"realtime" fs mirror application (backup, Python and Linux inotify)

Hello:

Recently I started an open source project "cutils" on the sourceforge:
http://sourceforge.net/projects/crablfs/

The document can be found at:
http://crablfs.sourceforge.net/#ru_data_man

This project's mirrord/fs_mirror tool is a near realtime file system
mirroring application across 2 or more hosts, something like MySQL's
replication, but it's for the file system especially with a great
amount
of small files, such as the php scripts and images of a website or the
(vitual) websites.

There are several ways to use this tool. The simplest is to mirror a
host's file system to another host for backup, and use the rotate
function(in the future version) or rotate scripts to get a daily or
hourly snapshot with the hard link.

Or futhur more, you can use it this way:
This graph should be displayed with monospaced fonts:

+----------+
| worker | -[mirrord] -----------\
+----------+ |
...... |
|
+----------+ |
| worker | -[mirrord] -----------\
+----------+ |
V
[fs_mirror]
|
+----------+ +----------+
| worker | -[mirrord] ---> | backup |
+----------+ +----------+
| |
[take_over] |
| |
V |
+----------+ |
| rescue | <------------------- NFS
+----------+

This is the multi to one backup, which is cost efficient. If one of
the
worker hosts fails, you can subsitute the failed worker with the
rescue
host, with the aid of any high available method, such as heartbeat
project. By this way, you can use 1 or 2 hosts to support the HA of
more
than 3 servers.

Or you can also use it as an IDS(Intrusion Detection System) like a
realtime "tripware", or you can make a mirror chain that a host B
mirrors
from A and be mirrored by C, etc ... I will also try to research a way
to use it as a distributed implemetation with one write and multi-read
model.

mirrord/fs_mirror makes use of inotify, which is a function afforded
by
the recent Linux (from 2.6.12). It is a counterpart of FAM, since
Linux
FAM has stopped so long.

Now it works for me, on a RHEL4 system and the LFS 6.2, I hope this
tool
can be useful to you too.

Thanks.

Now I meet a strange problem.

After the first sync init, it enters to the realtime replication
state. I deployed them on 3 machines, and have run near half month.
Suddenly one day, a host, I don't know what's wrong, I found fs_mirror
get the empty
records from its mirrord agent. In normal conditions, these records
should
be:
"CREATE:/var/www/html"
"FWRITE:/var/www/html/index.php"
"DELETE:/var/www/html/temp"
"MOVE'/var/www/html/aa', '/var/www/html/bb')"
....
But should be no empty records. This lead to fs_mirror to a dead
infinite loop.

I restart the fs_mirror from the broken point, but the problem
remains, after DEBUG I found the problem occurs at the same serial
number every time(I use Berkeley DB as the log record(wmLog), and
serial numbers are the keys), so I suspect that the problem is BDB,
but I don't know how to test
and locate to the right place.

I tried to open the orignal db file in Python:
>>> import bsddb
>>> x = bsddb.btopen("/var/mirrord/wmlog"
>>> len(x)
623748
>>> x["6854"]
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/usr/lib/python2.5/bsddb/__init__.py", line 223, in
__getitem__
return _DeadlockWrap(lambda: self.db[key]) # self.db[key]
File "/usr/lib/python2.5/bsddb/dbutils.py", line 62, in DeadlockWrap
return function(*_args, **_kwargs)
File "/usr/lib/python2.5/bsddb/__init__.py", line 223, in <lambda>
return _DeadlockWrap(lambda: self.db[key]) # self.db[key]
KeyError: '6854'
>>> x[str(6854)]
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/usr/lib/python2.5/bsddb/__init__.py", line 223, in
__getitem__
return _DeadlockWrap(lambda: self.db[key]) # self.db[key]
File "/usr/lib/python2.5/bsddb/dbutils.py", line 62, in DeadlockWrap
return function(*_args, **_kwargs)
File "/usr/lib/python2.5/bsddb/__init__.py", line 223, in <lambda>
return _DeadlockWrap(lambda: self.db[key]) # self.db[key]
KeyError: '6854'
>>> x.first()
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/usr/lib/python2.5/bsddb/__init__.py", line 278, in first
rv = _DeadlockWrap(self.dbc.first)
File "/usr/lib/python2.5/bsddb/dbutils.py", line 62, in DeadlockWrap
return function(*_args, **_kwargs)
_bsddb.DBNotFoundError: (-30990, 'DB_NOTFOUND: No matching key/data
pair found

Even I have stopped the mirrord daemon, the errors remain.

Then I tried to copy and move out the database file, and open the new
dbfile:
>>> import bsddb
>>> x = bsddb.btopen("/tmp/wmlog")
>>> len(x)
0
the length is 0, and getitem get the same errors above.

Why this occurs when I copy the db file? Especially the len() is 0?!

I can only restart the mirrord, to rebuild the BDB data file, and so
far,
this problem does not occurs again.

I don't know why there is a occasional problem like this? Is there any
one be familiar with BDB can give me several advices?

Thanks.