Lulu Poetry

Lulu Poetry or Poetry.com, announced on May 1, 2011 that they would close four days later on May 4, deleting all 14 million poems. Archive Team members instantly amassed to find out how to help and aim their LOIC's at it. (By the way, I actually mean their crawlers, not DDoS cannons.)

Site Structure

The urls appear to be flexible and sequential:
(12:13:09 AM) closure: http://www.poetry.com/poems/archiveteam-bitches/3535201/ , heh, look at that, you can just put in any number you like I think
(12:15:16 AM) closure: http://www.poetry.com/user/allofthem/7936443/ same for the users
There are apparently over 14 million poems. The numbers go up to http://www.poetry.com/user/whatever/14712220, though interspersed are urls without poems (author deletions?).

Howto

1. Claim a range of numbers below.
2. Generate a hotlist of urls for wget to download by running this, editing in your start and end number: perl -le 'print "http://www.poetry.com/poems/archiveteam/$_/" for 1000000..2000000' > hotlist
3. Split the hotlist into 100 sublists: split hotlist
4. Run wget on each sublist: wget -x i xaa
5. To avoid getting too many files in one directory, which some filesystems will choke on, recommend moving into a new subdirectory before running each wget on the sublist.
6. For the daring, here's how to run all wgets on all the sublists in parallel, in subdirs: for x in ???; do mkdir $x.dir; cd $x.dir; wget -x -i ../$x &; cd ..; done

Important note: Everyone's getting a lot of 500 errors, probably because we're whacking their server. Because of this, make sure you keep all your log files. Then you can search them later to generate a list of urls to retry. Suggested ways to do this:

• run the command grep -h -B1 "ERROR 500" *.log | grep ^http | sed 's/:$//'
• use a perl script to search your download directory for missing folders in the sequence

Recommended wget command: wget -E -k -T 8 -o logfile.log -nv -nc -x -i urls.txt

wget Options Translation
short	long version	meaning
-E	--adjust-extension	adds ".html" to files that are html but didn't originally end in .html
-k	--convert-links	change links in html files to point to the local versions of the resources
-T	--timeout=	if it gets hung for this long (in seconds), it'll retry instead of sitting waiting
-o	--output-file	use the following filename as a log file instead of printing to screen
-nv	--no-verbose	don't write every little thing to the log file
-nc	--no-clobber	if a file is already present on disk, skip it instead of re-downloading it
-x	--force-directories	force it to create a hierarchy of directories mirroring the hierarchy in the url structure
-i	--input-file	use the following filename as a source of urls to download

Coordination

Who is handling which chunks of urls?
IRC name	starting number	ending number	Progress
closure	0	200,000	complete
closure	200,000	999,999	in progress
jag	1,000,000	2,000,000	in progress
notakp	2,000,000	3,000,000	in progress
no2pencil	3,000,000	3,999,999	in progress
d8uv	4,000,000	4,499,999	in progress
Qwerty01	4,500,000	4,999,999	getting started
Awsm	???	???	in progress?
underscor	???	???	in progress?
BlueMax	???	???	in progress?
SketchCow	???	???	in progress?
[yourusernamehere]	x,000,000	x,999,999	in progress
[seriouslyeditme]	x,000,000	x,999,999	in progress

Correction

For server maintenance issues, I created the following correction script :

flist=`grep "currently performing site maintenance" *.html | cut -d: -f1`

x=0 for file in ${flist}; do

 if [ -f ${file} ];
 then
   echo correcting ${file}
   wget -E http://www.poetry.com/poems/archiveteam/${x}/ -O poem${x}.html 2>/dev/null
   echo done...
   x=`expr ${x} + 1`
 fi

done

if [ ${x} -eq 0 ]; then

 echo Directory clean

else

 echo ${x} files corrected

fi