(01:37:22 AM) ccustine: BTW, I started using JUnit 4.4 last week and
it works fine with our builds...  but I only played with it very
briefly
(01:37:28 AM) elecharny: oh
(01:37:43 AM) ccustine: but we really have to restructure the abstract
test classes
(01:37:45 AM) elecharny: I would be very interested if you can commit
such a test in a sandbox
(01:38:12 AM) ccustine: what I have not tried is using the annotations
in the abstract classes
(01:38:19 AM) elecharny: I was also thinking of using some kind of
scripted tests
(01:38:21 AM) ccustine: to get consistent behavior
(01:38:24 AM) elecharny: yes
(01:38:26 AM) ccustine: sure...
(01:38:28 AM) akarasulu: ccustine: yeah that would be broken up into
@Before and @After classes for pre and post
(01:38:49 AM) akarasulu: elecharny: I have an idea on this stuff btw
(01:38:54 AM) elecharny: akarasulu: I think we can use @Before and
@Afterclass on the main class
(01:38:56 AM) akarasulu: about reverting to previous state
(01:39:23 AM) elecharny: the only way to be sure you revert correctly
is to do it by hand
(01:39:48 AM) elecharny: or to remove the added elements and nd to
injects the new one
(01:40:09 AM) elecharny: ditching the tested partition may work too
(01:40:31 AM) elecharny: for instance, if all tests are done in a
dc=test, dc=org partition,
(01:40:52 AM) elecharny: then you can set it up on startup and delete
it on teardown
(01:41:21 AM) elecharny: (but for @BefoeTest and @AfterTest, not
during a @BeforeClass and @AfterClass)
(01:41:59 AM) akarasulu: elecharny: the biggest cost is in setting up
a partition
(01:42:08 AM) akarasulu: the record manager grabbing extents is the big problem
(01:42:45 AM) akarasulu: You need to generate an ANTI-LDIF using the
changelog interceptor.  Then play back the ANTI-LDIF.
(01:42:59 AM) akarasulu: that fixes the problem
(01:43:21 AM) akarasulu: minus a few minor details hehe
(01:43:27 AM) akarasulu: but they will be ok
(01:43:38 AM) akarasulu: (talking about timestamps)
(01:44:09 AM) akarasulu: but tests don't depend on this and timestamps
are updated anyways when creating new partitions and filling them
(01:44:33 AM) akarasulu: meaning it's not a full revert as a
transaction rollback would have but it will do the job for us
(01:46:39 AM) akarasulu: elecharny: still there?
(01:46:56 AM) elecharny: yep
(01:47:20 AM) elecharny: just an idea :
(01:47:43 AM) elecharny: what about simply removing all the entries
starting at a specific date creation ?
(01:47:50 AM) elecharny: simple, easy, and fast
(01:47:51 AM) akarasulu: hmmmm
(01:48:01 AM) elecharny: whenyou start a test, you inject some entries,
(01:48:07 AM) akarasulu: that could work to yeah
(01:48:14 AM) elecharny: and then at the end, you remove all the
crated entries based on the creation Date
(01:48:21 AM) akarasulu: so you have S0 (server started state)
(01:48:25 AM) elecharny: no complex anti-ldif to compute
(01:48:29 AM) akarasulu: you take the zulu time there
(01:48:42 AM) akarasulu: then delete everything created after that
(01:48:46 AM) akarasulu: hmm there are problems
(01:48:48 AM) elecharny: you simply create a fake entry, and lookup
for it, asking for its creation date
(01:48:59 AM) akarasulu: what if your test modifies existing things
(01:49:03 AM) akarasulu: like admin user
(01:49:12 AM) akarasulu: admin changes their password for example
(01:49:16 AM) elecharny: for existing partitions, like the ou=system,
(01:49:27 AM) elecharny: this is something we won't test this way

SNIP ...

(01:55:41 AM) akarasulu: however the approach I talk about is wholistic
(01:55:48 AM) akarasulu: and really works across partitions
(01:55:52 AM) elecharny: you mean the anti-ldif ?
(01:55:54 AM) akarasulu: and can be used for many other things
(01:55:56 AM) akarasulu: yeah
(01:56:11 AM) elecharny: yes, but this is pretty complex
(01:56:16 AM) akarasulu: it's not that hard to implement really
(01:56:22 AM) akarasulu: nah man not really
(01:56:32 AM) akarasulu: ersin did this before at some point
(01:56:32 AM) elecharny: for add, no
(01:56:37 AM) akarasulu: subversion does this to rollback
(01:56:38 AM) elecharny: but for delete for instance ?
(01:56:52 AM) akarasulu: did you read my email or just glance through it
(01:56:56 AM) akarasulu: the rules are simple
(01:56:59 AM) elecharny: I'm commenting it
(01:57:12 AM) akarasulu: Delete => Add (anti-op)
(01:57:30 AM) akarasulu: this can be used for so many things down the line
(01:57:42 AM) elecharny: yes, but then you have to read the data
before deleting it, otherwise you cant' add it
(01:57:53 AM) elecharny: del use a single DN
(01:57:55 AM) akarasulu: interceptor has this info
(01:58:19 AM) akarasulu: when you do a delete you most of the time
have a handle on the object to be deletede
(01:58:21 AM) elecharny: but if you add an interceptor, you are
modifying the server behavior
(01:58:22 AM) akarasulu: when you do a delete you most of the time
have a handle on the object to be deleted
(01:58:44 AM) elecharny: del just need a DN
(01:58:47 AM) akarasulu: ok heisenberg
(01:59:53 AM) akarasulu: this is really not that hard at all to do and
ersin has wrote such an interceptor already here ...
(02:00:16 AM) akarasulu:
https://svn.safehaus.org/repos/sandbox/workshop/interceptor/src/main/java/workshop/interceptor/ChangeLogInterceptor.java
(02:00:22 AM) akarasulu: it works really well
(02:00:36 AM) akarasulu: this can be used with tooling to control
snapshotting in the server
(02:00:47 AM) akarasulu: just think oops you just deleted a bunch of *#(@&^@%
(02:00:53 AM) akarasulu: how do you roll it back
(02:00:58 AM) akarasulu: this is after a transaction
(02:01:06 AM) akarasulu: (if we even had transactions)
(02:01:28 AM) akarasulu: then you &#^@(*$& your pants and think you just
lost 4 hrs to restore from backups since you did not use mitosis
(02:01:30 AM) akarasulu: :D
(02:01:33 AM) elecharny: man, just have a look at the Del ope in this
interceptor
(02:01:35 AM) akarasulu: but this (*#$&@^&@ can save you
(02:01:47 AM) elecharny: you have _no clue_ on the attributes which
have been deleted
(02:01:54 AM) elecharny: haw possibly can you restore them then ?
(02:02:01 AM) akarasulu: del op deletes and entry
(02:02:08 AM) akarasulu: elecharny: you can do it man
(02:02:12 AM) elecharny: it delete an entry using its DN
(02:02:19 AM) akarasulu: I'm telling you this is not impossible and
not hard to do
(02:02:20 AM) elecharny: if you want to restore the entry,
(02:02:28 AM) elecharny: you will have to read all of it's attribute first
(02:02:35 AM) akarasulu: you can do this
(02:02:37 AM) elecharny: yes, we can do it
(02:02:56 AM) akarasulu: but not hard at all
(02:03:19 AM) elecharny: but I don't like at all the idea to inject an
interceptor to test the server
(02:03:20 AM) akarasulu: really if you gave it a try you can have it
knocked out in less than 2 days
(02:03:35 AM) akarasulu: well this interceptor will be optional even
in production time
(02:03:46 AM) akarasulu: and it can track your log of requests to the server
(02:03:52 AM) akarasulu: this was what you were looking for
(02:03:55 AM) elecharny: for modifyDN, it will be complex
(02:04:07 AM) akarasulu: nah just switch src and target
(02:04:10 AM) akarasulu: in anti op
(02:04:15 AM) elecharny: transaction would save our *&@^#$^@ ... ;)
(02:04:39 AM) akarasulu: hmm for this little case it can yes but
however it will not save users who commit incorrectly ;)
(02:04:47 AM) akarasulu: you still need this rollback capability
(02:04:53 AM) akarasulu: this snapshoting
(02:05:03 AM) akarasulu: this is how subversion does it
(02:05:13 AM) elecharny: transcations without rollback are useless, true
(02:05:16 AM) akarasulu: it's all a series of diff operations
(02:05:28 AM) akarasulu: you apply anti-diff to get back to earlier state
(02:05:33 AM) akarasulu: but you go forward
(02:05:35 AM) akarasulu: in txn
(02:05:47 AM) akarasulu: but even with txn your ($*&Q&^@ after you commit
(02:06:00 AM) akarasulu: if you commit wrong )(*#&^@^ as a user you still
deleted #&@^#$*@& you cannot get back
(02:06:10 AM) akarasulu: and with replication you just replicated it
to the rest of the cluster
(02:06:20 AM) akarasulu: so good luck
(02:06:51 AM) elecharny: it can work
(02:06:53 AM) akarasulu: this feature can be done fast and can be used
to kill 2 big birds with one stone
(02:07:01 AM) akarasulu: think about studio with a plugin for ADS
(02:07:02 AM) elecharny: bt we should not do complex things
(02:07:11 AM) akarasulu: this is not that complex
(02:07:14 AM) akarasulu: search is complex
(02:07:21 AM) akarasulu: this is simple with clear rules
(02:07:22 AM) elecharny: like, if we mod an entry, to rollback, we
just restore the old entry
(02:07:40 AM) elecharny: if we del an entry, we rollback by restoring
the previous entry
(02:07:40 AM) akarasulu: nah we do anti-op know why?
(02:07:49 AM) akarasulu: this also serves as a recording mechanism
(02:08:03 AM) akarasulu: if you just resusitate instead of anti
operation you don't capture the full log
(02:08:17 AM) elecharny: you want to have this rollback interceptor, then
(02:08:19 AM) akarasulu: you can roll forward and back and forward and
back to your hearts content
(02:08:22 AM) elecharny: and use it for test ?
(02:08:26 AM) akarasulu: for now
(02:08:30 AM) akarasulu: but have it in the server
(02:08:45 AM) elecharny: ok, I buy your idea
(02:08:54 AM) elecharny: becaus eit won't be only for tests
(02:09:04 AM) akarasulu: exactly
(02:09:05 AM) elecharny: but can be used for someting else
(02:09:11 AM) akarasulu: it will actually help for 3 things
(02:09:21 AM) akarasulu: (1) test speed boost (minimal value)
(02:09:35 AM) akarasulu: (2) recording and playback for complex test cases
(02:09:53 AM) akarasulu: (3) for snapshoting and rolling back the
server to previous states
(02:10:03 AM) elecharny: I would add a 4 case :
(02:10:06 AM) akarasulu: oh?
(02:10:21 AM) elecharny: 4) To create journals to be used if the server crash
(02:10:38 AM) akarasulu: ahhh yeah sure totally can be used for that
(02:10:46 AM) elecharny: that is very importantn
(02:10:48 AM) akarasulu: I'd inject some meta info for that
(02:10:58 AM) akarasulu: also we can encode snapshots in there
(02:10:58 AM) elecharny: as we only synch data on disk every 15
seconds (default value)
(02:11:05 AM) akarasulu: even expose this data as a special partition
(02:11:11 AM) elecharny: hmmm
(02:11:20 AM) elecharny: that would be costly
(02:11:25 AM) elecharny: I would prefer a text file
(02:11:37 AM) akarasulu: oh was talking about snapshots
(02:11:46 AM) elecharny: ok
(02:11:48 AM) akarasulu: like gimme a snapshot
(02:11:59 AM) elecharny: let's talk about those ideas on the ML
(02:11:59 AM) akarasulu: but in ldif we can encode with special comment
(02:12:07 AM) elecharny: there are reallly interesting for the community
(02:12:14 AM) akarasulu: # SNAPSHOT: ( id, date )
(02:12:26 AM) akarasulu: yeah this is cool 
(02:12:36 AM) akarasulu: been dying to do this stuff for years
(02:12:42 AM) akarasulu: thought ersin would do them and he came close
(02:12:48 AM) akarasulu: then we moved on as usual
(02:12:52 AM) akarasulu: but now we have serveral drivers
(02:12:56 AM) akarasulu: several
(02:13:02 AM) akarasulu: so it may actually materialize
(02:13:47 AM) akarasulu: hey btw we can even have a special extended
op that sends a gzipped version of an anti-log or a segment of it back
to studio to be looked at
(02:13:56 AM) elecharny: wiki !
(02:13:58 AM) elecharny: wiki !
(02:13:58 AM) akarasulu: perhaps we can expose a read only partition
to show this info as well
(02:14:00 AM) elecharny: wiki !
(02:14:06 AM) akarasulu: aye you got the text :D