Hi,
> Note: the counter used to store the next sequence number is a signed int
> (4bytes) maintained by the parent node, the counter will overflow when
> incremented beyond 2147483647 (resulting in a name "-2147483648").
I wrote a test[1] and found that overflow of cversion results in
NODEEXIST
I wanted to add 2 things.
1. Increasing to a 64-bit counter certainly solves the problem, but this
might require conversion of zk data when the current counter is stored as 32-bit
2. A client that relies on a unique version that it uses as a reference
outside of zk should verify that a ve
The invariant is that the value should be increasing except in failure
modes. You found a somewhat surprising failure mode.
Please compute how long it would take for a 64-bit counter to overflow if
incremented a million times per second. (hint, half a million years).
Remember that zk only does thi
Thanks a lot for your inputs, Ted.
On Thu, Jun 15, 2023 at 2:52 PM Ted Dunning wrote:
> Breaking a semantic invariant is a dangerous solution here.
Totally agree. We should not break a semantic invariant if there is one.
What's the semantic invariant here and how ZK is supposed to behave in the
Breaking a semantic invariant is a dangerous solution here. It would be
better to use a longer integer.
On Thu, Jun 15, 2023, 13:35 Li Wang wrote:
> Thanks for the response, Enrico.
>
> Please see comments below.
>
> On Thu, Jun 15, 2023 at 5:47 AM Enrico Olivelli
> wrote:
>
> > Li,
> > thank
Andor Molnar created ZOOKEEPER-4705:
---
Summary: Restrict GitHub merge butten to allow squash commit only
Key: ZOOKEEPER-4705
URL: https://issues.apache.org/jira/browse/ZOOKEEPER-4705
Project: ZooKeepe
Thanks for the response, Enrico.
Please see comments below.
On Thu, Jun 15, 2023 at 5:47 AM Enrico Olivelli wrote:
> Li,
> thanks for reporting your problem.
>
> Most likely you have found a bug.
>
> I have one question, related to your use case,
> is the problem that the numbers are not "uniqu
Yeah, it works fine with the hardcoded classifier.
I'll make this Pulsar approach in the patch.
Andor
On Thu, 2023-06-15 at 19:50 +0200, Enrico Olivelli wrote:
> I may be wrong but Epoll is only a Linux thing probably.
>
> You don't have it on Mac or Windows
>
> Enrico
>
> Il Gio 15 Giu 2023
I may be wrong but Epoll is only a Linux thing probably.
You don't have it on Mac or Windows
Enrico
Il Gio 15 Giu 2023, 19:49 Andor Molnar ha scritto:
> Interesting that with only the BOM included and the dependencies
> without the classifier, Netty doesn't load the native epoll selector,
> bu
Interesting that with only the BOM included and the dependencies
without the classifier, Netty doesn't load the native epoll selector,
but loads the native SSL library.
I'm confused.
In the Pulsar example the sub-project dependency has a hardcoded
classifier:
io.netty
netty-transpo
Thanks Enrico, we've made a mistake though: discussed that fips-mode
will be enabled by default on master branch and disabled by default on
branch-3.8.
Let me create a separate pull request for that.
Andor
On Thu, 2023-06-15 at 14:39 +0200, Enrico Olivelli wrote:
> Il giorno mer 14 giu 2023 al
Li,
thanks for reporting your problem.
Most likely you have found a bug.
I have one question, related to your use case,
is the problem that the numbers are not "unique" or that the number is
not monotonically increasing ?
Do you have 2147483647 concurrent sessions and you found that two
sessions
Your patches have been merged to master branch and they will be
shipped with 3.9.0.
Thank you
Enrico
Il giorno sab 10 giu 2023 alle ore 18:13 Kezhu Wang
ha scritto:
>
> Hi all and committers,
>
> [ZOOKEEPER-4471][1] reported that `AddWatchMode.PERSISTENT` could be
> partially removed by `Watcher
Thanks for volounteering
Il giorno mer 14 giu 2023 alle ore 14:31 Szalay-Bekő Máté
ha scritto:
>
> Hello ZooKeepers,
>
> release 3.8.1 happened this January and I volunteered to do 3.8.2 soon,
> maybe in June or early July. Let me know if someone would also like to do
> it, I am happy to hand it
Il giorno mer 14 giu 2023 alle ore 13:43 Andor Molnar
ha scritto:
>
> PR has been created with the proposed resolution:
>
> https://github.com/apache/zookeeper/pull/2008
Committed to master and branch-3.8
Thank you
Enrico
> Please review.
>
> Thanks,
> Andor
>
>
>
> On Sat, 2023-06-10 at 11:25
This is our Jenkins job
https://ci-hadoop.apache.org/view/ZooKeeper/job/ZooKeeper-Java-EA/
The latest build passed.
We are using openjdk-21-ea+21 on Linux.
it seems that it is not the very latest EA (is it +26 ?)
>From the output of the jenkins job run:
[ZooKeeper-Java-EA] $ /home/jenkins/tool
I think that the best way currently is to add these dependencies:
Import the Netty BOM in the main pom.xml
io.netty
netty-bom
${netty.version}
pom
import
declare netty dependencies without setting the version and the classifier
io.netty
netty-tcnative-boringssl-static
This is
Il giorno gio 15 giu 2023 alle ore 13:58 Andor Molnar
ha scritto:
>
> Hi folks,
>
> There're 64 open tickets which has fixVersion = 3.9.0
> I'll remove the fixVersion from all of them except the ones that we
> marked as release blockers.
I agree.
It is useless to keep the "fixVersion" on an issue
Hi folks,
There're 64 open tickets which has fixVersion = 3.9.0
I'll remove the fixVersion from all of them except the ones that we
marked as release blockers.
Currently:
- ZOOKEEPER-4393 Problem to connect to zookeeper in FIPS mode
- ZOOKEEPER-4622 Add Netty-TcNative OpenSSL Support
- ZOOKEEPE
Hi,
I've come across the following when working on the support of native
SSL libraries. Currently ZooKeeper supports loading the native epoll-
based event loop of Netty, but a build profile which would download the
required dependencies is not shipped with our product.
This is perfectly okay sinc
20 matches
Mail list logo
