Mailing-List: contact user-help@cassandra.apache.org; run by ezmlm
Precedence: bulk
Reply-To: user@cassandra.apache.org
Date: Fri, 25 Mar 2016 00:32:15 +0000
From: Aleksey Yeschenko <aleksey@apache.org>
To: Dikang Gu <dikang85@gmail.com>
Cc: dev@cassandra.apache.org, cassandra <user@cassandra.apache.org>
Message-ID: <etPan.56f4870f.66ab2716.4d08@AY15rMBP.home>
In-Reply-To: 
 <CAPw3+DhC2rGEeitPZXczNfFCaC=R0HSzGPNwUQ0n_jYqHN-_YQ@mail.gmail.com>
References: 
 <CAPw3+DjJ6Eg4EncGtMZEZAE+PinZEvew7nTXV6j4JxsF34RoLg@mail.gmail.com>
 <etPan.56f42bab.2ac8cc70.4d08@AY15rMBP.home>
 <CAPw3+DhC2rGEeitPZXczNfFCaC=R0HSzGPNwUQ0n_jYqHN-_YQ@mail.gmail.com>
Subject: Re: Counter values become under-counted when running repair.
MIME-Version: 1.0
Content-Type: multipart/alternative; boundary="56f4870f_2e3aaae2_4d08"

--56f4870f_2e3aaae2_4d08
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: quoted-printable
Content-Disposition: inline

Best open a JIRA ticket and I=E2=80=99ll have a look at what could be the=
 reason.

--=C2=A0
AY

On 24 March 2016 at 23:20:55, Dikang Gu (dikang85=40gmail.com) wrote:

=40Aleksey, we are writing to cluster with CL =3D 2, and reading with CL =
=3D 1. =20
And overall we have 6 copies across 3 different regions. Do you have =20
comments about our setup=3F =20

During the repair, the counter value become inaccurate, we are still =20
playing with the repair, will keep you update with more experiments. But =
do =20
you have any theory around that=3F =20

Thanks a lot=21 =20
Dikang. =20

On Thu, Mar 24, 2016 at 11:02 AM, Aleksey Yeschenko <aleksey=40apache.org=
> =20
wrote: =20

> After repair is over, does the value settle=3F What CLs do you write to=
 your =20
> counters with=3F What CLs are you reading with=3F =20
> =20
> -- =20
> AY =20
> =20
> On 24 March 2016 at 06:17:27, Dikang Gu (dikang85=40gmail.com) wrote: =20
> =20
> Hello there, =20
> =20
> We are experimenting Counters in Cassandra 2.2.5. Our setup is that we =
=20
> have =20
> 6 nodes, across three different regions, and in each region, the =20
> replication factor is 2. Basically, each nodes holds a full copy of the=
 =20
> data. =20
> =20
> When are doing 30k/s counter increment/decrement per node, and at the =20
> meanwhile, we are double writing to our mysql tier, so that we can meas=
ure =20
> the accuracy of C* counter, compared to mysql. =20
> =20
> The experiment result was great at the beginning, the counter value in =
C* =20
> and mysql are very close. The difference is less than 0.1%. =20
> =20
> But when we start to run the repair on one node, the counter value in C=
* =20
> become much less than the value in mysql, the difference becomes larger=
 =20
> than 1%. =20
> =20
> My question is that is it a known problem that the counter value will =20
> become under-counted if repair is running=3F Should we avoid running re=
pair =20
> for counter tables=3F =20
> =20
> Thanks. =20
> =20
> -- =20
> Dikang =20
> =20
> =20


-- =20
Dikang =20

--56f4870f_2e3aaae2_4d08
Content-Type: text/html; charset="utf-8"
Content-Transfer-Encoding: quoted-printable
Content-Disposition: inline

<html><head><style>body=7Bfont-family:Helvetica,Arial;font-size:13px=7D</=
style></head><body style=3D=22word-wrap: break-word; -webkit-nbsp-mode: s=
pace; -webkit-line-break: after-white-space;=22><div id=3D=22bloop=5Fcust=
omfont=22 style=3D=22font-family:Helvetica,Arial;font-size:13px; color: r=
gba(0,0,0,1.0); margin: 0px; line-height: auto;=22>Best open a JIRA ticke=
t and I=E2=80=99ll have a look at what could be the reason.</div> <br> <d=
iv id=3D=22bloop=5Fsign=5F1458865867251725056=22 class=3D=22bloop=5Fsign=22=
><div style=3D=22font-family:helvetica,arial;font-size:13px=22>--&nbsp;<b=
r>AY</div></div> <br><p class=3D=22airmail=5Fon=22>On 24 March 2016 at 23=
:20:55, Dikang Gu (<a href=3D=22mailto:dikang85=40gmail.com=22>dikang85=40=
gmail.com</a>) wrote:</p> <blockquote type=3D=22cite=22 class=3D=22clean=5F=
bq=22><span><div><div></div><div>=40Aleksey, we are writing to cluster wi=
th CL =3D 2, and reading with CL =3D 1.
<br>And overall we have 6 copies across 3 different regions. Do you have
<br>comments about our setup=3F
<br>
<br>During the repair, the counter value become inaccurate, we are still
<br>playing with the repair, will keep you update with more experiments. =
But do
<br>you have any theory around that=3F
<br>
<br>Thanks a lot=21
<br>Dikang.
<br>
<br>On Thu, Mar 24, 2016 at 11:02 AM, Aleksey Yeschenko &lt;aleksey=40apa=
che.org&gt;
<br>wrote:
<br>
<br>&gt; After repair is over, does the value settle=3F What CLs do you w=
rite to your
<br>&gt; counters with=3F What CLs are you reading with=3F
<br>&gt;
<br>&gt; --
<br>&gt; AY
<br>&gt;
<br>&gt; On 24 March 2016 at 06:17:27, Dikang Gu (dikang85=40gmail.com) w=
rote:
<br>&gt;
<br>&gt; Hello there,
<br>&gt;
<br>&gt; We are experimenting Counters in Cassandra 2.2.5. Our setup is t=
hat we
<br>&gt; have
<br>&gt; 6 nodes, across three different regions, and in each region, the=

<br>&gt; replication factor is 2. Basically, each nodes holds a full copy=
 of the
<br>&gt; data.
<br>&gt;
<br>&gt; When are doing 30k/s counter increment/decrement per node, and a=
t the
<br>&gt; meanwhile, we are double writing to our mysql tier, so that we c=
an measure
<br>&gt; the accuracy of C* counter, compared to mysql.
<br>&gt;
<br>&gt; The experiment result was great at the beginning, the counter va=
lue in C*
<br>&gt; and mysql are very close. The difference is less than 0.1%.
<br>&gt;
<br>&gt; But when we start to run the repair on one node, the counter val=
ue in C*
<br>&gt; become much less than the value in mysql, the difference becomes=
 larger
<br>&gt; than 1%.
<br>&gt;
<br>&gt; My question is that is it a known problem that the counter value=
 will
<br>&gt; become under-counted if repair is running=3F Should we avoid run=
ning repair
<br>&gt; for counter tables=3F
<br>&gt;
<br>&gt; Thanks.
<br>&gt;
<br>&gt; --
<br>&gt; Dikang
<br>&gt;
<br>&gt;
<br>
<br>
<br>-- =20
<br>Dikang
<br></div></div></span></blockquote></body></html>
--56f4870f_2e3aaae2_4d08--