-
warp
Swarup: you can run the update script which has that function I think.
-
kepstin-work joined the channel
-
-
Swarup
which update script?... admin/sql/updates/20120731-constrain-catno-whitespace.sql - i don't see this in my sandbox
-
should i just run CreateFunctions.sql
-
ha nm..i see it ..sorry..
-
-
is that ok to continue
-
luks
warp: as far as I can tell, that script is only for master servers
-
warp
luks: ok
-
luks: do you have an suggestions on how Swarup should continue?
-
luks
I guess creating just the functions from that script will work
-
but I never managed to run the whole upgrade process, so I'm not sure
-
Swarup
ok let me try that..but i would expect others to have issue ..right?
-
luks
after hitting the postgresql 8.4 issue I gave up
-
Swarup: yes, everybody I've seen trying to upgrade today ended up with a reimport instead :)
-
Swarup
hmm
-
ok probably that's safer in anycase..let me go that route..
-
luks
it seems that creating the functions should work though
-
-
ruaok
Swarup: we're trying to get a fix for the 8.4 problem out. but a reimport will likely be faster. :(
-
warp
ruaok: note that this was not the array_agg/ORDER problem
-
ruaok
ah.
-
things looked so promising yesterday. :(
-
Swarup
we're on posgres 9 i believe..i just loaded those missing functions manually...running it now..if it still fails will go with re-import
-
ruaok
good. then you should be fine, Swarup
-
warp
ruaok: both machines here upgraded without trouble yesterday. but that's with the friday dump on ubuntu supplied 9.x.
-
ruaok
same here.
-
but I should've gone with the stock 8.4 pg install for my testing.
-
I've noted the problem in the migration doc, so we that do our testing with the lowest version of pg supported
-
warp
ruaok: in the months leading up to the schema change we should perhaps ping our customers and ask what postgres they're running exactly.
-
that will give us a better idea which versions are important to test.
-
ruaok
yeah.
-
I think for the next release we should specify a min of 9.x.
-
Swarup
Rob..do u recommend to upgrade search server too along with api servers for this patch..
-
ruaok
I think once I start the schema change process on feb 1, I will mail our customers and tell them that.
-
warp
obviously that's easiest for us, we're all on 9.x anyway. I don't know how easy it is for customers to upgrade.
-
reosarevok
Someone mentioned debian stable has 8.4
-
ruaok
Swarup: can't hurt. the upgrade is easy -- you just need to put a new war in place for the search servers.
-
reosarevok: debian stable is also ancient
-
Swarup
ok will do..thanks
-
ruaok
good luck Swarup
-
warp
ruaok: if we still have a few customers on 8.4 we can perhaps support a (re)import on 8.4, but not upgrade.
-
reosarevok
Sure, but I imagine some people prefer "stable" to "new", especially companies
-
ruaok nods at warp
-
adhawkins is on Debian with 8.4
-
warp
reosarevok: there _should_ be a new debian stable in february.
-
adhawkins
No real reason why I couldn't update my VM with MB server to testing though
-
That's all it does.
-
ruaok
oh, I need to get a new VM going too.
-
adhawkins
ruaok: I think I mentioned this last time, but if you could create a second disk and mount it for postgres it'd help people
-
Make it much easire when you run out of space.
-
easier
-
ruaok
yes.
-
and then we could put out data updates without having to update everything else.
-
adhawkins
That's true.
-
ruaok
1GB root and then a 10GB /mnt/db or something?
-
adhawkins
Something like that sounds good.
-
A dynamic disk will mean the second bit will only be as big as the data on it and can grow as necessary
-
ruaok
should I do a 64bit or 32 bit install?
-
warp
32bit seems safer.
-
adhawkins
Yeah, a lot of linux installs are still 32 bit these days (mine included).
-
I plan on changing that next time I reinstall though.
-
warp
You can run a 32bit VM on both a 32bit and 64bit machine. I'm not sure a 32bit host can run a 64bit guest yet.
-
adhawkins
warp: I don't think mine can.
-
luks
I often do that (64bit vm on 32bit host)
-
for building
-
adhawkins
Depends on the processor and VM software I think luks
-
kepstin-work
a 32bit os with a 64bit processor with hardware virtualization can run a 64bit guest
-
adhawkins
kepstin-work: I don't think my atom can though :)
-
kepstin-work
adhawkins: yeah. requires hardware virtualization extensions; atom doesn't have that.
-
warp
kepstin-work: looks like a 32bit os on a 32bit CPU can also run 64bit guests if the CPU has the appropriate VT support.
-
kepstin-work
warp: that's what I said.
-
er, wait, no
-
it has to be a 64bit cpu
-
a 32bit cpu can't run 64bit programs no matter what, unless you do slow software emulation.
-
ruaok
I think I'll just continue with a 32bit one for now.
-
kepstin-work notes that in x86, running a 32bit os on a 64bit cpu is stupid, because you lose half your cpu registers.
-
luks
but you also get 64bit pointers that you don't need
-
kepstin-work
in general, having double the registers gives you more performance than you lose by having bigger pointers
-
a few unusual cases don't.
-
adhawkins has just caught up with replication packets and restarted his server.
-
adhawkins
All good.
-
kepstin-work
(this is what that 'x32' stuff recently added to linux is supposed to work around - keep the extra registers, but keep using short pointers)
-
warp
aren't larger pointers useful when you have >4GB ram?
-
kepstin-work
warp: most userspace programs don't need >4gb ram
-
but those that do, really do.
-
warp
(I have no idea how much overhead PAE is)
-
kepstin-work regularly uses the gimp with 8gb of in-ram data
-
kepstin-work
warp: pae only lets the kernel spread processes out over more ram. processes are still individually limited to 4gb
-
er, closer to 3gb in most setups, really.
-
warp
right.
-
so your gimp example would suffer badly on a 32bit os.
-
kepstin-work
yep
-
you could run a 64bit kernel with mixed 32bit/64bit userspace
-
but then you have a ram and cache penalty, because some shared libraries can't be shared between the two.
-
Freso
kepstin-work: Isn't that the most common scenario for people running 64-bit kernels?
-
kepstin-work
Freso: most linux installs use 64bit kernel with mostly-64bit userspace
-
warp
I wouldn't expect libraries to use a significant amount of memory compared to the data you're processing.
-
Freso
I know I have a fair share of lib32 packages installed.
-
kepstin-work
notable exceptions are wine and proprietary applications that are 32bit only
-
Freso
kepstin-work: But if it's "mostly-64bit userspace", is it not then also mixed? :)
-
kepstin-work
warp: instruction cache penalties when multitasking hurt - the more shared code, the better :)
-
Freso
kepstin-work: Actually, Wine can be 64bit now too.
-
kepstin-work
Freso: 64bit wine can only run 64bit windows programs
-
Freso
kepstin-work: If you compile it to be. But its support isn't very good, even for Wine standards, yet.
-
kepstin-work
you need 32bit wine for 32bit windows programs (i.e. almost all of them)
-
gentoo and some others do a multibuild wine that installs both 64bit wine and 32bit wine together.
-
Freso
Arch \o/
-
kepstin-work
windows is more or less the other way - the kernel is 64bit, but most programs are 32bit with a few exceptions.
-
Freso has set WINEARCH=win32 in his .zprofile
-
I think some newer versions of e.g. photoshop might have 64bit versions?
-
Freso shrugs
-
warp
windows is for steam, and not much else. :)
-
Freso
kepstin-work: I was merely commenting on how "mixed" was the most likely scenario. Not the degree of "mixed". :)
-
kepstin-work
either way, my desktop has 12gb ram, and I can use large amounts of that for processing big scans in the gimp :)
-
Freso: yeah; although I did actually run my laptop pure-64bit for a while
-
before I installed wine
-
Freso
Yeah. I need more RAM, I've realised. :)
-
* :|
-
warp
I thought I needed more RAM
-
Freso
Since I've begun pushing a lot of filesystem paths into hourly sync'd tmpfs file systems...
-
kepstin-work
For something silly like $120, I can put 24gb ram in my desktop.
-
warp
I've always been fine on the desktop with 4GB, but after replacing debian with ubuntu I started to have some trouble after a few weeks.
-
kepstin-work
well, that's ubuntu :/
-
luks
I never had the need for that much ram on a desktop machine
-
Freso
kepstin-work: That's pretty cheap. For 600 DKK, I could probably get a single 8 GB block. :|
-
warp
so I went out and bought another 4GB, installed it, and now I'm back to less than 3GB usage. turns out it probably just needed a reboot.
-
Freso
Haha.
-
kepstin-work
luks: you've never had the opportunity to extract raw cd audio to a ramdisk then encoding it :)
-
warp
(obviously a reboot is required for installing memory :)
-
kepstin-work has actually decoded dvd music videos to raw video on a ramdisk before
-
Freso
Yeah, I thought I'd be good with 8 GB. Before I started pushing stuff into tmpfs's.
-
kepstin-work
most compiling I do is on tmpfs as well
-
I can compile libreoffice completely in ram :/
-
warp
lol
-
Freso
All compiling I do is on tmpfs. :p
-
luks
well, most of the day all I need it a browser, music player and terminal with ssh, don't need much ram for that :)
-
Freso
But then, I don't compile LO.
-
luks
ram is for the servers I ssh to
-
kepstin-work
at this point tho, I probably use my laptop more often than my desktop
-
and my laptop only has 4gb.
-
ruaok
luks: I'm on the same page with you there.
-
my laptop hardly ever does any work. but the machines I ssh to, lots.
-
kepstin-work would say 4gb is a good amount for a typical-usage machine, and most people would be fine with a 32bit os.
-
luks
since I switched to gnome 3, I often feel sad then I alt+tab and see only three/four icons there :)
-
it made me realize how much I really use the laptop
-
kepstin-work
luks: there's an extension if you want to see separate icons for your terminal windows :)
-
it might make you feel better.
-
warp
heh, when I'm working on non-musicbrainz projects the machine I SSH into does most of the work as well.