simpler like have ZK coordinate shard masters
that each use conventional methods for handling transactions
+1 on moving to JDK 7
I have been through this decision a few times now. The first time I was -1
(eventually -0), the second time +0 and lately I am +1 on this change.
ZK probably can't move to require JDK7, but 7 really is better.
On Tue, Jul 9, 2013 at 9:09 AM, Patrick Hunt
, a lot of big application
have that). Within one cluster, it can done by enable/disable quorum.
On Mon, Jul 1, 2013 at 8:37 PM, Ted Dunning ted.dunn...@gmail.com wrote:
Embedding zk in a product is a fine thing. Embedding zk in an
application
is (in my estimation) a bad idea. The problem
Embedding zk in a product is a fine thing. Embedding zk in an application is
(in my estimation) a bad idea. The problem is that you generally want to
coordinate all the way down to zero, especially in upgrade situations.
Combining the uptime constraints of zk with those of your app is bad
This sounds like you have a bad configuration for your cluster.
On Tue, Jun 11, 2013 at 2:48 PM, NoamBC n...@cloudon.com wrote:
Hi,
We use ZK 3.3 in production, a quorum of 5 nodes, and found an odd
behaviour
where if we take one node offline, the quorum keeps serving requests, but
if
That is always true due to pesky things like propagation delay. The idea
of before and after are very slippery unless you have and use primitives
like sync to force boundaries.
On Fri, Mar 1, 2013 at 1:38 PM, Yasin yasinceli...@gmail.com wrote:
So, if the read request is made by some other
Yes. Sync doesn't guarantee up to date. It guarantees an ordering. It
guarantees that if event A involves a ZK update and if you can guarantee
that A occurs before sync, then any read on a client C that is done after a
sync on C will see a successor state of A.
On Fri, Mar 1, 2013 at 2:01 PM,
I think that Camille's points are well founded.
However, I have just had two years experience with the opposite approach in
which we use ZK as a way of doing application level load balancing and I
really have come to like it.
The situation is that the job tracker in MapR is a Hadoop Job Tracker
Yes.
And in general, you can't have precise distributed lock control. There
will always be a bit of slop.
So decide which penalty is easier to pay. Do you want at-most-one or
at-least-one or something in between? You can't have exactly-one and
still deal with expected problems like partition
Wait-free means that from the user's point of view, all operations will
proceed or fail immediately. Synchronization internal to the server to
ensure coherent memory access doesn't really apply here.
The point is that ZK is not a lock manager and there is no corollary to a
wait operation. This
On Fri, Dec 7, 2012 at 6:50 PM, Ian Kallen spidaman.l...@gmail.com wrote:
No swap, we keep swap turned off following the better to OOM than
slow down theory.
Good theory except that it may compromise getting cores.
I'm gonna try to get GC's plotted for all of these
JVMs. I'm also finding
Next test would be to incorporate a small patch to seed the shuffling of
server names by the client ID or something.
On Thu, Dec 6, 2012 at 12:32 PM, Brian Tarbox briantar...@gmail.com wrote:
I killed all the clients and then restarted them do its not a reconnection
issue .
Is there any swap activity on the client or server boxes?
On Thu, Dec 6, 2012 at 8:25 PM, Ian Kallen spidaman.l...@gmail.com wrote:
d) ZK swapping out due to inactivity during memory pressure?
Can you cite an explanation or explain this here? I'm not sure what to
look for. It wouldn't be
Kishore,
That should be a good explanation, but it depends on where the returning
node gets put into the replication chain.
Most replicating systems get put at the end of a replication chain since
that causes the least disruption.
I don't know what ZK does, but this can be tested by determining
Shuffle depends on Math.random which is seeded by time of start. That
should be just fine.
On Wed, Dec 5, 2012 at 11:35 PM, Camille Fournier cami...@apache.orgwrote:
If you're using the Java ZooKeeper client, you can see in the code that the
way connections are established is that we parse
THis looks like very low load.
What is the rate of change on znodes (i.e. what is the desired watch signal
rate)?
On Wed, Dec 5, 2012 at 10:10 PM, Ian Kallen spidaman.l...@gmail.com wrote:
We have an ensemble of three servers and have observed varying
latencies, watches that seemingly don't
the lock any more.
On Nov 25, 2012, at 11:40 AM, Ted Dunning ted.dunn...@gmail.com wrote:
To expand on this, when a client is disconnected, it receives a
disconnect
event.
From that time until the client is reconnected, the client doesn't know
what has happened. The session may have
Why do you say leader and follower? Normally all members of a quorum can
take on any role.
On Tue, Nov 20, 2012 at 11:38 PM, Maru Panda song...@gmail.com wrote:
Hi guys
we have 1 zookeeper cluster
On server A, there're 2 zookeepers, port: 2180 ( follower ) and 2181 (
leader )
another one
In particular, you can do something like this
contents, version = get(x/y/z)
multi([delete(x/y/z, version), create(a/b/c, contents)])
The multi will fail if x/y/z has been updated between the get and the
multi. The multi will also fail
if a/b/c cannot be created. If the multi succeeds,
You specify the MINIMUM number of copies when you define the number of
nodes in your ZK cluster.
The idea is that ZK requires strong consistency and provides guarantees to
that effect. The only way to provide those guarantees is if a majority of
the ZK cluster agree to and persist all changes.
form server. So I think that need to takes a few
milliseconds to receive session expried events from full gc finish, that's
right?
2012/10/29 Ted Dunning ted.dunn...@gmail.com
Yes. When the GC finishes, it will get the disconnect and session
expiration event.
You can experiment
You don't have Cloudera Zookeeper. You may have Apache Zookeeper that is
packaged with Cloudera's software distribution.
On Fri, Oct 26, 2012 at 8:31 AM, Roshan Punnoose rpunno...@texeltek.comwrote:
I have my cloud era zookeeper (3.3.5) running with the dataLogDir and the
dataDir going to two
Using 6 nodes for ZK is a bit odd. Actually, it is a bit even.
If all of the nodes are involved in the quorum, you will get lower write
throughput than with 5 nodes and slightly higher chance of failure since it
is more likely to get 3/6 node failures versus 3/5 failures.
What motivated your
Ahh. Better.
On Wed, Oct 17, 2012 at 6:29 PM, yang.li yang...@baifendian.com wrote:
Thank you for your advicse, Ted.
Actually, the sixth node is set to observer mode,
we just put it there to stand by.
Can you provide sample code and more detailed replication instructions?
On Fri, Oct 12, 2012 at 6:06 PM, Deepak Jagtap deepak.jag...@maxta.comwrote:
Sure, just reported this on jira!
Thanks Regards,
Deepak
On Fri, Oct 12, 2012 at 1:01 PM, Michi Mutsuzaki mi...@cs.stanford.edu
wrote:
Two disks could be a significant advantage.
You should also experiment with ways to avoid VM induced gaps in time.
Finally, if you really are going to be write heavy, 3 nodes are likely to
perform better than 5.
On Tue, Oct 9, 2012 at 2:03 PM, Brian Tarbox briantar...@gmail.com wrote:
In
I have some experience with these.
On Tue, Oct 9, 2012 at 4:47 PM, Eric Fleischman e...@ericfleischman.comwrote:
* Using high IOPS instances
Haven't tried this.
* Using multiple disks and doing OS level RAID across them to present a
single logical disk to the app
This typically is very
Yes.
And Patrick's experience is not unexpected. There is, however, a huge
variation with different types of flash memory. The software driving the
flash can also result in very different experience. The experiences that
he alludes to are likely with a conventional SSD packaging of flash
AFAIK) to
engineer out latency spikes. I'd imagine they started with a strong
vendor and not a low end device, but of course this is just
speculation.
On Thu, Oct 4, 2012 at 12:28 PM, Ted Dunning ted.dunn...@gmail.com
wrote:
Yes.
And Patrick's experience is not unexpected
Another option is to use a proxy.
On Mon, Sep 17, 2012 at 6:44 PM, Morris Bernstein
mor...@systems-deployment.com wrote:
The obvious solution would be running multiple Zookeepers hierarchically.
Each server would take, say, 1000 clients. Each server would be a client
of an uber Zookeeper.
Well... you can set one of the 2 as an observer. That way you can have a
quorum of one of 2, but you can't have a quorum of the other of two.
Probably not what you want.
You can also set up a third member who refuses to become leader. They
basically serve as a tiebreaker. You can lighten the
client or something else? (memory
corruption on client possible?)
Patrick
On Fri, Aug 31, 2012 at 12:56 PM, Ted Dunning ted.dunn...@gmail.com
wrote:
But isn't the larger ZXID pretty stunningly large?
The epoch number is nearly 2 billion and the transaction id is 845
million
But isn't the larger ZXID pretty stunningly large?
The epoch number is nearly 2 billion and the transaction id is 845 million.
These seem implausible from a starting point of 0.
On Fri, Aug 31, 2012 at 12:54 PM, Patrick Hunt ph...@apache.org wrote:
seen zxid 0x636c65616e2d7374 our last zxid
That link seems to point to the what is exhibitor page.
Can you explain how this feature is not fantastically dangerous? Do you have
long time constants? How do you avoid obvious crazy cases like 5 new hosts
jumping onto a cluster where all of the original hosts in the quorum are down
for
Yes. Updating and restarting works fine.
One does wonder why you have *four* servers now. That generally doesn't
provide any advantage over having three servers.
On Mon, Jun 18, 2012 at 10:28 AM, David Nickerson
davidnickerson4mailingli...@gmail.com wrote:
Say I have a quorum of four
, 2012 at 2:35 PM, Jordan Zimmerman
jzimmer...@netflix.comwrote:
FYI - have a look at our Exhibitor system which makes upgrades/additions
easier to manage.
https://github.com/Netflix/exhibitor
On 6/18/12 11:26 AM, Ted Dunning ted.dunn...@gmail.com wrote:
Yes. Updating and restarting
?]
Another way to do what I would like would be to create a znode structure
that is not attached to the existing one at all, then once I have it the
way it should be, attach it to an existing node. However, I am quite
certain this is not possible
On 5/16/2012 7:21 PM, Ted Dunning wrote
It looks like you are assuming that the versions of two znodes are
synchronized. That seems pretty dangerous.
What is the higher level intent here? Would it better to simply build a multi
to update both znodes?
Sent from my iPhone
On May 16, 2012, at 11:59 AM, Joe Gamache
.
(com.netflix.curator.framework.recipes.queue.DistributedQueue) But feeling
it is a
problem of ZooKeeper.
Thank you.
Yuhan
On Mon, Apr 30, 2012 at 10:50 PM, Ted Dunning ted.dunn...@gmail.com
wrote:
How many objects are in that directory?
Sent from my iPhone
On Apr 30, 2012, at 8:24 PM, Yuhan Zhang yzh
The client can't think it has succeeded with a deletion if it is connected
to the minority side of a partitioned cluster. To think that, the commit
would have to be be ack'ed by a majority which by definition can't happen
either because the master is in the minority and can't get a majority or
As long as the old quorum constitutes a quorum in the new cluster you should be
fine.
In the replace a dead node scenario you can reconfigure the survivors before
bringing up the new node. That only works if you start with 5 nodes, but being
down a node in a three node cluster is a problem
Alexander's comment still applies.
VM's can function or go away completely, but they can also malfunction
in more subtle ways such that they just go VY slowly. You
have to account for that failure mode. These failures can even be
transient.
This would probably break your approach.
On
point to a new IP (and
hence cause the old server to be replaced).
Did I understand your objection correctly?
Von: ext Ted Dunning [ted.dunn...@gmail.com]
Gesendet: Donnerstag, 15. März 2012 19:50
Bis: user@zookeeper.apache.org
Cc: shra...@gmail.com
-inc.com wrote:
I’ve been wondering about this for a while, and suspect that this check
doesn’t exist in the code… but I may be wrong.
From: Ted Dunning [mailto:ted.dunn...@gmail.com]
Sent: Wednesday, March 07, 2012 4:55 PM
To: Alexander Shraer
Cc: user@zookeeper.apache.org
Subject: Re
** **
*From:* Ted Dunning [mailto:ted.dunn...@gmail.com]
*Sent:* Thursday, March 08, 2012 12:32 AM
*To:* Alexander Shraer
*Cc:* user@zookeeper.apache.org
*Subject:* Re: Possibility / consequences of having multiple elected
leaders
** **
The whole point of the zab protocol is to ensure
Can you resolve the pronouns here?
It looks like ZK is running, but that Finagle will not.
Is that what you meant to say?
The log messages make it look like somebody assumes that a directory
/twitter/servers does exist, but is finding that this directory doesn't
exist. Isn't that an
://screencast.com/t/TWG529FGV0R
Finagle added the paths to ZK, now when I try to restart, it looks like ZK
is trying to replay some transaction, failing, and quitting. If I knew
where the data file was, I could just delete the data file and start fresh.
-Ryan
On 3/8/12 4:03 PM, Ted Dunning
It won't be any different than a temporary state when one of 3 or 5 nodes
is down.
On Thu, Mar 8, 2012 at 4:10 PM, Jordan Zimmerman jzimmer...@netflix.comwrote:
AlsoŠ
I thought that ZK ensembles need to be odd in number. How would ZK handle
a temporary state where there is an even number?
Any even number that is greater than half of the configured number of nodes
is fine.
The only *really* bad even number of servers is 0.
On Thu, Mar 8, 2012 at 4:24 PM, Ted Dunning ted.dunn...@gmail.com wrote:
It won't be any different than a temporary state when one of 3 or 5 nodes
is down
This can be emulated on Linux by simply pausing the process.
The correct behavior is that the old leader will freeze and if it comes
back relatively soon, it will still be recognized as leader.
If the pause is long enough, then the other members of the quorum will
decide that they have lost
Not off the cuff and I have to run away right now.
On Wed, Mar 7, 2012 at 4:07 PM, Alexander Shraer shra...@yahoo-inc.comwrote:
Such a commit will be rejected due to an old epoch.
Ted, can you please point me to the place in the code where this check is
performed ?
Thanks a lot,
Alex
If you were able to connect to a 1 out of 3 servers, then you have a very
serious problem since that isn't enough machines to form a quorum.
I suspect you have a configuration error.
On Tue, Mar 6, 2012 at 9:08 AM, Patrick Hunt ph...@apache.org wrote:
This is windows I take it? What version of
OK. Trying again against the correct thread.
I trimmed down your summary to something that jumped out at me. If you
have three ZK servers in a cluster, then taking 2 down should not allow
normal operation.
I think you have a config error.
On Tue, Mar 6, 2012 at 8:15 AM, Scott Lindner
Look at the Curator library.
On Tue, Mar 6, 2012 at 9:50 AM, Shelley, Ryan ryan.shel...@disney.comwrote:
Ok, I'll go back and see if I can find the Leader Election examples and
rethink the problem a bit more.
Thanks for the valuable insights. I truly appreciate it.
-Ryan
There are
Yes ^ 4
On Mon, Mar 5, 2012 at 3:59 PM, Shelley, Ryan ryan.shel...@disney.comwrote:
If I'm correct after reading the docs, if my client sets a watch on a
znode, disconnects, then automatically reconnects, any appropriate pending
watches will be fired. I'm assuming this is only as long as the
On Mon, Mar 5, 2012 at 3:59 PM, Shelley, Ryan ryan.shel...@disney.comwrote:
Additionally, I'm curious if there is a way to know if there are any
watches currently on a znode. Since we can't do ephemeral parent nodes,
yet, and if I have a client that fails (which results in losing their
Ryan,
Lots of assumptions here.
On Mon, Mar 5, 2012 at 5:02 PM, Shelley, Ryan ryan.shel...@disney.comwrote:
The reason I was thinking that it might be useful to know if there are any
watches on a node is for the lack of ephemeral parent nodes. If a node
doesn't have a watch on it, I can
I don't think that your test tests what you think it tests.
You won't necessarily get a notification every time a value changes. If a
value changes and a watch is queued, you won't get another notification
until you get the data and set the new watch. If another change happens
before you reset
On Fri, Mar 2, 2012 at 3:33 PM, Amirhossein Kiani amirhki...@gmail.comwrote:
Oh.. you are right. The value was skipped because I was getting two of
some values (newer but not older)
So I should restructure my code to not make decisions that rely on locks
created based on node data (unless I
.
On 2/29/12 7:16 PM, Ted Dunning ted.dunn...@gmail.com wrote:
On Wed, Feb 29, 2012 at 7:04 PM, Marshall McMullen
marshall.mcmul...@gmail.com wrote:
Yes, Ted's right. The multi has to fail as that's part of the contract
it
guarantees.
The only thing you could do, which will significantly
(which, even if it did work, the
CheckResult object doesn't include the path, so I'm just hoping the
results are in the same order as the operations were listed in, as I have
to infer the path I'll need to build from the position of the OpResult in
the list).
On 3/1/12 3:05 PM, Ted Dunning
Well, in 3.4, you can use multi to do this.
On Wed, Feb 29, 2012 at 4:08 PM, Shelley, Ryan ryan.shel...@disney.comwrote:
Is it possible to create all nodes in a path in one step?
For example, my ZK is empty of any nodes. I want to create:
/lorem/ipsum/foo/bar
Do I have to create each
at 4:26 PM, Shelley, Ryan ryan.shel...@disney.comwrote:
Can you reference some docs? I looked but couldn't find anything regarding
multi.
On 2/29/12 4:23 PM, Ted Dunning ted.dunn...@gmail.com wrote:
Well, in 3.4, you can use multi to do this.
On Wed, Feb 29, 2012 at 4:08 PM, Shelley, Ryan
On Wed, Feb 29, 2012 at 7:04 PM, Marshall McMullen
marshall.mcmul...@gmail.com wrote:
Yes, Ted's right. The multi has to fail as that's part of the contract it
guarantees.
The only thing you could do, which will significantly narrow the race
condition, is as you're *building *the multi,
ZK doesn't do this out of the box, but it is very easy to wrap up ZK
primitives to get this behavior.
On Wed, Feb 22, 2012 at 12:45 AM, Shelley, Ryan ryan.shel...@disney.comwrote:
Hi folks. Very new to ZooKeeper and evaluating it for a project. The ZK
directory-like layout is perfect for what
These results are about what is expected although the might be a little more
extreme.
I doubt very much that hbase is mutating zk nodes fast enough for this to
matter much.
Sent from my iPhone
On Feb 14, 2012, at 8:00, N Keywal nkey...@gmail.com wrote:
Hi,
I've done a test with
status written in . On paper, if we need 0,02s per node, that
makes it to the minute to recover, just for zookeeper.
That's theory. I haven't done a precise measurement yet.
Anyway, if ZooKeeper can be faster, it's always very interesting :-)
Cheers,
N.
On Tue, Feb 14, 2012 at 8:00 PM, Ted
Yes it is possible.
With a loaded server, each group of transactions wiill take about one
rotation. But the time from when they arrived to the time that they are
committed will be roughly 0 ... 8 ms for a 7200 RPM drive because the
transactions will be arriving at different times.
There will be
On Fri, Feb 3, 2012 at 4:01 PM, Jason Harmon jh0...@att.com wrote:
My preference, of course, would be to have three datacenters...if one is
partitioned off, zookeeper would not respond, which would be perfect. In
that scenario, our other two would still be up and running in the other two
One of the required functions of Zookeeper is that network partition should
not result in inconsistent results. If you have 2 servers, then partition
leaves a symmetrical situation that you have to break in order to have a
reasonable way to continue.
Since you can't, in general, distinguish
I am completely clueless about what you are asking here.
Can you be a bit more explicit?
- do you mean that you want to increase the number of threads in the
listeners?
- do you mean that you want to increase the number of listeners?
- what do you mean by load? Load on Zookeeper? Load on
A backup can still be useful. It is a common property that a database
backup is known to be slightly out of date.
Such a backup can still be very useful. In many systems, the most common
cause of error is simple human intervention. This especially applies to
file systems and databases, but can
1 - 100 of 258 matches
Mail list logo
