Lesson learned – Array.insert should probably be named Array.insert! – it changes the underlying array, and sometimes that’s not what you mean to do.

http://dev.rubyonrails.org/ticket/8383

Now we’ll see if it gets committed.

What we really want to do is use EC2 – look at the gigavox success story on amazon’s site – we need almost the same things, and want to implement almost the same arch – except using rails and activemessaging.

The problem is the EC2 closed beta – ugh – we’ve been waiting, oh so impatiently. In the meantime everyone shuold go get their feet wet with RightScale. You get 10 hours to play with, even without your own EC2 keys.

At PRX we wanted to be able to upload/resize/attach images and files, have a preview mode to see what the home page would look like before we publish parts of the content, and also we wanted to be able to access session info so we could make some content vary based on info in a user’s session (this is also part of how the ‘preview’ mode works).

Folks on the forums also seem to be trying to figure this kind of thing out, so I humbly offer what I have gotten to work.

So without exposing all the work I am cleaning up on extending comatose, here is a bit of code showing how to add the session info to the comatose pages so it can be accessed and used.

As for the rest of my enhancements – well, I need to make it into a plugin I think, that gets installed on top of comatose. That will have to wait, as I am too busy adding enhancements at the moment to stop and package it….

Have a look, there’s also a very short prez I gave at the IMA 2007 Conference showing how we use S3, ActiveMq, Rails, ActiveMessaging and other junk like that to make our systems hum.

Here it is again:

Also, I should mention, I try to avoid installing anything using fink or mac ports. Not because there is anything wrong with them per se, but I like control over what I install and where it goes, and as 9 times out of 10 I am going to be installing the same thing on at least 1 other OS, I want to do it from scratch. I am probably insane.

When I went to search for some easy instructions for installing php5 and apache2, best thing I could find was here. Wasn’t bad, just change the instructions to prefix under “/usr/local/”, and all went well.

Then I hit the PHP5 make, and it died, the error turns out to be pretty well known, and is related to PHP5 only compiling against MySql 4, whereas I live in this century and run MySql 5. Apparently there is a bug report for this. A quick install of mysql 4 from the tar.gz package, and updating the softlink for /usr/local/mysql, and everything compiled. Flip the soft link back to MySql 5, and so far everything runs fine.

So once again, with a bit more work, compiling from src triumphs!

• When dragging the end of a control, like a Text Input, to be the input to another form value, like in a URLBuilder, the drag and drop is very imprecise. In a list of three Query Parameters, I could never drop onto the middle element, it just wasn’t able to only select that one middle parameter
• Add a For Each: replace that does an element in a feed item, rather than the whole element. Or adds to it, hard to actually combine feedw without something like this
• Have a smarter Union with options like interleaving rather than just putting all of one feed after another
• The URLBuilder does not put the arguments into the query string in the order you have them in the form
• While I love the debugger, there is no way to immediately run a pipe, you can only see debugger output, I want a way to preview the run
• The output from a given source can only go to one widget – it would be great if the output of a fetch could be used for different processing, then recombined. I did this, but needed 2 almost identical Fetch controls to make it work, and adds another request to the processing time
• Save is buggy – I have made changes to the URLBuilder, but Save would be disabled till I changed a different control. I also had once where I had saved some changes, then gone back to edit after running, and the changes were goe…
• Consistent behavior of the fetch/content analysis/for each replace. I see times when I run it that no results come back, other times they work great. I have often seen that running a Content Analysis operation would flake out and show no results, even though the fetch above it would have results when I debug it. This same inconsistency seems possibly related to the general inconsistency of results mentioned above.
• Also saw another bug after adding a For Each where I could not add the sub source to the For Each. Had to save, close up, and come back in to get it to work.

So all in all a good toy, but not ready for anything real – but I see the potential. Really the UI is unbelievable, and while a bit buggy still, I do not know how they made it such a rich UI – the pipes alone are an amazing bit of work.
Having used lots of GUI based tools like Informatica Powermart or the Weblogic Integration tools, or even just Rational Rose or Visio, I see where something like this is not far from being a very competitve interface, and incredibly light weight and responsive.

So we switched to Contegix not that long ago for prx, and I have to say, these guys are a class act. It’s not that our last hosting provider was bad – they weren’t – it’s just that Contegix is that much better.

The support just may be too good. They are spoiling me; what will I do if I have to deal with a different host?

I fear I may never get better at any sysadmin tasks, it is all too easy to simply write a request, and in moments you have a response, and in minutes you have it done. I have never had better support, even when I had a dedicated team of admins around the corner. And I’d say about half the time I ask for something, the response includes analysis and suggestions on how to do it better.

So this week alone, I can’t say enough about the strong work from Craig, Eric, Joe and Brian – thanks again guys!

Works well – we are streaming large audio files over it, so we needed something better than the ruby s3 bindings amazon released (though they are very clear in their code that they do not handle big files well).

Only hitch was setting metadata when creating an AWS::S3::S3Object – couldn’t get it to work.
Tried creating a 1 byte file to start, then updating with metadata, that was fine, but the initial create was erroring out, or the metadata would not take.

Luckily, there is a patch offered by ‘Lars’ which makes it all work for me now.

I posted this to the list, but here is some sample working code for setting metadata on create (once the patch is applied):

1) Using the store class method:

AWS::S3::S3Object.store(
'myfile-81805',
open('/data/production/amazon/myfolder/myfile-81805'),
'my-s3-bucket',
:content_type=>'audio/mpeg',
'x-amz-meta-my-file-name'=>'my.mp3')


2) Using the Bucket new_object method, and then store, you don’t have to include the metadata prefix ‘x-amz-meta-’:

my_bucket = AWS::S3::Bucket.find('my-s3-bucket')
s3o = my_bucket.new_object
s3o.key = 'myfile-81805'
s3o.value = open('/data/production/amazon/myfolder/myfile-81805')
s3o.content_type = 'audio/mpeg'
s3o.metadata['my-file-name'] = 'my.mp3'
s3o.store


So with that working, back to ActiveMessaging, and getting changes fully tested and checked in.

This should make the changes to a13g cleaner, and we can now take advantage of having the reconnect wait time configurable.
All in all, it makes a13g closer to implementing a ‘reliable’ listening service.

There is one sorta ugly bit, the __old_receive method was a fine way to do this when I was extending the Connection class, but could use refactoring…after I get a good check in of fixes to a13g, I’ll submit something prettier. But as we used to say back in Stanley Eisenstat’s classes (which I wish I had taken more of): good is great, but done is better.