 | 2010-09-09 The Yes Men Fix The World |
 | 2010-09-07 ed is not dead |
 | 2010-08-26 Installing Perl modules in a non root environment |
 | 2010-08-22 Magic self leviation |
 | 2010-08-20 Google Chrome does not support offline Gmail |
 | 2010-08-19 The number 48 |
 | 2010-08-12 Welsh trout mini HOWTO |
 | 2010-08-04 Fooling a NetCache proxy into fetching forbidden files |
 | 2010-07-30 The world will end on May 21, 2011 |
 | 2010-07-28 Hiding or showing a textbox with image animation using JQuery |
 | 2010-07-27 Manipulating browser cookies using Javascript |
 | 2010-07-25 Survival of the fittest book |
 | 2010-07-23 Pastafarians in Spain |
 | 2010-07-22 You have two sheep |
 | 2010-07-09 Highway bank fire |
 | 2010-07-08 Setting up a remote git repository |
 | 2010-07-06 Bye bye trusted old Macbook |
 | 2010-06-28 John Cleese on Football |
 | 2010-06-23 ABN Amro and the Pathetic Customer Service Dept. |
 | 2010-06-22 Wally does not like criticism |
 | 2010-06-14 Soccermatch Netherlands vs Denmark |
 | 2010-06-13 Lazy Cat |
 | 2010-06-08 Reading public Buzz using the Google API |
 | 2010-06-07 A Personal Letter from Steve Martin |
 | 2010-06-05 Sushi Saturday |
 | 2010-06-04 Suppressing the Enter key with Javascript |
 | 2010-05-31 Temporal spacial anomaly on the Dutch highway |
 | 2010-05-23 Greenhost will not log your traffic |
 | 2010-05-10 Jarlsberg Webapp Exploits |
 | 2010-05-04 A Thought Experiment |
 | 2010-05-03 SafeEdit information updated |
 | 2010-05-01 Microproxy now supports ftp |
 | 2010-04-30 What could get Data angry |
 | 2010-04-29 Lego Mindstorm solving the Rubik Cube |
 | 2010-04-28 Crossroads 2.65 is out |
 | 2010-04-17 Goggomobil in its natural habitat |
 | 2010-04-14 Bacon Time |
 | 2010-04-11 104 More friends to connect with |
 | 2010-04-10 Bacteria infested radio reporter |
 | 2010-04-07 The Kubat STAR |
 | 2010-03-30 Homework Essay |
 | 2010-03-29 C++ mutexes again |
 | 2010-03-20 Weird Eyechart |
 | 2010-03-15 Microproxy 1.01 |
 | 2010-03-05 Microproxy |
 | 2010-03-03 Sven Kramer and the wrong lane |
 | 2010-02-26 Endearing Babe Magnet |
 | 2010-02-17 Speed of light measured using chocolate and a microwave |
 | 2010-02-17 Never again expires after 65 years |
 | 2010-02-16 encfs on the Mac |
 | 2010-02-15 Hyves.nl and sexual predators |
 | 2010-02-10 Funny textbook |
 | 2010-02-09 DNS failing after sleep wake cycle |
 | 2010-02-06 Blast from the past |
 | 2010-01-28 Simple and straight Perl HTTP::Proxy |
 | 2010-01-15 Avatar the Movie |
 | 2010-01-08 Slightly NSFW Linux Ad |
 | 2010-01-07 WTF |
 | 2010-01-05 Stop Software Patents in the EU |
 | 2009-12-05 HammerServer 1.02 |
|
Whee! I've been quite recently playing around
with CouchDb. It's
really neat and nice: a networked, document-oriented database that
does simple things, and does them really well: storing stuff, and
letting you retrieve it. It doesn't provide referential integrity or
built-in types - everything is just a string - but it's scalable and
reliable, it's got built-in features for replication and what not. All
stored documents are even automatically versioned and revisioned, so
that data can always be rolled back to a previous state.
Interesting!
And if you happen to like Perl, the
module Net::CouchDb will make your life really a lot
easier. Great fun to play with.
So to get some practical fun with CouchDb, I incorporated it as a back
end type in my
tamper-resistant HammerServer. By
definition, there is more network overhead when storing data in
CouchDb, since all is sent over HTTP. So when compared to a flat file, or a
PostgreSQL database, CouchDb is slower. However, CouchDb's strenghts
should become apparent when the datastore is spread over many nodes
which balance-out inserting and retrieving records (which I didn't try
out).
So if you like, have a look
at HammerServer 1.02.
One of the fun things I found was the following. The HammerServer
relies on sequential ID's to store data. Each record at a given ID is
signed using its own data, but also using the signature of the
previous ID (which is ID-1). This is done to form a "chain" of secure
data, where deletions and insertions are detected. Therefore, the
correct working of the HammerServer relies on atomic increments of
some ID counter. Concurrent requests to insert new data may never wind
up using the same ID!
In RDBMS terms this is easily accomplished as something like:
select nextval(sequence)
or in Oracle terms,
select sequence.nextval from dual
The problem is that CouchDb is a document-oriented datastore, meaning
that making such an operation atomic isn't straight forward. In any
case, such an atomic update isn't built in. How can the
application enforce atomic operations in CouchDb, to avoid
concurrency-issues?
A too simplistic application-oriented attempt might be:
- Initially, create a document in the database, named
e.g. lastid, with a data element nr initialized to the
value 0.
- When getting the next value, the process might be:
- Fetch the document lastid from the database,
- Get its field nr,
- Increment it,
- Write it back to the database.
The problem here is of course that this approach isn't designed to be
atomic. Two concurrent requests might get the same ID, which would
break HammerServer's security.
Fortunately, CouchDB's revisioning comes to the aid. Internally
CouchDB also fetches revision data from the database when a document
is requested. When the same document is sent back for storage, the
revision data must still match - else, CouchDB issues an error stating
that there is a revision conflict. Brilliant! The right pseudo-code is
then:
- Fetch the document lastid from the database,
- Get its field nr,
- Increment it,
- Attempt to write it back to the database,
- If there is a conflict error, then redo all - starting at the
top.
- Otherwise we're done, the last value of nr is the
atomically incremented ID.
The HammerServer furthermore spices it up by retrying for no more than
60 seconds (just incase there's something really wrong at the back
end). As an example here is some Perl code that
uses Net::CouchDb to illustrate this. In the real HammerServer
sources the code is quite different and you can find it in the
method getnextid() of mod/db/CouchDB.pm. The below code
is only an illustration of the concept.
# Connect to the CouchDb server
my $connection = Net::CouchDb->new('http://hostname:5984')
or die("Can't connect to server\n");
# Connect to the database
my $cdb = $connection->db('database')
or die("Can't connect to the database\n");
# The next ID to determine:
my $nextid;
# Fetch the "lastid" document, increment the field "nr". If an update
# conflict occurs, then redo until we're out of time. Oh, and die if
# some other weird error occurs.
my $start_t = time();
my $try = 1;
while (1) {
# Fetch the counter document, increment number.
my $doc = $cdb->get('lastid');
$doc->nr += 1;
# Write it back to the database.
print("Try $try: incrementing to ", $doc->nr, "\n");
my $result = $cdb->post($doc)
or die("Failed to re-contact the CouchDB server\n");
if ($result->{ok}) {
# Data committed. We're done.
$nextid = $doc->id;
last;
} elsif ($result->{error} eq 'conflict') {
# Conflict error. Die miserably if this has been going on
# for 60 seconds or more.
die("Failed to increment lastid counter after 60 sec\n")
if (time() - $start_t > 60);
# The 1 min hasn't elapsed yet, let's retry.
print("Concurrency error, retrying\n");
next;
} else {
# Some other error.
die("CouchDB error: ", $res->{error}, ' ',
$res->{description}, "\n");
}
}
# All done.
print("ID succesfully and atomically incremented to $nextid\n");
Conclusion: Yep this works like a charm. I only had to fire up two
"trains" of insertion requests to detect logfile messages stating that
retries were necessary. All in all, this approach is perfectly
feasible. One might say: yech, that's a lot of work to just
emulate select nextval(sequence). True. On the other hand,
consider that CouchDb's philosophy is that tasks - and hence code and
complexity - should be "pushed out" towards calling applications, so
that the central database can remain lean and mean and fast. That
argument is as good as any.
|
|
|
 | 2009-11-28 Perls Automagical Autoloading |
 | 2009-10-07 Office Poster |
 | 2009-10-06 The nr 1 Nerdjoke |
 | 2009-10-04 WoW Startscript for my Mac |
 | 2009-09-27 HammerServer section is online |
 | 2009-09-26 The BING HQ |
 | 2009-09-26 Digging a WOW Tunnel |
 | 2009-06-29 Wee Todd |
 | 2009-06-23 The On Off Switch Revisited |
 | 2009-06-22 Meatspace |
 | 2009-05-30 My old houses |
 | 2009-05-11 LOLcats are funny |
 | 2009-05-11 Civic Duty WIN |
 | 2009-05-10 Vote for the baby, Sky Radio promo FAIL |
 | 2009-05-05 My secure data center |
 | 2009-02-15 My Valentine is sending me a dot exe |
 | 2009-02-05 MacPorts trash: .mp_123456 savefiles cleaning |
 | 2009-02-01 Truecrypt 6 on Linux and the ext3 filesystem |
 | 2009-01-28 www versus nl.youtube.com |
 | 2009-01-27 Songsmith and The Police |
 | 2009-01-25 My own Ministery of Silly Walks |
 | 2009-01-09 CoolIris Mini HOWTO |
 | 2008-11-04 UDP and DNS balancing |
 | 2008-11-02 Life in graphs |
 | 2008-11-01 Skeined yet? |
 | 2008-10-30 New Crossroads on the horizon |
 | 2008-10-28 Thread safe or not |
 | 2008-10-15 WOW patch 3 on a case sensitive MacOSX filesystem |
 | 2008-10-15 Surprising C++ optimizations |
 | 2008-10-14 Weird system message |
 | 2008-10-08 Data mining against terrorism does not work |
 | 2008-09-16 Crossroads at the top of Freshmeat.net |
 | 2008-09-09 Stupid spammers at Computable |
 | 2008-09-06 Spam prevention with Postfix and Postgrey |
 | 2008-09-03 The Gnomish Flying Machine |
 | 2008-08-27 Bank customer data on eBay |
 | 2008-08-26 Mutexes in C++ Threads |
 | 2008-08-22 4M dataloss in the UK last year |
 | 2008-08-21 Dropping spam with Postfix and Spamassassin |
 | 2008-08-18 Bayes and the War on Photography |
 | 2008-08-13 Good marital advice |
 | 2008-08-12 Squid proxy for personal usage |
 | 2008-08-11 Posix threads in C++ |
 | 2008-08-09 Crossroads mailing list |
 | 2008-08-08 Crossroads 2.00 is out |
 | 2008-08-01 Fail Pics |
 | 2008-07-14 The Fish Dance |
 | 2008-07-01 Big Bother and Massive Data Storage |
 | 2008-06-30 MMV One of omitted Unix tools |
 | 2008-06-08 Even anonymous breadcrumbs can give you away |
 | 2008-05-29 Crossroads in Argentina |
 | 2008-05-20 The Party at the Company Outing |
 | 2008-05-19 Crossroads 1.80 is out |
 | 2008-05-18 Where does technical innovation really come from |
 | 2008-05-16 Corporate bs generator |
 | 2008-05-15 Even the Vatican has to adapt |
 | 2008-05-12 Big Brother is watching your dog |
 | 2008-05-09 666 all over the place |
 | 2008-04-17 Security and privacy are incompatible |
 | 2008-04-16 The Hallmark E Card |
 | 2008-04-15 Crosroads Solaris port is out |
 | 2008-04-04 Identity theft can cost you dearly |
 | 2008-04-03 Crossroads can already do that |
 | 2008-03-31 A dagerous safari |
 | 2008-03-28 Why some Java J2EE projects are inefficient |
 | 2008-03-26 The Hummingbird |
 | 2008-03-25 The Easter delusion |
 | 2008-03-18 McAfee detects mass hack of 200.000 webpages |
 | 2008-03-17 More predictive statistics |
 | 2008-03-10 Backwards conclusions even on Slashdot |
 | 2008-02-18 A fractal photograph |
 | 2008-02-15 Kaprekar revisited |
 | 2008-02-14 Kaprekar numbers |
 | 2008-02-12 A tale of the criminal ineptitude |
 | 2008-02-10 Irritating Selfregistered users in PHPBB |
 | 2008-02-08 B2B Spam in the Netherlands |
 | 2008-02-06 Surprising iSight Capture |
 | 2008-02-05 Breadcrumbs at WickedLasers.com |
 | 2008-01-29 iSight Capture Utility |
 | 2008-01-28 The Male Brain |
 | 2008-01-26 Searching for the next Uri Geller |
 | 2008-01-24 Opt in for b2b spam |
 | 2008-01-14 Bokito Revisited |
 | 2008-01-13 Top Crossroads User |
 | 2008-01-12 World of Warcraft Dancing |
 | 2008-01-12 Justice dispensed better late than never |
 | 2008-01-11 Jeremy Clarkson and Identity Theft |
 | 2008-01-10 Terrorism in the Netherlands |
 | 2007-12-07 The mind and bodysnatchers are among us |
 | 2007-12-05 Bruce Schneier and Hildo |
 | 2007-12-04 Bye bye, good Christian soul |
 | 2007-12-03 Confusing mail message |
 | 2007-11-30 Medion MD 85276 reviewed |
 | 2007-11-29 Recent cases of data exposure |
 | 2007-11-20 Bayes bites |
 | 2007-11-19 Japan starts fingerprinting foreigners |
 | 2007-11-14 Privacy, Yahoo and the Strange World |
 | 2007-11-14 Privacy, Fall through algorithms, and Securing data |
 | 2007-11-07 European airlines to retain data |
 | 2007-11-03 BloggEd |
 | 2007-10-30 Wilders and Marktplaats.nl |
 | 2007-10-28 The goldplated Mac |
 | 2007-10-26 More morons |
 | 2007-10-26 Dilbert nails it again |
 | 2007-10-23 Rough yet funny |
 | 2007-10-05 Another silly Trojan mail |
 | 2007-10-01 So ugly it is beautiful |
 | 2007-09-28 Here is a nickel kid |
 | 2007-09-23 Spy Shredder |
 | 2007-08-29 Web svn view 1.08 |
 | 2007-08-24 Caught in THE Process |
 | 2007-08-21 Stupid Trojan attack |
 | 2007-08-21 Back in 1994 |
 | 2007-08-20 A girly iPod |
 | 2007-08-17 Crossroads for RDP connections |
 | 2007-08-15 Firewall art |
 | 2007-08-14 jpeginfo |
 | 2007-08-13 Good People |
 | 2007-08-07 The Real Crossroads |
 | 2007-07-30 BBC Documentaries in the Netherlands |
 | 2007-07-12 No problems with Crossroads so far |
 | 2007-07-11 Politically correct ad nauseam |
 | 2007-07-02 Waka Waka Poem |
 | 2007-07-02 Voyage of the rubber ducks |
 | 2007-06-28 The On Off Switch |
 | 2007-06-27 No free lunch |
 | 2007-06-25 Crossroads web interface |
 | 2007-06-25 Blinkenlights |
 | 2007-06-21 There is no silver bullet |
 | 2007-06-18 Motto of the week |
 | 2007-06-18 Do not feed the troll |
 | 2007-06-17 Which programming language are you |
 | 2007-06-13 Crossroads support request |
 | 2007-06-12 Bokito glasses |
 | 2007-06-07 Apache mod_proxy balancer description |
 | 2007-06-05 A ticketnumber is not support |
 | 2007-06-05 403 Hammertime |
 | 2007-06-04 Playground Fun |
 | 2007-05-24 Ascii man |
 | 2007-05-07 Cannot find the damn server |
 | 2007-05-02 The BFG200 |
 | 2007-04-27 Crossroads Top User |
 | 2007-03-30 Crossroads Usage |
 | 2007-03-25 The guy with the dark motorhelmet |
 | 2007-03-22 The Process and The Result |
 | 2007-03-21 Quotes attributed to Jos |
 | 2007-03-20 A really nice comment about Crossroads |
 | 2007-03-18 Kubat in the air |