DomainKey-Signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma;
	h=mime-version:in-reply-to:references:date:message-id:subject:from:to
	:cc:content-type:content-transfer-encoding;
	b=dH0ZmlBuIon+NdA7Sf4x5dUliAIr4O+/vScFzq4orBpLSpMoGXigju+qhr1IDWpURI
	+M1YhLwg/O6lddgeyyhi0pmL5flSXkNcF0PqVD752ppsfiaYtv0AmffB82adDYp9Z2pb
	8W+qqGaWMjKmX1s6JAQwNny6cl0A6JklfgsfU=
MIME-Version: 1.0
In-Reply-To: <BANLkTim1EoDEmdrOSfMqVkVe9Pa4KP_TYQ@mail.gmail.com>
References: <AANLkTim1aO2hG+D4h_VFXas8GTL6TQMLj7aDc7tJ0vok@mail.gmail.com>
	<AANLkTinjsS7Aj0ZoMVq3ZDhKepBsZ_HOpBqEjRrabk_J@mail.gmail.com>
	<BANLkTinq7KSstviqjGfnaKh7YOfan7sjig@mail.gmail.com>
	<BANLkTin9bB+LAScMMBnuJn9uROXOkPsQLw@mail.gmail.com>
	<BANLkTi=j-+Cauczk-q+pJuASp_PPOjXgEQ@mail.gmail.com>
	<BANLkTi=Yr7QBdU-_Gc_9S+gUi0VO6s9u+A@mail.gmail.com>
	<BANLkTim1EoDEmdrOSfMqVkVe9Pa4KP_TYQ@mail.gmail.com>
Date: Mon, 18 Apr 2011 13:19:03 +0800
Message-ID: <BANLkTim4e8PQKHofUSOcQHsPsLgcPmnEVw@mail.gmail.com>
Subject: Re: REINDEX takes half a day (and still not complete!)
From: Phoenix Kiula <phoenix.kiula@gmail.com>
To: Scott Marlowe <scott.marlowe@gmail.com>
Cc: Merlin Moncure <mmoncure@gmail.com>, pgsql-performance@postgresql.org
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable

Btw, hardware is not an issue. My db has been working fine for a
while. Smaller poorer systems around the web run InnoDB databases. I
wouldn't touch that with a barge pole.

I have a hardware RAID controller, not "fake". It's a good quality
battery-backed 3Ware:
http://192.19.193.26/products/serial_ata2-9000.asp


On Mon, Apr 18, 2011 at 1:14 PM, Phoenix Kiula <phoenix.kiula@gmail.com> wr=
ote:
> Thanks for these suggestions.
>
> I am beginning to wonder if the issue is deeper.
>
> I set autovacuum to off, then turned off all the connections to the
> database, and did a manual vacuum just to see how long it takes.
>
> This was last night my time. I woke up this morning and it has still
> not finished.
>
> The maintenance_men given to the DB for this process was 2GB.
>
> There is nothing else going on on the server! Now, even REINDEX is
> just failing in the middle:
>
>
> # REINDEX INDEX new_idx_userid;
> server closed the connection unexpectedly
> =A0 =A0 =A0 =A0This probably means the server terminated abnormally
> =A0 =A0 =A0 =A0before or while processing the request.
> The connection to the server was lost. Attempting reset: Failed.
>
>
> What else could be wrong?
>
>
>
>
> On Mon, Apr 18, 2011 at 2:38 AM, Scott Marlowe <scott.marlowe@gmail.com> =
wrote:
>> On Sun, Apr 17, 2011 at 10:59 AM, Phoenix <phoenix.kiula@gmail.com> wrot=
e:
>>> TOP does not show much beyond "postmaster". How should I use TOP and
>>> what info can I give you? This is what it looks like:
>>
>> We're basically looking to see if the postmaster process doing the
>> vacuuming or reindexing is stuck in a D state, which means it's
>> waiting on IO.
>> hot the c key while it's running and you should get a little more info
>> on which processes are what.
>>
>>> =A04799 postgres =A015 =A0 0 =A0532m =A094m =A093m D =A00.7 =A01.2 =A0 =
0:00.14
>>> postmaster
>>
>> That is likely the postmaster that is waiting on IO.
>>
>>> VMSTAT 10 shows this:
>>>
>>> =A0r =A0b =A0 swpd =A0 free =A0 buff =A0cache =A0 si =A0 so =A0 =A0bi =
=A0 =A0bo =A0 in =A0 =A0cs us sy id wa
>>> =A03 14 =A099552 =A017900 =A041108 7201712 =A0 =A00 =A0 =A00 =A0 =A042 =
=A0 =A011 =A0 =A00 =A0 =A0 0 =A08 34 41 16
>>> =A02 17 =A099552 =A016468 =A041628 7203012 =A0 =A00 =A0 =A00 =A01326 =
=A0 =A084 1437 154810 =A07 66 12 15
>>> =A03 =A07 =A099476 =A016796 =A041056 7198976 =A0 =A00 =A0 =A00 =A01398 =
=A0 =A096 1453 156211 =A07 66 21 =A06
>>> =A03 17 =A099476 =A017228 =A039132 7177240 =A0 =A00 =A0 =A00 =A01325 =
=A0 =A068 1529 156111 =A08 65 16 11
>>
>> So, we're at 11 to 15% io wait. =A0I'm gonna guess you have 8 cores /
>> threads in your CPUs, and 1/8th ot 100% is 12% so looks like you're
>> probably IO bound here. =A0iostat tells us more:
>>
>>> The results of "iostat -xd 10" is:
>>> Device: =A0 =A0rrqm/s wrqm/s =A0 r/s =A0 w/s =A0rsec/s =A0wsec/s =A0 =
=A0rkB/s =A0 =A0wkB/s
>>> avgrq-sz avgqu-sz =A0 await =A0svctm =A0%util
>>> sda =A0 =A0 =A0 =A0 =A00.00 =A0 7.41 =A00.30 =A03.50 =A0 =A02.40 =A0 87=
.29 =A0 =A0 1.20 =A0 =A043.64
>>> =A0 23.58 =A0 =A0 0.13 =A0 32.92 =A010.03 =A0 3.81
>>> sdb =A0 =A0 =A0 =A0 =A00.00 =A0 0.00 =A00.00 =A00.00 =A0 =A00.00 =A0 =
=A00.00 =A0 =A0 0.00 =A0 =A0 0.00
>>> =A0 =A00.00 =A0 =A0 0.00 =A0 =A00.00 =A0 0.00 =A0 0.00
>>> sdc =A0 =A0 =A0 =A0 =A00.00 =A018.32 158.26 =A04.10 2519.32 =A0180.98 =
=A01259.66
>>> 90.49 =A0 =A016.63 =A0 =A013.04 =A0 79.91 =A0 6.17 100.11
>>
>> 100% IO utilization, so yea, it's likely that your sdc drive is your
>> bottleneck. =A0Given our little data is actually moving through the sdc
>> drive, it's not very fast.
>>
>>> Device: =A0 =A0rrqm/s wrqm/s =A0 r/s =A0 w/s =A0rsec/s =A0wsec/s =A0 =
=A0rkB/s =A0 =A0wkB/s
>>
>>> 8GB memory in total. 1GB devoted to PGSQL during these operations.
>>> Otherwise, my settings are as follows (and yes I did make the vacuum
>>> settings more aggressive based on your email, which has had no
>>> apparent impact) --
>>
>> Yeah, as it gets more aggressive it can use more of your IO bandwidth.
>> =A0Since you
>>
>>> What else can I share?
>>
>> That's a lot of help. =A0I'm assuming you're running software or
>> motherboard fake-raid on this RAID-1 set? =A0I'd suggest buying a $500
>> or so battery backed caching RAID controller first, =A0the improvements
>> in performance are huge with such a card. =A0You might wanna try testing
>> the current RAID-1 set with bonnie++ to get an idea of how fast it is.
>>
>