brad's life - diskchecker.pl [entries|archive|friends|userinfo]
Brad Fitzpatrick

[ website | bradfitz.com ]
[ userinfo | livejournal userinfo ]
[ archive | journal archive ]

diskchecker.pl [May. 9th, 2005|08:49 pm]
Previous Entry Share Next Entry
[Tags|, ]

Dear Slashdot comment flood: I didn't write the summary in the Slashdot story. The submitter did. I know the disks themselves don't handle an fsync().

I know fsync() only tells the operating system to flush. This script tests whether fsync works end-to-end. A database in userspace can only fsync... it can't send special IDE or SCSI "flush your buffers" commands to the disks. So that's what I care about and what I want to test: that a database can be Durable.

The problem is manufacturers shipping SCSI disks with a dangerous option (hardware write-caching) enabled by default. It makes sense for consumer ATA stuff, but for SCSI disks that already have reliable TCQ, there's much less point. And any respectable raid card should just disable write-back caching on the disks if the raid card has its own nvram-backed cache, but LSI doesn't anymore (they used to, but stopped).

And I'm glad Linux is finally starting to tell the disks to flush on an fsync. But months ago I was stuck with databases that couldn't survive power outages and needed a way to test whether everything from the filesystem to the block device driver to the disks themselves were doing what the database expected when it did an fsync. That's no component's job alone, so I needed to test everything together.


Remember my disk-checker program I wrote about before? I'd never released it because it was too hard to use, but now it's dead simple, so here it is:

http://code.sixapart.com/svn/tools/trunk/diskchecker.pl Edit: now https://gist.github.com/3172656

Run it and be amazed how much your disks/raid/OS lie. ("lie" = an fsync doesn't work)

It seems everything from PATA consumer disks to high-end server-class SCSI disks lie like crazy. Yes, that includes SATA there in the middle. I'll discuss fixing your storage components in a second.

In a nutshell, run it like this:

Tester machine (machine that won't crash):
$ diskchecker.pl -l

And then just let it chill. (the el is for listen). This program will listen (on port 5400 if no number follows -l) and will write one tiny file per host to /tmp/. It can be run as any user.

Machine being tested (machine you're going to pull the power cable on)
$ diskchecker.pl -s TESTERMACHINE create test_file 500

That creates a 500 MB file named "test_file" and it reports everything it's about to do and does to the TESTERMACHINE (which can be an IP or hostname).

Now, pull the power cable on the machine being tested. Don't turn it off nicely. Don't just control-C the program. Wait a couple seconds and plug your testee machine back in and reboot it. When it's back up, do:

$ diskchecker.pl -s TESTERMACHINE verify test_file

If the server process is still running, the machine you just killed will connect to the server and information about what's supposed to be where. The client will then verify it and produce an error report.

What you should see is:

Total errors: 0.

But you probably won't. You'll probably see an error count and a histogram of errors per seconds-before-crash. Most RAID cards lie (especially LSI ones), some OSes lie (rare), and most disks lie (doesn't matter how expensive or cheap they are). They lie because their competitors do and they figure it's more important to look competitive because the magazines only print speed numbers, not reliability stats. They must figure people who care about their disks working know how to test/fix their disks.

Ways to maybe fix your disk:

hdparm -W 0 /dev/hda -- worked on a crap office PATA disk (and it failed otherwise)
scsirastools -- need this on lots of SCSI disks. you'll probably have to remove your SCSI disks from your RAID card and fix the disks directly, since RAID cards very often won't disable it for you

Anybody have anything else to add?

Enjoy.
LinkReply

Comments:
Page 1 of 2
<<[1] [2] >>
[User Picture]From: wetzel
2005-05-10 04:07 am (UTC)

(Link)

will this bit of script work on windows (w/ activeperl), or will it freak on me like it's retarded?

or is the only way to find out to test it?
[User Picture]From: brad
2005-05-10 05:20 am (UTC)

(Link)

Should work. I use nothing fancy. All pure, base Perl.
[User Picture]From: electromage
2005-05-10 04:14 am (UTC)

(Link)

There's far too little emphasis on reliability in benchmarks and other comparisons.
[User Picture]From: claystorm
2005-05-10 04:56 am (UTC)

Linux only?

(Link)

Is this Linux only, or can it be run in a Windows environment with a perl shell running in Windows?
[User Picture]From: brad
2005-05-10 05:20 am (UTC)

Re: Linux only?

(Link)

Should work on Windows unmodified.
[User Picture]From: valiskeogh
2005-05-10 05:17 am (UTC)

(Link)

um, yes, an executable or com would be nice for us non linux freaks :)
From: sherm
2005-05-10 05:26 am (UTC)

(Link)

It's perl. You can't expect many perl developers who give stuff away for free to build a Windows executable for you.

Try ActivePerl. It's free, but they probably send you a bunch of crap if you give them a real e-mail address and don't unsubscribe.
(Deleted comment)
[User Picture]From: endquote
2005-05-10 05:30 am (UTC)

(Link)

Maybe send a link to this off to the folks that review storage for us normal folk? It would be interesting to see Cnet or PC Mag include information like this along with the speed and volume specs.
[User Picture]From: brad
2005-05-10 05:39 am (UTC)

(Link)

Go for it.
[User Picture]From: gholam
2005-05-10 05:48 am (UTC)

(Link)

Isn't this... caching? By pulling the plug, you kill whatever data is pending write in drive's volatile RAM cache, and that's what they have UPSes and stupidly expensive battery-backup-capable RAID controllers for.
[User Picture]From: brad
2005-05-10 05:58 am (UTC)

(Link)

Re-read my post. I'm testing whether the entire storage stack respects the fsync() system call. All the way from the OS to the drivers to the raid array to the disks themselves.

The fsync() system call says: "Stop caching, it's very important that everything I've given you now must be on disk, and don't return to me with an answer until it has."

So this program tests that your fsync() works as advertised and some part of the storage stack isn't faking the fsync. (it's usually the disks themselves, against specs, and unknown the operating system, which thinks the disks are behaving)

Otherwise caching's just fine and it's done all over. My complaint is when it's done when you tell it not to.
(no subject) - (Anonymous) Expand
[User Picture]From: ydna
2005-05-10 07:34 am (UTC)

(Link)

Sweet, Brad. Perfect timing. I've got an ATA-over-Ethernet evaluation unit from Coraid on its way. I'm planning several tests on this equipment for a demo and presentation up here in June (background: see my rant and LUG discussion). It'll be fun to add this test to the mix (and pull the plug at different places in the connection to see which part does most of the lying). Thankee much.
[User Picture]From: brad
2005-05-10 07:52 am (UTC)

(Link)

I still don't really get AoE. Who's it supposed to be for?

BTW, you mentioned in one of those posts capturing ethernet frames in userspace to make an AoE server. Look at "tap". In the kernel source, read:

./Documentation/networking/tuntap.txt

And docs for CONFIG_FILTER=y and CONFIG_PACKET=y. Between the three, you could start to write an AoE server.
[User Picture]From: boggyb
2005-05-10 08:50 am (UTC)

(Link)

Some Windows-related details of disk caching that I've found out about (and may be useful to those of you using Windows).

Under Windows NT (that's 2k, XP, 2k3 as well - remember those are NT 5.x), you can disable disk caching by going to the properties for the hard disk and then going to the Settings/Properties/Policies tab (is named differently under different versions but does the same thing). By default write cache is enabled except on disks containing the Active Directory. It is also disabled by default for removeable disks (USB, memory cards (but not all!), iPods, etc.) and apparently anything it thinks is SCSI (this includes some IDE/SATA controllers). I think Windows tells the disk itself to explicitly disable the disk cache as well.

When using CreateFile(), you can set the FILE_FLAG_WRITETHROUGH flag, which tells Windows to flush write caches quicker than usual, or you can set FILE_FLAG_NO_BUFFERING, which bypasses all Windows caches but has some restrictions (you have to be careful with memory alignment and read in multiples of the sector size). These are set in the dwFlagsAndAttributes parameter.
[User Picture]From: mendel
2005-05-10 02:26 pm (UTC)

(Link)

5.8.1's Getopt::Long ignores this, but 5.6.1's chokes on it.
--- diskchecker.pl?rev=HEAD     2005-05-09 20:28:18.000000000 -0400
+++ diskchecker.pl      2005-05-10 10:23:39.647304453 -0400
@@ -22,3 +22,3 @@
 usage() unless GetOptions('server=s' => \$server,
-                          'listen:5400' => \$listen);
+                          'listen:i' => \$listen);
 usage() unless $server || $listen;
[User Picture]From: mendel
2005-05-10 02:37 pm (UTC)

(Link)

Hah, no, that's not enough! You and your little ands and ors.
--- diskchecker.pl?rev=HEAD     2005-05-09 20:28:18.000000000 -0400
+++ diskchecker.pl      2005-05-10 10:33:01.437798948 -0400
@@ -21,10 +21,11 @@
 my $listen;
 usage() unless GetOptions('server=s' => \$server,
-                          'listen:5400' => \$listen);
-usage() unless $server || $listen;
-usage() if     $server && $listen;
+                          'listen:i' => \$listen);
+usage() unless $server || defined $listen;
+usage() if     $server && defined $listen;

 # LISTEN MODE:
-listen_mode($listen) if $listen;
+my $port = $listen ? $listen : 5400;
+listen_mode($port) if defined $listen;

 # CLIENT MODE:
[User Picture]From: granting
2005-05-10 07:50 pm (UTC)

(Link)

I sent you via e-mail a 1.5 meg PDF file regarding hard drives that you might be interested in. Or maybe not. But this post made me think you might.

Do you mind if I mention this tool elsewhere?
[User Picture]From: scosol
2005-05-11 01:48 pm (UTC)

(Link)

does it work on windows????? :P
[User Picture]From: pne
2005-05-11 04:16 pm (UTC)

(Link)

It should, as several previous comments mentioned.

You'd need a Perl interpreter, of course, such as ActivePerl, IndigoPerl, or Cygwin's perl.
From: orbadelic
2005-05-11 08:25 pm (UTC)

scsirastools....

(Link)

so the scsirastools are used to disable write caching on the drives themselves? Going to run this later today with a freshly minted dell running freebsd 5.4, I wonder if there are any similar tools for it.
[User Picture]From: vyhuhol
2005-05-13 08:03 am (UTC)

(Link)

Isn't this hdparm -W option causes severe performance loss?
From: (Anonymous)
2005-05-13 08:04 am (UTC)

We need a "name and shame" list or wiki

(Link)

This is very interesting, but we can only check the disks that we have physical access to.

What we need is a public "name and shame" list of disk models, so folks who *require* reliable data integrity can make informed choices.

Anyone interested in setting up an online database or Wiki?

Matthew
From: (Anonymous)
2005-05-13 03:20 pm (UTC)

Re: We need a "name and shame" list or wiki

(Link)

Matthew,

I'll talk to my boss and see if he'll allow me to create a small space on our webserver for a "name and shame" type thing that you're talking about (if no one else has done it, that is). It might not be a Wiki (just a static webpage at first) but it might be a good excuse for me to work on getting the company Wiki up and running 80).

S. Garcia
SLM Industries
~NOSPAM~ steven &DOT& garcia %AT% slmindustries ~DOT~ com ^SPAMISEVIL^
From: (Anonymous)
2005-05-13 08:15 am (UTC)

Nothing strange here, its your assumptions that are wrong

(Link)

It is strange that after finding out that fsync() never functions the way you expect, you didn't doubt your assumptions or your code!
Your assumption is wrong! fsync() is required to flush the hard-disk cache. It's function is just to flush all buffers inside the OS and the device driver to the hard-disk and it does that perfectly.
As you see, these are two different things: flushing to the device and flushing to the disk. fsync() flushes to the device but you expect it to flush to the disk (which it does not). Flush to the disk usually can only be done in a system level program (like a device driver) and not a user program.
[User Picture]From: brad
2005-05-13 02:24 pm (UTC)

Re: Nothing strange here, its your assumptions that are wrong

(Link)

Yes, but that's all a userspace program like a database can do. And while I know that fsync() says it can't promise it makes it to disk if write-caching is enabled, this tool is a great way to see if your disk's write-caching is indeed on.

The old version of this script used the raw(8) interface to bypass all filesystems and kernel buffers, but it produced the same results as the fsync version, so I made it just use fsync so it was easier to use. (don't need a spare block device on the disk(s) being tested handy....)
[User Picture]From: error10
2005-05-13 08:29 am (UTC)

hdparm -W 0 /dev/hda

(Link)

I needed this two months ago! I suffered massive filesystem corruption due to exactly this problem. Well, this and the fact that ext3 still sucks. And why doesn't Red Hat do reiserfs anyway?
From: (Anonymous)
2005-05-13 09:45 am (UTC)

OS shutdown?

(Link)

Does anyone know how the disk knows to do the write during an OS shutdown?

There must be a trick otherwise we'd be losing data at every reboot..
[User Picture]From: wdomburg
2005-05-13 10:11 am (UTC)

(Link)

Did you disable writeback cache on both the controller and drive? I only care whether that fails to ensure fsync is writing to stable storage.
[User Picture]From: brad
2005-05-13 02:26 pm (UTC)

(Link)

Yes, we do. And that's what I wrote the script for: to verify it all works. Because we have got bitten by certain hardware (LSI RAID cards) where it doesn't always work.
From: parthurb
2005-05-13 11:22 am (UTC)

Nice one

(Link)

I have deep suspicions that device mapper is not very good at making sure fsyncs() actually hit disk.

In response to iSCSI, if you are going to push a lot of data the cost of iSCSI (Cisco) gear is higher than low end fcal gear, and there is a big benefit with fcal which is, fibre doesn't carry electric shocks :-).
From: (Anonymous)
2005-05-13 11:37 am (UTC)

Apparent

(Link)

I didn`t know this untill I was asked to supply an identical disk for a recovery job in a raid-0 crash. I had 2 80gb WD disks, and one of them had crashed making the whole raid array unreadable. When we bought a spare disk of *exactly* the same factory type, but from a later period, it appeared that the new disk (of the exact same user size) had a media that was a tiny bit smaller (it had less so called 'spare blocks' that are used by the disk to park data that can be damaged sooner or later). The end result was that the raw data block of the old disk didn`t fit on the new disk.. in the end we ended up using 2 maxtor drives of 120Gb and praying, but it worked.

So yes, most modern HD disks seem to have hardware-driven relocation and repair strategies, moving blocks into spare space when a error is detected (or maybe predicted) in a block.

From: (Anonymous)
2005-05-13 11:46 am (UTC)

WHat make you sure it's the HW?

(Link)

I know for a fact that bugs in ext3 file system (on Linux of course) will cause an fsync() operation to succeed even if the disk itself is turned off (it was a FC external disk).

The problem was tracked to a bug in ext3 (since then fixed I believe, but still found in RHEL3 at the very least).

In short - yes, fsync() lies 0- but what makes you sure it's the hardware fault?

Gilad Ben-Yossef
http://benyossef.com
[User Picture]From: brad
2005-05-13 02:30 pm (UTC)

Re: WHat make you sure it's the HW?

(Link)

Because the script produces the same results when it's done via the raw(8) interface[1], bypassing the filesystem and kernel buffers. And it produces near identical results with any filesystem.

But once I fix the disk's write-caching, all tests work, filesystem or not.

[1] You should see the ugly Perl required to do aligned writes.
From: (Anonymous)
2005-05-13 11:58 am (UTC)

Do you have write cache enabled?

(Link)

Maybe this /. post applies here.
http://hardware.slashdot.org/comments.pl?sid=149349&cid=12517596

excerpt:

From Linux --

NOTES
In case the hard disk has write cache enabled, the data may not really
be on permanent storage when fsync/fdatasync return.
[User Picture]From: brad
2005-05-13 02:32 pm (UTC)

Re: Do you have write cache enabled?

(Link)

Read my fucking post. Yes, I know. This whole script is used to detect if write caching is enabled. It should never be, though. The kernel does caching for you, so there's no reason for your disk to be doing it. Unless you want your applications not to be able to fsync.
From: (Anonymous)
2005-05-13 12:38 pm (UTC)

Official release of the tool? Any experience with Spew?

(Link)

Brad,

I just added to the InnoDB Configuration (http://dev.mysql.com/doc/mysql/en/innodb-configuration.html) section of the MySQL manual (http://dev.mysql.com/doc/mysql/en/) a warning message about fsync() that I had prepared some time ago. I would like to mention your tool there, to give users an easy tool for pull the plug tests. However, I would like to link to a more official address than a CVS gateway.

Does anyone have any experience with Spew (http://spew.berlios.de/)? Could it be used in place of Brad's script?

Unexpected shutdowns are not the only source of InnoDB corruption. Sometimes the operating system or hardware writes data to wrong offset in the file, or only writes part of the data, or the memory gets somehow corrupted, especially on some fairly recent 64-bit systems.

Marko Mäkelä
Senior Software Engineer
Innobase Oy
[User Picture]From: desh
2005-05-13 01:21 pm (UTC)

(Link)

Congrats, Brad Fitzgerald, on making Slashdot with this. So what if they don't know your actual name...
From: (Anonymous)
2005-05-13 04:49 pm (UTC)

MacOS X

(Link)

This darwin-dev article may also shed some light:

http://lists.apple.com/archives/darwin-dev/2005/Feb/msg00072.html

MacOS X provides a F_FULLFSYNC fcntl to provide the best possible chance of getting data written to disk. Is there any way you could test with that turned on, to see if that behaves better?
Page 1 of 2
<<[1] [2] >>