version2beta archive

Ramada Plaza Milwaukee kicks my 70 year old mother out at midnight

2012-05-20T02:21:00Z

Dear Ramada Worldwide,

Your Wisconsin Ramada Plaza Hotel on 13th at College Ave in Milwaukee just kicked my 70 year old mother out of her hotel room at midnight, because I complained about accessibility.

I know that sounds absurd. It is absurd, and it actually gets worse. You see, what I actually complained about was that the hotel told my sisters to cart my mother to their room on a luggage rack. You know, the rack the bellboy pushes.

Mom has had bypass surgeries, back surgeries, neck surgeries, heart surgeries, you name it. Last month they did an operation on her carotid artery. A month before they replaced her pacemaker. She also has primary progressive MS. Mom is a modern medical miracle. One would think that when you arrive with a 70 year old woman and ask for a wheel chair, you might recognize the kind of situation that requires a level head for both customer service and legal compliance. You might not want to tell the guests to load mom up on a luggage cart and skate her up to her room.

Upstairs, and all the way to the other end of the building. To a room where, when one sits on the foot of the bed, the head of the bed rises 12 inches. (Try to picture my mother sitting on the edge of the bed putting on her shoes, with the other end of the bed floating away. It’s okay if you need to laugh.)

My younger sister didn’t just tell me about it, she called up Ramada and complained. Nothing happened.

Trying to do something about this worried my older sister - with good reason, as it turned out. After all, look what happened when I did try to do something about it. They got kicked right out on their asses at midnight. Duh!

My mother didn’t want to be a bother. To be honest, I think she feels bad she can no longer walk from the lobby to room 256.

On the other hand, I was willing to be a bother. Here’s what I did:

I tweeted about it.¹ ² ³ ⁴ ⁵ Really, this is some absurd stuff. Push the old lady up on a luggage rack? It’s almost funny, you know? I mean, if she’s not your mom.
I called Ramada Worldwide customer service and told them about it. They told me to call the hotel. I said I would, but I still wanted to talk to a manager. Turns out, Ramada Worldwide customer service is one of those departments that is set up to prevent the customer from getting service. I had to tell the woman to (a) get a manager, or (b) take my information so that a manager could get back to me, or (c) hang up on me. She chose option (b) and promised a manager would return my call by Thursday, five days hence.
I called the Ramada Plaza Milwaukee, where they were staying. Scratch that. I tried to call the hotel, but THEIR NUMBER WAS DISCONNECTED. Yes, the 800-303-8002 number on every page of their website⁶ was disconnected. Wtf?
I drove down to the hotel and I talked to a manager, Jennifer, who was actually very sweet and helpful. She was aghast about the luggage rack. Jennifer’s indignation made me feel good. She also found an accessible room on the first floor. She gave me keys so I could move my mom and my sisters (who were visiting with my son at his hippotherapy) to the new room. As it happens, she couldn’t find one of the aforementioned luggage racks so I had to make five trips. But she did take some notes on my concerns (like the see-sawing bed and the out-of-service phone number) to leave for the owner.

The owner didn’t call me. When my sisters returned at midnight, their room key didn’t work. When they checked in at the counter, my sisters were informed that they were no longer guests and would be escorted out, with their things.

Here’s the really fun part - mom wasn’t even there because she was at the hospital for a heart attack! My sisters were returning to the hotel room after we got mom admitted to St Luke’s Hospital. I’m sitting in mom’s hospital room where it’s still not clear if she had a heart attack or if it’s just a stage one blockage, and my sisters - after one hell of an evening - are getting kicked out of their hotel.

The reason they were given is that I was combative.

Some people might think I was being reasonable to ask that they help find a more appropriate room. But some people might call it reasonable that the hotel didn’t have any wheelchairs and didn’t offer help beyond a luggage rack for her transport. My mother could have chosen to travel with her electric scooter, except that she wouldn’t have been able to fly here from Maine. Or maybe my sister should have reserved a handicapped room - I don’t actually know whether she reserved a handicap room, or if she even knew they exist. If she didn’t reserve an accessible room, why should that be the hotel’s problem?

(Except for the ADA. And Wisconsin law. I don’t know if the city has any laws on the subject.)

Just to be clear: I was not there when my sisters wheeled our mother up to the room on a luggage rack, but all three of them told me it happened, that it was what the hotel told them to do and they had no other way to get mom to the room. I did sit on the seesawing bed. I did talk to Ramada Worldwide customer service. And I did talk to Jennifer at the hotel, who is the person who answered my questions when I asked for a manager.

My conclusion can only be that my elderly mother and two sisters were kicked out of the Ramada Plaza Milwaukee because I complained about accessibility. To be fair, what I call complaint they call combat. And I did tweet about them five times. So maybe I deserved to feel like a total shit when my sister called me crying at half past midnight to tell me they kicked them out of their room because I was combative.

I fear many people don’t actually understand what accessibility is, and why it’s important. Some of them work in (or even own) businesses that deal with accessibility challenges on this scale every day, and often they are handled poorly. I encourage you to share this story for the value of a bad example, and reinforce why accessibility is important.

A Response to Tom Henrich's Document The Why

2012-03-26T10:15:00Z

On his blog, Tom Henrich¹ calls for better developer documentation² on why a design decision was made, rather than just what decision was made. In this post I share my example.

From the source code for a Box.net³ uploader snippet I wrote in MODX:⁴

/*
 * @name      Down and Dirty Box.net Uploader
 * @author    Rob Martin <rob@qmuxs.com>
 * @copyright (c) 2011 Quintessential Mischief LLC
 * @link      http://www.qmuxs.com
 * @license   GPLv3
 * @version   0.1.0
 * 
 * The need for a down & dirty uploader came about after reviewing the 
 * box.net PHP library code at https://github.com/boxplatform/box-php.
 * It's just not very good, and some of it is totally borked. It's not
 * even a surprise - I found comments on blog posts dated from February
 * 2008 complaining about a bug (can you call a whole missing function
 * - promised in the docs - a bug?) that has not changed in over three
 * years.
 *
 * I was about to start over. :-( But I did get in touch with Sean ****
 * at Box.net who threw some salt on the open wounds. He let me know
 * that the new PHP library is **just days from release**. 
 *
 * And no, he couldn't get it for me. So sorry.
 *
 * The Python code isn't much better, though I haven't read through it
 * as much. However, documentation such as:
 *
 * > """Exception class for errors received from Facebook."""
 * 
 * makes me think it might not be. Facebook has nothing to do with this
 * code.
 *
 * **NOTE**: Just so you know, this code doesn't do any of the
 * authentication work. You've gotta do that manually. You need an API key,
 * an auth_token, and a folder ID for where you want the files to land.
 *
 * Here's how:
 *
 * 1. Get a ticket. It'll come back as <ticket>...</ticket>.
 * https://www.box.net/api/1.0/rest?action=get_ticket&api_key={YOUR_API_KEY}
 *
 * 2. Using the ticket, log in.
 * https://www.box.net/api/1.0/auth/{YOUR_TICKET}
 *
 * 3. Request the auth_token. It'll come back as <auth_token>...</auth_token>.
 * https://www.box.net/api/1.0/rest?action=get_auth_token&api_key={YOUR_API_KEY}&ticket={YOUR_TICKET}
 *
 * 4. Get the ID of the folder you want. It'll look like <folder id="xxx" name="this-one".../>
 * https://www.box.net/api/1.0/rest?action=get_account_tree&api_key={YOUR_API_KEY}&auth_token={YOUR_AUTH_TOKEN}&folder_id=0&params[]=nozip
 * 
 * So that's the story. Here's the code:
 */

MODX can beat up your content management system

2011-12-13T10:48:00Z

Tonight’s Milwaukee Web Design Meetup featured a CMS MegaPanel - eight pros in eight content management systems, head to head on development, features, ecosystems, costs. I represented for MODX.

Here is my Prezi¹ on MODX:²

Sorry, Flash is required here. I don’t like it either, and I know I really need to learn a different presentation system.

MODX can beat up your Content Management System. on Prezi

A quick note on MongoDB replica sets

2011-11-29T11:49:00Z

MongoDB is replying to a replica-set add command with “Need most members up to reconfigure, not ok”. The problem is with my configuration.

Here’s the error:

PRIMARY> rs.add("nosql1-private:27017")
{
    "assertion" : "need most members up to reconfigure, not ok : nosql1-private:27017",
    "assertionCode" : 13144,
    "errmsg" : "db assertion failure",
    "ok" : 0
}

The problem is in my configuration on the slave Mongod. Note that I’m using a configuration file, rather than command line options, which is how the tutorials are written. This fits in with my init file, which I can share if anyone wants an Ubuntu init for MongoDB 2.0.2.

fork = true
bind_ip = nosql1-private
port = 27017 # default port
logpath = /var/log/mongodb/mongod.log
logappend = true
cpu = true # log CPU activity periodically
journal = true
auth = true
replSet = gedc

I disabled journal (shouldn’t be needed on a slave, right?) and auth, which seems to actually be the problem. After, the rs.add() works fine.

PRIMARY> rs.add("nosql1-private:27017")
{ "ok" : 1 }

I’m not sure I don’t have to go back and get auth reenabled, though, because now my status says that the second server is “still initializing” and “(not reachable/healthy)”. But when I try to add authentication credentials, I get this error:

uncaught exception: error { "$err" : "not master and slaveok=false", "code" : 13435 }

So, I did discover what I really needed was a keyFile directive¹ in my mongod.conf. The file itself contains a base64 text key the servers will use for communicating.

...
keyFile = /etc/mongodb/key.txt
...

My 18-monitor setup

2011-11-29T09:04:00Z

Sitting right now at my desk at Top Floor Technologies, I’ve got 18 monitors available to me as I work. Excessive? Nah. 18 is almost enough.

I’ve had a variety of computer configurations. Hell, I have a variety of configurations right now, ranging from a decent setup in my home office that’s stuck in the middle of an OS update I started last month and haven’t gotten back to, to a setup at my QM¹ with a 32” LCD TV as the main monitor (with a 23” on the side), to my Version2beta² laptop and 19” Dell monitor that I take with me when working onsite with customers.

In the past, I’ve done 3-monitor configurations. Once even a 4-monitor setup. They’ve been good, sometimes even fun.

Now though, at Top Floor,³ I count 18 monitors on my setup.

As you might guess, these aren’t all full, separate physical monitors. My desk is not so big, nor my chair so rolly. In fact, no one computer (there are three) has more than one physical monitor installed. But I have lots of virtual workspaces, and my fingers are very rolly.

My setup. Count along with me.

Main: Intel Core i7 @ 3.4GHz with 12 GB RAM. 27” LED monitor. 8 virtual monitors configured in pairs.
- General web on Chrome (1) and general consoles (2). Stuff like filling out my time card, logging into Rackspace, etc.
- Email (3) and Google Talk (4).
- Project web on Chromium (5) and project consoles (6). Whatever development project I’m working one.
- Interuptions, web (7) and consoles (8). The stuff that gets done inbetween the stuff I’m supposed to be working on.
Laptop: Intel i5 @ 2.4GHz with 6 GB RAM. 17” widescreen display. 8 more virtual monitors configured in pairs.
- General (personal) web (9). My webmail clients, twitter, etc. Also, general consoles (10).
- Google talk (11) and network tests (12). Yes, it’s the same Google Talk account. Sometimes I don’t want to switch away on my main monitor.
- Version2beta project work browsers (13) and consoles (14).
- Overflow (15) and (16). Okay, I have more monitors than I need on my laptop. Usually need.
iPad: first generation. It has a processor and some memory, but I don’t recall how much.
- Just one display (17). It often has my Twitter client on it, but sometimes I’ll throw email on screen, especially if I need to look at one of my more obscure personal accounts.
Dell 19” widescreen (18). Currently sitting on the floor next to my desk, unused. I need to buy an HDMI cable for my main system so I can run this one off the secondary output.

I configure CCSM (Compiz Configuration Settings Manager) to use Ctl-Alt-Arrows to move me from virtual monitor to virtual monitor, alt-tab (and alt-shift-tab) to get me between applications on my current workspace, and alt-super-tab (and alt-super-shift-tab) to switch me between all applications on all virtual monitors. It’s almost as fast as moving my eyes from one screen to the next.

Drupal, on GlusterFS, on Rackspace Cloud

2011-11-28T16:12:00Z

I ran Drupal on GlusterFS for a few days. I didn’t like it. Here’s why.

I’m building an eight-node server cluster on Rackspace Cloud for a Drupal site. Clearly, I want it to go fast.

It is a busy site, peaking at 20,000 unique visitors a day, maybe 100,000 pageviews. But the real fun part is that it has 1.6 million users.

Putting it on Drupal is a decision that predates me (and I withhold judgement - I don’t have nearly enough Drupal experience yet), but the rest of the structure has been up to me. I have specific goals, like 1.5 second response times even for users who are logged in.

My architecture goes like this:

Two load balancers, running nginx. Nginx is also serving all static content and boost cache from Drupal. Upstream servers are selected by ip_hash.
Two application servers upstream of nginx run lighttpd and fastcgi. Hello, Drupal.
A dual-master pair of MySQL servers back up the application servers. I wanted to run PostgreSQL, and I was really happy when I remembered that Drupal works with PostgreSQL. But Drupal only works with Postgres for some values of the word “works”. Modules may or may not support it. CCK has some issues too.
Two MongoDB servers hold up the rear and contain user-specific information for all 1.6 million users. A custom Python program manages authentication and data interface.

Originally, I intended to use DR:BD¹ much like I do with my high-availability Linode virtual private servers.² It turns out that you can do that on Rackspace a couple different ways, neither of which are recommended. The first requires you to make a loopback device, since it’s really difficult to get more than one block-level device on a Rackspace cloud server. The second requires you to get more than one block-level device on a Rackspace cloud server, which is not (as you might guess) only really difficult to accomplish, but it can also break things like backups and snapshots.

So bzzzzt to DR:BD, this time.

Next, I tried GlusterFS.³ The docs sucked⁴. At least I think they would have sucked if I’d read them, nearly all of them, because that’s how far you have to get into them before you can see what really needs to be done. In part, this highlights the absence of a useful quickstart guide. But there’s this sad thing, too. Once you manage to infer the organization of the configuration, it’s really rather elegant. You could think of it as a string of Unix pipes, each volume handing off to the next, but I picture it like a stack of matryoshka dolls. This volume (the littlest doll) provides a POSIX device. This volume (the next littlest doll) provides locks. The next volume (we’re getting bigger now) provides multiple I/O threads. And so on.

When all the dolls were stacked, I had a working GlusterFS filesystem across four servers.

Since I’m not the only person on this project (which is a good thing, or else we’d really be in trouble) we also had a development server at WebEnabled.⁵ I compared the Drupal devel module timing between the two systems and realized that I had a serious problem. What WebEnabled was doing in 1,200ms, my setup was taking 4,200ms.

Ultimately I tracked this down to GlusterFS. It appears (and I haven’t really tested this) that Drupal is I/O intensive enough that the delays of GlusterFS made a noticeable (and unacceptable) difference.

Here’s the timing for one page (logged into Drupal, so boost isn’t helping) with GlusterFS:

Executed 229 queries in 294.47 milliseconds. Page execution time was 3820.92 ms.

Here’s the same page without GlusterFS:

Executed 229 queries in 325.98 milliseconds. Page execution time was 839.68 ms.

Now I’m not here saying you can’t run Drupal on GlusterFS. I’m not even saying I had it configured right. (Remember, I didn’t read the docs - at least not all of them.) But I’m reasonably confident I had it configured and optimized well, and I got considerably better results when not using GlusterFS. In my case, I got acceptable results only without GlusterFS.

I still need to synchronize filesystems, so I tried another new-to-me program, lsyncd.⁶ I like it, and it works well for this application, where the file system is not changing all that often. Perhaps I’ll write something up on lsyncd another day.

Elavon Phishing Scam

2011-11-28T11:20:00Z

A customer received 40,000 mailer-daemon messages yesterday. I bet that was fun.

I’ve seen only four or five of the emails, but it was enough to demonstrate that someone had credentials to use the SMTP server to send phishing messages pretending to be from Elavon Merchant Services.

The text of the message goes something like this:

- Elavon Virtual Merchant Account Alert -

Your RETAIL account has expired. If you intend to use our services in the future you have to update your online info now.

- Download the attached login page in this email to continue. -

*** Please note : Only Retail accounts are locked. (Example : Market Segment : Retail )

Thank you - Elavon Merchant Services Team.

The attached HTML file looks like a login form for Elavon, but sends the data to a Japanese website http://depot-abc.com, deep linked into a Wordpress installation, probably hacked.

While it looked like the client’s computer might have been compromised, it turned out the messages were being sent from colocated or shared servers, at least the few I tracked down. Most of the messages I looked at were sent from The Planet in Texas, which I believe is still the world’s largest privately held hosting facility (hosting services, not like Facebook or Google which undoubtedly have bigger facilities.) Some of the other messages came from a machine hosted at Netelligent.ca, providing hosting services in and around Montreal. In this case, the machine seems to be in Laval, QC.

Tryton as an e-commerce back-end - Web Framework

2011-11-19T16:44:00Z

This is part three in a series exploring how Tryton might fare as the heavy-lifting portion of an e-commerce package. This third part explores what it means to do e-commerce.

The whole series:

First a review

If you read my last two posts on this topic, you’ll recall I’m in the process of discovering the whys and hows of using Tryton as the back end of an e-commerce system.

To review, Tryton¹ is an open source ERP² platform licensed under the GNU General Public License,³ written in Python⁴ and working with a PostgreSQL⁵ database server. Phew. In short, it’s a cool way to set up One Software To Rule Them All.⁶

In this post we get into the meat and potatoes: what is e-commerce anyway?

What’s e-commerce anyway?

There are many ways to define e-commerce. Perhaps most comprehensive would be a list of features that are implemented in a given e-commerce system. I considered creating such a list for this blog post. Ultimately, I decided to keep it simple, because the most simple view also exposes the biggest holes in almost all e-commerce systems.

Here’s the simple definition: e-commerce is the implementation of a business’ customer interface on the internet. If it involves a customer, a company, and a computer, it’s probably e-commerce.

Obviously this covers all the typical interactions: Buying a shirt, choosing an XL size. Reviewing my order before I check out. Paying for my order, getting my order confirmation. Getting a note when my shirt ships. Returning it when I discover the hole in the underarm. As you can see, this is all part of a process.

A lot of this typical process is well defined by best practices, and implemented in (almost) every sort of e-commerce platform. So why do we need to consider Tryton?

The problem is that my process, or your process, might actually be different. In fact, I can say my process is definitely different - I’m not selling shirts, or widgets, or anything concrete. I sell services as diverse as disinfecting a compromised web server to building an e-commerce web site. Many of my sales are projects, made up of milestones and tasks and billing stages. Sure, when I order a shirt online I might wish to see when it’s going to ship, but when my customer orders a web site, she might wish to see where we’re at in the process of producing it.

My work is not unique in its e-commerce needs. Heck, my field isn’t even unique. Just about everyone who sells anything other than widgets online needs something that’s either custom, or developed special for their vertical market. Lawyers, doctors, veternarians, carpet cleaners, housekeepers, daycare providers, doggie daycare providers, landscapers, painters, roofers, artists, appliance repair persons, gutter cleaners, chimney sweeps, HVAC cleaners, pest control applicators, plumbers, electricians, home theater installers, home automation designers, car mechanics, small engine mechanics, massage therapists, optomotrists, window cleaners, holiday window painters, office plant carepeople, office fish carepeople, snow removers, community theater, professional theater, cinema, parking space rental, storage space rental, apartment rental, house rental, hoteliers and moteliers, concierge and personal assistants - the list goes on.

Why can’t I pay my rent online? Schedule my car service? Schedule a gutter cleaning, a chimney cleaning, and a tooth cleaning - all for the same day? After all, I can practically watch Domino’s make my pizza with their new system.

E-commerce is the ERP interface

ERP is a system that manages documents as they progress through a pre-defined process. The documents very often represent something else — a purchase order represents some product that needs to be received and paid for, a sales order represents a set of products that need to be shipped and paid for, a work order represents some raw product that will be transformed into something new. We call the process they go through “workflow”.

The more accurately we model the real world with our ERP analogs, the more accurate the information we retrieve from the ERP will be. If we count incoming deliveries before marking an order received, we’ll probably have better inventory numbers and fewer overcharges from vendors. If we eschew back door sales, chances are good our sales figures will be closer to reality and more profitable too.

Normally speaking, our customer doesn’t have much more access to our ERP system than glancing at my screen as I fill out their order. That is, until we connect a website to it, and then much of the interaction they have is - or should be, as I propose - interfaced to the ERP.

Let’s start at the beginning with a brand new customer, and fictitious business called Hearty Dog Treats that sells natural treats for dogs. Jane Doe goes shoppping for some new treats for her Newfoundland (his name is Jayne, but he’s named for Jayne Cobb from Firefly and Serenity, not because it sounds just like Jane’s name) and she finds Hearty Dog Treats through a series of fortuitous events. Here is her process, and how it could and should represent an interface tightly integrated with ERP:

Jane visits the home page of the website to see who the company is. A sidebar advertisement is promoting 12” bully sticks.⁷ This advertisement might be promoted based on an overstock, or special pricing received, or other condition modeled within the ERP.
Jane clicks through to the product page for bully sticks. She reads the ingredients (wow - she didn’t realize what they were actually made of!) She looks at the price and stock status. The three primary pieces of information Jane pursued all come from the ERP system. A food’s ingredient list really must be kept up to date and managed in a system that allows for appropriate controls and distribution. Something like an ERP system would work. The stock status is certainly derived from the ERP’s inventory of bully sticks. And Jane’s price comes from the default web pricelist. Too bad she’s not a repeat customer or a dealer - she could create an account on the website, and get access to better pricing. Her pricing.
Jane is actually looking for duck hearts. She returns to the “Treats” section of the catalog. The catalog as a whole, and the navigation for it, may very well come directly from the ERP system. Indeed, if Hearty Dog Treats maintains multiple sales channels (say, online sales, a print catalog, and flyers to hand out at dog shows) they might wish to maintain the official copy for marketing materials in the ERP system as well, associated directly with the product or product line. Besides, the high school kids with the after school jobs running a point of sale terminal at Hearty Dog Treats’ retail location doesn’t know much about the product, so putting good product information into the ERP system will help them make the sale too.
Jane clicks through to Freeze Dried Duck Hearts. (These are real, by the way.⁸ I often claim that either of my Basset hounds would balance a piano on their nose for a freeze-dried duck heart.) See step 2 above.
Jane decides to buy. She fills out her order and submits it. She’s just created a new document in the ERP system and started the process of filling an order.
Jane gets a confirmation email from the ERP system. She clicks the link and sees that her order has been picked from inventory and moved to shipping. She thinks to herself how cool it is to be able to see the progress of her order.
When Jane’s order is ready to ship, the ERP system captures her credit card funds. Sure, it used to be that e-commerce systems grabbed the money when the order came in, regardless of whether it could ever be filled, but today’s system knows it’s better all around to charge for what you’re shipping, not what you hope to ship as soon as possible.
Jane receives an email that her order has shipped, containing a tracking number so that she can check the progress of her delivery. She does. The truck is out for delivery.
The day after delivery, she gets an email from Hearty Dog Treats⁹ saying she should have received her order, and asking if everything okay with it. It also suggests she might want to subscribe to their email newsletter, or follow them on Twitter, or fan them on Facebook. In fact, maybe Jane would like to click the link below and share Hearty Dog Treats with her friends on Twitter and Facebook.

Every step of Jane’s order interfaced with the ERP system. Many, many websites work this way, but far from all of them. Most tier 3 and 4 businesses aren’t running ERP at all. They’re getting by on Quickbooks, Peachtree, or sometimes even the shoebox accounting method. Many e-commerce systems don’t connect to any ERP system, or even to Quickbooks. Since that is the case, some companies actually use their e-commerce as an ERP package even though the design is weak at best.

That brings us to the next point: ERP is the company’s interface to the e-commerce system.

ERP is the e-commerce interface

It stands to reason that, if ERP is the flip side of the e-commerce coin, the e-commerce is the flip side of the ERP coin too, right? (If not, that would make a cool magic trick…)

For each step Jane when through, there were one or more steps that Hearty Dog Treats went through as well. They need to order the product and have some in stock. They need to process her order and issue a pull slip or pick ticket or delivery order or whatever Hearty Dog Treats is using to represent a stock move from inventory to shipping to the customer. They need to order more stock, if their inventory has fallen below the trigger point. They need to reconcile her payment on the back account. All of these tasks are triggered by, and performed in their ERP system.

In fact, the reason Hearty Dog Treats has ERP is so that these processes can be automated as much as possible. Hearty Dog Treats wants things to run smoothly and profitably, even when the boss is away.

Wow, so Tryton can do all this?

Nope.

Tryton can’t do it. OpenERP can’t do it. Magento can’t do it. Quickbooks, Peachtree, Sage — I don’t really think any ERP or accounting system, open or closed source, designed for small business, can do what I’m describing. (If you think I’m wrong about this, comment or tweet at me right away. I don’t want to redesign the wheel. I do want a system that does all of this, scales down to a mom-and-pop shop and up to tens of thousands of products and customers.)

I can’t let that sit. I feel like those statements need to be expanded upon. I hear someone saying, “There’s someone wrong on the Internet! He uses the handle Version2beta!”

First, Tryton can do it, because Tryton is a framework for developing business applications. Given a framework and a developer, much can be done. It’s my opinion after months of planning this project that Tryton can do it well. Nonetheless, Tryton cannot do it yet.

According to OpenERP SA, OpenERP will integrate with a number of popular open source e-commerce platforms, including OSCommerce, Magento, and Joomla.¹⁰ And they do - I know some of the people who have worked on the Magento Connector.¹¹ In the very next paragraph of the same page, OpenERP says that “to benefit from all the power of OpenERP … you can also use our own eCommerce system that automatically plugs in an existing website. Add a few lines of HTML in your website and OpenERP converts it to a full featured eCommerce platform.”¹² The rest of the page lists cool things you can do, a broken image link, and some comments from visitors asking how. If you dig deeper into their site, you can find the e-commerce module, version 5.0.1.0 (which often indicates it is a module for OpenERP version 5, even though it’s listed under OpenERP version 6). This module is not an official module, not quality certified, and reimplements Partner, Address, Payment, Shop, Category, Sale Order, and Order Line — all standard OpenERP objects — in order to achieve e-commerce. And so I hold that OpenERP cannot do this.

Magento can’t do this.¹³ Now that they’ve been bought by eBay, Magento is doing a lot of cool things, and it would not surprise me if they moved in this direction down the road. (I’m certain many companies will.) They’re even providing a US$1 million stimulus to encourage small business e-commerce.

Quickbooks might be able to do it, if you go for the Enterprise version and add on third party software. It’s my impression (earned in part by having used Quickbooks) that it’s a great accounting package for small businesses, but the software’s enterprise features are flawed with usability issues. Then there’s the third party problem for the e-commerce software - how integrated is it with Quickbooks? Both Quickbooks and the e-commerce package are proprietary; how well are they kept in sync? Is the quality of the code there? I’m not trying to engage in FUD¹⁵ — I’d be thrilled to hear about a solid Quickbooks e-commerce solution that goes beyond widgets. Let me know and I will list it right here.

Peachtree can’t do it — or if it can, we’re looking at the same situation as Quickbooks at best. But Peachtree’s parent company, Sage, can do it. Once again, though, you need to use their proprietary module (eBusiness Web Services) combined with a third party e-commerce application.¹⁶ Besides, Sage is pretty pricey for a small business. We had a customer gut OpenERP (and us, along with it) and put in Sage. Rumor has it the licenses alone were more than US$80,000, plus integration, plus e-commerce software and hosting.

Netsuite can probably do it. With Netsuite Ecommerce, a lot of set-up, and US$1,200 a month, a mom-and-pop shop can sell widgets online and get the benefit of Netsuite’s accounting, finance, and supply chain management. At that price, it is probably affordable for a business doing a million a year in sales. (Mom and pop are good at this widget business.)

You know who can do it, with open source software even? Apache OFBiz.¹⁷ Honestly I know very little about OFBiz (which stands for “Open For Business”), but it clearly does a lot, and if you’re looking for a community open source ERP project that does web e-commerce right now, this might be your choice. If you use it, let me know what you think?

What does Tryton need?

I think Tryton needs a web framework.

I’m somewhat at odds with Cédric Krier¹⁹ on this, I know. We’ve chatted on IRC, and he assures me that Tryton doesn’t need a full framework; JSON¹⁸ is enough. And he’s right - JSON is enough. Tryton itself is a framework, and with an easily implemented interface like JSON, one can do all I describe above. But unless there’s something on the other end of the JSON, it’s useless to my customers, and in turn useless to my customers’ customers.

Cédric’s “something on the other end of the JSON” approach puts Tryton as a rather smart back end to a web application (any application, really - I can totally see using it behind an Android or iPhone app), without building an actual framework into Tryton. But I’m not sure that a real web framework built into Tryton wouldn’t be very, very useful. I have ideas on how to do this, involving some of my favorite tech tools - MongoDB²⁰, Jinja2²¹ or Restructured Text²², maybe a little Flask²⁴ action. Some Redis²⁵ and Celery²⁶ to make sure we never ask Tryton to try too hard.

I’m certain this can be done. I’m pretty sure I can do it. I strongly suspect it’s already been done by at least one company, for private use.

A Tryton web framework

When I first started this blog post almost two months ago, my work life was considerably different than it is right now. At that time, I thought I may be days from starting the web framework project for at least one, and maybe two different customers.

Hell, when I first started this series of blog posts on Tryton as an e-commerce back-end, I thought I’d have a Tryton-based e-commerce platform done by now. Funny, that. But then life knocked on my door - kicked it in, more like. Among other things, I worked with one of the best Tryton programmers in the world for nearly 40 hours straight, and at the end of it, I had lost my biggest customer. (Correlation, not causation - he didn’t steal my customer or anything like that.)

I still have prospects for using Tryton as an e-commerce back end. If any of them come through I may continue the series. Until then I hope it’s been a useful discussion.

What else is there?

Part 1: About Tryton. An overview of Tryton – architecture, business model, features, license.
Part 2: Stalking the Tryton community. There are some impressive people working on this project. I’m pleased to get to know them.
Part 3: The crux of the matter. What does a Tryton-backed e-commerce system need to do?

If I’ve inspired a response from you (probably the type of response that deserves an apology), mention @version2beta on Twitter and I’ll see it there. Or, if you can’t comment in less than 126 characters (my handle takes 14 characters with a space), blog about it and tweet that.

DjangoCon, part four

2011-09-11T16:55:00Z

Here’s the recap of day three, the last day of the talks. Apparently I took good notes on day three, since this is by far the longest post, and the one it took me longest to produce. It was, indubitably, an excellent day.

DjangoCon agenda

Keynote: Glyph Lefkowitz on Why does Django hate Python

In 2008 Cal Henderson gave a DjangoCon speech entitled “Why I Hate Django”. Apparently it provided an opportunity for Django core developers to demonstrate gracefulness and generosity. The speech has also created a legacy. 2011’s Why I Hate Django keynote was given by Glyph Lefkowitz, founder of the Twisted framework project. Glyph’s particular flavor was not why he hates Django (indeed, he’d only tried to install it once before writing the speech) but why Django hates Python. His premise is that, if Django didn’t hate Python, we might start using more Python (and fewer tools that are not based on Python.) “It is not enough that your system be well designed. It needs to be consistently designed as well.”

In order to assist Djangonauts and developers in moving toward a more Pythonic environment, Glyph recommended a number of projects we might wish to consider. It may be more accurate to say that Glyph challenged the Django community to use more Python, including these projects. (I’m breaking my style guideline on footnotes over in-line links for this list, since it’s basically a list of links.)

Fabric: Fabric is a library and framework for performing systems administration tasks, both locally and over a network using SSH. I’d heard of Fabric prior to DjangoCon, but not used it. Using Fabric will be one of the key benefits I got from this year’s conference.
WSGI and Web servers
- FAPWS: FAPWS stands for Fast Asynchronous Python Web Server. It’s WSGI-compliant with two primary goals: keep it small, and keep it asynchronous.
- Spawning: Spawning is a multi-process, multi-threaded, non-blocking WSGI compliant web server. Eric Florenzano wrote about using Spawning and Django together.
- Tornado: Tornado is the open-source version of a Web server written to run FriendFeed. It is non-blocking, relatively fast, and includes some of the tools they use at FriendFeed.
- Gunicorn: Gunicorn is a Python fork of the Ruby Unicorn project, a prefork worker model WSGI web server.
- Cogen: Cogen is a library for network programming using coroutines via enhanced generators. I don’t understand it either, but it provides a WSGI server and is asynchronous.
- Twisted Web: Twisted Web is the HTTP portion of Twisted, an event-driven networking engine.¹ It may be important to note that this is the project that Glyph heads.
Lamson Project: Lamson is a Python-based MTA, SMTP server and library. While that probably says enough, I want to add it can use anything Python can use for storing it’s data, which of course includes the Django ORM as well as CouchDB, MongoDB, and many other options. I’m pretty excited about Lamson.
PyDNS: PyDNS is a Python library for performing DNS lookups.
PyCron: Glyph, maybe you were just stretching at this point, looking for more applications? I found two pycrons. Neither are in PyPi. One doesn’t seem to include source. Here is the other.
Python Sched library: sched is the Python event scheduler library.

Glyph had much more to say in his keynote, much of it about the overall monolithic structure of Django. Really, I think it boils down to “Django is not particularly Pythonic.” But that’s beyond my pay grade right now — I am improving my Django skills so that might work with more customers. I am not asking to join the core development team in planning Django 2.0.²

Advanced django forms

Danny Greenfeld and Miguel Araujo proposed and then demonstrated a more thoughtful application of The Zen of Python³ and Spartan Programming⁴ to Django forms. Some key points for me:

Limit vertical complexity
Limit conditionals
Flat is better than nested

A few more notes I took for making Django forms easier for me in the future:

See Jacob Kaplan-Moss’ post on Dynamic Form Generation
Check out django-uni-forms, especially since it is Section 508 compliant
Also look at django-floppyforms.

Danny Greenfeld and Miguel Araujo’s slides

Teaching Django to comrades

Issac Kelly gave some excellent starting advice, which is good since I’m still starting in many ways.

Some key troublespots:

The ORM
Reverse relations
Migrations
Forms, forms, forms
Signals
URLs
Deployment

Of course, it might have been easier to list the areas that are not key troublespots…

I particularly liked Issac’s Axioms:

There is a method for that. Corollary: If you’re new and it’s hard, you’re probably doing it wrong.
There is a Django way of doing this. Corollary: Don’t fight the system until you are assured of victory.

Deployment, daemons, and datacenters

I wasn’t sure if I’d need to read five years of Andrew Godwin’s blog in order to describe him well enough. He is young and brilliant, in a polite, relatively unassuming way. I like Andrew; he impresses me. And his presentation was good too — solid mechanics and good advice, some of which I recorded:

ep.io’s standard stack is nginx (for static files and gzipping), gunicorn, postgres 9, redis, and virtualenv
For higher loads, Andrew says:
- Use Varnish or other upstream caching
- Use HAProxy or nginx for load balancing
- Give PostgreSQL more resources
Keep staging identical to production
Redis is easy to setup and use
Ship long-running tasks off. Use Celery or a worker solution.
Don’t use SQLite in production
Use transactions
Plan so you can always add another machine.
Keep things loosely coupled - simple components loosely connected are easy to troubleshoot
Anything you’ve done at least three times should be automated
ZeroMQ is really, really nice to have.
MongoDB and Redis will both lose data when they are running low on disk space.
Regarding backups and redundancy
- A Postgres slave is not a backup
- RAID is not a backup
- Backup in multiple forms at diverse locations
- Make sure you can restore from backup on a somewhat regular basis
- Always make the backup machine pull the backups, not production push

I didn’t know about that last one. In fact, for much of my backup process that wouldn’t work - Amazon S3, for example, cannot pull files. But I’m curious about this.

I don’t know that Andrew exactly promoted this, but I put in my notes to make backups progressive. I’ve thought of this before, too. Perhaps ep.io is doing it. Here is a progressive schedule of backups to keep:

First of year
First of month
First of week

Content ain’t a thing: web scraping

I missed the first half of this session, and decided not to take notes on the second half. While much of what the presenter offered was technically interesting, I found myself increasingly bothered by the blasé disregard for intellectual property rights on the web. It’s not that the speaker did not give some homage to the concept, just that a cultivated ability to split hairs was obviously a part of her professional upbringing.

So no further notes on the topic.

Cache rules everything around me

Especially on Thursday, the last day of talks, there were quite a few tag-team presentations. Some of them were so highly tuned, they finished each other’s sentences. Each one I saw was impressive.

Jacob Burch and Noah Silas presented on caching. Jacob works with Frank Wiles and now Jacob Kaplan-Moss at Revolution Systems. I don’t actually know where Noah is from, but he certainly knew his stuff too.

First, I liked the warnings.

Don’t cache:

Caching adds complexity. Don’t do it.
Caching adds additional points of failure. Don’t do it.
Modern databases are “stupidly optimized” (by which they seem to mean fast). Maybe you don’t even need to cache. Don’t do it.
As the quote goes, there are only two hard things in computer science. Naming things and cache invalidation. Don’t do it.

If you do cache:

Your application should never ever rely on caching.
The Super Awesome Fun Time Rule for Minimizing Sadness: You should have one canonical data source, and it is not your cache.

From here, they went into a number of caching patterns. I’m not going to translate all my notes here. I took too many for just part of one blog post. Suffice with just a few.

The “New Hotness” model:

Cache indefinitely - for as long as the back-end will allow.
Invalidate explicitly - replace, don’t delete

“Publish Cache”

Put entire responses into cache
Implementation in Django is easy - just drop in middleware
Cache invalidation is probably easy too, when the blog post, article, etc. is updated
Might be difficult with dynamic page elements

Django.core.cache is simple to setup and supports multiple caching backends. Don’t use these backends:

db.DatabaseCache
filebased.FileBasedCache
locmem.LocMemCache
dummy.DummyCache

These are okay:

memcached.MemcachedCache
memcached.PyLibMCCache

And these are good for the Publish Cache pattern:

django.views.decorators.cache.cache_page
django.middleware.cache.UpdateCacheMiddleware

Third party enhancements for Django:

django-newcache
johnny-cache
django-cache-machine
django-autocache

And finally, they recommend that it might be good to create a “Does Not Exist” cache entry, something like !!!DNE!!!. This can come in handy if you ever get slammed for something that, no matter how many times the back-end is asked for it, you cannot create it. Putting !!!DNE!!! in cache can prevent the extra load of discovering over and over that it does not exist.

DjangoCon, part three

2011-09-08T07:45:00Z

Day two is done, and I’m trying to do better at this. I definitely have better notes than yesterday. I have slightly fewer of them too, since the South session was absolutely worthless to me, and I blew off Tilting at Windmills to go up to the room with a stomach ache and a customer who needed help.

DjangoCon agenda

Here’s the recap of day two:

Lightning talks

By the schedule, there were 90 minutes of lightning talks happening in the morning. I guess when 9:00AM rolled around, there were only three submitted so they postponed the start time. This gave me enough impetus to put together my own, on using Prezi for presentations. Here it is. (This is my first time embedding media in this blog. It may look crappy until I spend more time formatting. If so, please excuse the visual disruption; I will try to make it better soon.)

A Prezi on Prezi for DjangoCon on Prezi

Keynote: Brad Fitzpatrick

Laid-back, easy-going guy who receives a lot of emails. Of course, he’s written some pretty important open source software, like memcached¹, and an Android app that runs on a server in his garage scanning for his phone’s SSID so that when he rides up on his motorcycle it can open the garage door automatically².

Brad Fitzpatrick’s slides

Restful API’s, promises and lies

Tareque Hossain gave an excellent, easy-to-follow introductory presentation on Django REST APIs, with illustrations from PBS children’s shows. Really - his opening slide had Elmo on it, which was only the start of visuals borrowed from his employer. His technical examples also came from PBS, of course.

One cool takeaway: JSONP³ for sharing APIs across multiple domains.

Tareque Hossain’s slides

A Little South Sanity

I really tried to get something out of this session. I can totally see how useful South could be, despite my initial misimpression that it was useful only in the largest, most drawn out installations. Unfortunately, Brian Luft did much of the presentation live at the terminal and it wasn’t working, so I have no clue which parts were South and which parts were Brian’s attempts to get it working. It was like I was the customer showing up for training, looking over the lead developer’s shoulder. Which is something I’m sure I’ve done with customers before. Many times.

Building APIs with TastyPie

Isaac Kelly did in fact convince me I could build an API with TastyPie. Someday I’ll have to prove that, right?

I hadn’t known about json.tool for pretty printing JSON from the command line, but that looks useful too:

$ echo '{"json":"obj"}' | python -mjson.tool
{
    "json": "obj"
{

Stop Tilting at Windmills: Spotting Bottlenecks

By the end of the JSON session I was already fielding voicemail, email, and text messages from a customer back home. One of us had managed to break his brand new Box.net uploader library.⁵ I spent the Stop Tilting at Windmills session tilting at windmills.

Safely Deploying on the Cutting Edge

This session was very cool. Best parts included:

a good branching model for git⁶
a vocabulary of deployment verbs implemented in scripts (pull, build, tag, sync, install, rollback, start, stop, restart, reload)
a reminder that the startup processes (init.d, upstart, etc) are just processes written by other developers, and sometimes it makes sense to roll your own

The session next door, “Best Practices for Front-End Django Developers”, was clearly very popular as well. Roughly half the conference attendees were in the same session I was. Next door, it sounded like they had about one-and-a-half times as many people clapping.

Benevolent Designer For Life’s Keynote - Designers Make It Go to Eleven

Idan is cool, an excellent choice. I’m guessing that, just as design implements the physical interface between human and machine - where the skin meets metal - Idan may do the same for the Django community.

Idan Gazit’s slides

DjangoCon, part two

2011-09-07T12:04:00Z

Oh my, I’m not good at this. There are people around live-blogging (and live-noting) the conference, and I’m scraping the sleepiness off my face enough to see the screen, only to realize I don’t actually remember yesterday very well. I may be exagerating a little, but I did not take adequate notes (sorry, PyDanny - I heard you on this, really I did.)

DjangoCon agenda

Here’s the recap of my first day:

David Eaves’ keynote on the science of community

David Eaves is awesome.¹ He doesn’t know much about the Django community, but he clearly knows about communities, including open source communities. Mostly he talked about two topics: using metrics to track and plan within a community, and using negotiation theory to include more people more effectively. Here was an important, but only peripheral, take-away: the involvement of women in an open source project is a key indicator of the health of the project.

From designer to Django’er in six weeks

I understand that Alex was ill in the morning, so his slot went to Adam Nelson for ‘Testing with Lettuce and Splinter’. I arrived about four minutes in, and left about eight minutes in. I’m sure it was a valuable talk. I’ve even heard that from someone who stayed. I missed the discussion on BDD (behavior driven development), and given that my testing experience is in many ways nascent, I decided that learning alternatives was further out than I was ready for.

What I did instead was listen to Tracy Osborn’s³ fascinating ‘From Designer to Django’er in Six Weeks: A Story from Solo Founder’.⁴ Anecdotes of learning not just Django, but of becoming an entrepreneur with Django.⁵

Alex did present in the afternoon, opposite Frank Wiles’ PostgreSQL Performance talk, so I did not see him then either.

Tracy Osborn’s slides

Confessions of Joe Developer

Danny claims to be lazy and stupid, which is not only a solid rhetorical tool but also a good starting point for learning complex systems. I half expected to hear a reference to Larry Walls’ three virtues of a good programmer.² (In fact, it would have fit right in, except for the language mismatch.)

A good portion of his presentation described various methods for saving time, not relying on memorization, and the merits of doing things “the dumb way”. His example contrasts urllib2 with the requests library⁶.

Also terrifically important, Danny talked about diversity at the conference - PyLadies,⁷ women in tech, our wives and daughters, the role of men in creating open and diverse environments. I like the thought, and believe Danny’s sincere desire to make it all better. At the same time, as men we risk being the patron by being involved at all. There’s much more to be said here, but I’m reluctant to say it for fear my position stands counter to the enlightened attitudes around DjangoCon. It does not. I salute the efforts to include women, to build an inclusive environment and to identify narratives and behaviors that interfere. I think I’ll stick with helping my daughter with her AP Physics, AP Calculus, and AP Statistics homework. If she even needs help.

Danny Greenfeld’s Joe Developer slides

The story and tech of Read the Docs

I like devops.⁸ I’m fascinated by it. Eric Holscher’s discussion on Read the Doc kept my attention as much as anything for how they solved the devops questions - clearly, that is the essence of Eric’s work, right?

Testing - The developer strikes back

This was the densest presentation I’ve seen at Djangon yet. Sandy⁹ laid it down at 10 MBPS, I swear. I knew little about “real” testing practices before I went in, and now I have enough familiarity with the universe of options, I’m highly likely to recognize them when I finally sit down and learn them right.

Sandy Strong’s slides

Secrets of PostgreSQL performance

I’ve known of Frank Wiles and Revolution Systems¹⁰ for about as long as I’ve used PostgreSQL, so much of what he offered wasn’t brand new to me. But some of it was new - tablespaces, for instance.

Frank Wiles’ slides

DjangoCon, part one

2011-09-04T12:33:00Z

It’s two days before DjangoCon, I’m on a plane to Portland, and I’ve mostly been spending my time reading the Django Book. That has me thinking about what I expect from the conference.

DjangoCon agenda

You’ll be able to recognize me at DjangoCon.¹ I have all of the sessions printed and three-hole punched inside a 1.5” binder, with my schedule inserted in the front cover. My schedule looks reasonably good - I have a session I want to see during every session, and in two or three slots, I have both sessions selected. I’m hoping my wife will be willing to attend the my counter-sessions.

Tuesday

Alex Gaynor, Summer in the wild

Optimization is good and I always want to feel reasonably confident I can make things faster, so I look forward to this discussion of PyPy. But I’m also curious to hear about Alex’s position at Quora, a company I expect to fail, which probably indicates it will be wildly successful.

Confessions of Joe Developer

Joe’s real name is @PyDanny,² which is why I plan to attend this session. Danny promises the benefit of his experience. I’ll buy that.

The story and tech of Read the Docs

When I first visited readthedocs.org,³ I wondered how they got anyone to put in the effort to maintain their centralized database of software manuals. I learned enough more to make the important connections between Sphinx, GitHub, and Django to make the automation connection. Now I’m fascinated with how Read The Docs came about and how it works. This session is one of my highest priorities.

Testing - The developer strikes back

I don’t know who Sandy is, but I hope she or he is going to help me greatly improve my understanding and use of unit tests. I’ve been coding a long time, like over thirty years now, but generally by myself. Shortcuts have happened, but I know I do not want to develop code without tests any more.

Secrets of PostgreSQL performance

I have over the years had a few opportunities to optimize databases, most recently PostgreSQL. I’ve used three different resources for that task, and they all came from Frank Wiles.⁴ Hell yeah I’m going to hear him talk.

Wednesday

RESTful APIs - promises and lies

You may have heard the term “mobile first”. I’m beginning to wonder if we shouldn’t be looking at the API goals of a website first. I’m serious. Even the semantic web has goals related to sites as applications - what’s my address, hours of operation, etc. What web site has no opportunity to share?

Real-world Django deployment using Chef

Scale is one big reason I am learning Django and attending DjangoCon at all. Besides, it sounds like Glyph is going to complain about Chef in his keynote address “Why does Django hate Python?” Might be good to have more context with Chef, as Glyph specifically mentions it.

Building APIs in Django with Tastypie

See above on APIs. Now add in Tastypie, which does sounds interesting.

Making the Django ORM multilingual

I’ve only had the opportunity to work on a couple of multilingual websites so far, but I expect that frequency to shift. I don’t have a clue yet how Django does i18n, translation, or localization. I’m worried it doesn’t do it well, based on the description of this session.

Stop tilting at windmills - Spotting bottlenecks

This session promises the kind of good advice you only get from someone experienced. But just as important to me, it appears that the best Django debugging tools will be demonstrated and potentially discussed.

Safely deploying on the cutting edge

Urban Airship’s Eric Holscher⁵ talking about Fabric, Gunicorn, Virtualenv, Rsync, Supervisord, and Python in fully isolated server environments? Oh yeah. Yes yes yes. This session is like sysadmin porn.⁶

Thursday

Advanced Django Form usage

@PyDanny’s back again, talking about forms. We have a customer who does a lot with forms, and I’ve already redesigned everything for them once, more than a year ago, to drastically improve the usability of their forms. That was the front side - how good can it get on the back side?

Pinax after three years: lessons learnt and the way forward

I want a content management system that my customers will rave about. I have no idea what that is with Django, if there even is someone complete and open source. Learning more about Pinax seems like a good direction to go, though.

Advanced security topics

‘nuf said.

Write the friendly manual

There is some really good stuff going on in the Python world with documentation. Sure, I’ve read most of the Django book, but have you looked at Flask’s documentation? Geez, who has time to not only code a cool platform like Flask, but also put out documentation of that quality? There has to be some secrets. I plan to uncover them.

Content ain’t a thing: Web scraping with our favorite Python libraries

I’m bracketing my DjangoCon experience, right? Share (APIs with Tastypie, etc.) and share (Web scraping) alike!

Cache rules everything around me

I’m back to scale, and systems administration, with this last formal session.

Find me?

So these are my picks for DjangoCon - find me at any of these sessions, strike up a conversation, and maybe even talk about sprints. I’m the guy with the long beard, big laugh, and contemplative look, trying to figure out how to put this new knowledge to use professionally.

jQuery.multiup plugin for multiple files

2011-08-30T20:21:00Z

I’m a progressive. By this I mean I like things to get progressively more complicated. Don’t you? Of course, I don’t mean that things should be more complicated than they are — in fact, I think they should be at most only as complicated as they are. But they don’t have to start off that complicated; they can get complicated progressively.

One example is file uploads. Sure, some of the time we only need to upload one file, and the way that HTML inputs (type=file) work is okay. But often, we need to upload more than 1 file. Maybe 10 of them. Maybe N of them.

So a progressive solution is to show one file uploader, but give the visitor the option to get another file. And another one. Right up until they’ve selected as many as you’re willing to allow them to upload, or as many as they have.

I did a review on this, and found some options. Malsup¹ has a nice option for Ajaxifying forms, even does an Ajaxian-type thing for file uploads, but it doesn’t let you build a package of files to upload. I found some nice stuff from the Stickman², but only in a MooTools dialect. Long story short, I wrote a plugin.

I guess that’s significant. It is my first jQuery plugin. Feedback welcome, please.

Here is an example of the jQuery.multiup plugin in action. Source code hosted at GitHub.

Keep the progressive idea in mind. I hope to write something about progressive authentication systems eventually too.

blog reset

2011-08-25T08:27:00Z

My blog had been hosted on a legacy shared server account that I don’t manage, running Wordpress. I was frankly embarrased to tweet links to it - often it wouldn’t come up at all. Now that’s all changed. Now my blog is on Django. (But only kinda.)

A month ago I heard from a potential customer in Chicago who needs to move a site from Wordpress to Django.¹ They’d heard about me in particular from a colleague, and wanted to know if I could help.

Of course I could! I’ve been learning Django for months. I even have the Django book.² I’ve even read the Django book. Parts of it, anyway. And I do have tickets for DjangoCon³ in Portland next month.

I’m afraid the colleague might have oversold me. It wasn’t complete radio silence when I said I didn’t have any production Django sites she could look at, but it was close.

“But I was thinking of moving my blog onto Django. I could probably have that done over the weekend.” I could hear her slight sigh of relief.

I didn’t move my blog onto Django that weekend. What I did with my weekend was build out two sets of high-availability server pairs and write up a how-to on the same.⁴ Then, at 5:30pm on the night before my 6:15am train to Chicago to meet with this customer, I sat down to figure out how I could get my blog onto Django, quickly.

I was already somewhat familiar with Hyde⁵. It’s a static site generator written in Python, inspired by a remarkably similar Jekyll⁶ written in Ruby. Conveniently, Hyde is also based on Django, using both the settings.py file and Django templates, so I’d be able to use some of what I already knew to show off what I might be able to do for a customer using Django. Because that’s what every customer needs, right? A small, unobtrusive, under-designed blog-like website that uses Django templates?

My whole blog did not get moved that evening, but I did post the high-availability write-up in a blog section of a website I called “proof of concept”. That was convenient, since the new site is hosted on one of the server pairs I’d just built, and the write-up was a key take-away for my presentation at Web414⁷ the next night on cloud hosting.

A couple of weeks passed before I got back to the project, but I did finally move my blog. I also improved the design some, though of course it’s still small, unobtrusive, under-designed, and blog-like. I did away with almost all of my inline links in blog posts, opting for footnotes instead. Not only does this make the reading easier (and less spammy-looking), it gives me more room to be snarky⁸ out-of-band. (There are many fine examples of footnoting bloggers, but as usual Barbara⁹ has the most influence on me.) I’d also been wanting to use Disqus for comments, and figured it’d be best to do that while I only have 8 comments to figure out what to do with. (Note: if you are one of the people who’d commented on my old blog, please accept my apology as I decide how, when, and if to move them.)

I haven’t heard back from the Chicago customer yet, which is probably okay because they wanted freelance, and I would want to run it through QMUXS.¹⁰ But in spite of the dubious benefit of the sales call, I worked through the new high-availability challenges, built up the new servers, got to work with Django more, and built a new fast fast fast website for version2beta.

I hope you enjoy it.

High Availability Linode Pairs

2011-08-08T20:08:08Z

Something like a recipe for creating database and application servers that failover on each other. This recipe is for Lighttpd and MySQL, but it can be extended easily for other applications. Later, I’ll do one for Tryton and PostgreSQL. I use Ubuntu 11.04 64-bit on two Linode virtual private servers.

Initial configuration, from within the Linode¹ dashboard

Repeat the following procedure for each member of the pair:

Go to the Remote Access tab on the Linode dashboard.
- Note the public IP. We’ll need that later.
- Add a private IP, and note it. We’ll need that later, too.
- If by any chance you have an IPv6 private address, note that. We’ll use it all. If you don’t have one, click on “Enable IPv6” and get one. We really are going to use it.
- Set a Lish password. I use APG² to generate new passwords, e.g. apg -n 100.
- Add an SSH public key to the Keys box. This is in addition to the Lish password. Or instead of it, if you prefer.
- This is also good time to write to Linode support. You’ll need to order an additional public IP address, and by default you are unable to purchase an additional IP address until you’ve requested and justified it specifically with support. You’ll also need another private IP address, which doesn’t cost anything but it takes a support person to do it - there’s no way to automatically add more than one private IP address. These addresses will float between the pairs, so it doesn’t matter which of the pair get these additional addresses. Make note of these floating addresses.
Go to the Settings tab in the Linode dashboard.
- Give the Linode a good name.
- Name a display group for your pair. Generally we use “location pair”, e.g. “newark pair”, “dallas pair”, etc.
- Change the email alert thresholds
  
  CPU Usage: 50% (Default 80% over two hours is too high)
  
  Disk I/O Rate: 2000 I/O Ops/sec (Default 1000 ops over two hours is too low)
Go to the Dashboard tab in the Linode dashboard, and click on “Create a new Disk Image”.
- Create a disk image called “var-www”, type “unformatted / raw”, with a size that’s half your available space (just as a rule of thumb).
- Create another disk image called “var-mysql”, also type “unformatted / raw”, with a size that’s half your remaining available space (again, a rule of thumb).
Also on the Linode dashboard tab, choose to deploy a Linux distribution.
- I like the latest Ubuntu³. Right now, that’s 11.04 64bit.
- Set the swap partition size next. Max it out, which on Linode seems to be 1 x RAM.
- Use everything that’s left for the deployment disk size.
- Set a good root password.
- Press the Deploy button. That takes you back to the dashboard.
The configuration profile name is too pedestrian. “My Ubuntu blah blah blah”. I already knew it was mine, after all I made it, didn’t I? Click the “Edit” link to the right of the profile.
- Change the label to something with some teeth. Something like “Ubuntu 11.04-64bit high availability pair”.
- Assign block devices for our DR:BD drives. Put var-mysql on /dev/xvdc, and var-www on /dev/xvdd. Save the changes, and you’ll be taken back to the dashboard.
Each server needs to be able to assume the IP address from the other. From the “Remote Access” tab for each server, select all of the addresses, both the public and private, for IP Failover.
Boot the system and log in, either through SSH using the root password you just assigned, or through Lish.

This is a good time to repeat the previous six steps on the other member of the pair. Once that is done, continue to the final step in the Linode dashboard.

Some notes before we get into it

We have a number of implementation specific details, stuff that changes for each cluster. I’d like to refer to things in here in a way that is internally consistent, yet easy to use. So here is an index of variable data:

Server names: alice and gertrude. If you’re running vi or vim against this document, and wanted to name your systems Fred and Wilma, you should be able to run :%s/alice/fred/g followed by :%s/gertrude/wilma/g.

IP addresses:

Public IP on alice: pub.a.a.a
Private IP on alice: priv.a.a.a
IPv6 on alice: ipv6:pub:a
Public IP on gertrude: pub.g.g.g
Private IP on gertrude: priv.g.g.g
IPv6 on gertrude: ipv6:pub:g
Floating public IP: pub.f.f.f
Floating private IP: priv.f.f.f

Initial configuration, operating system

Again, repeat the following steps for each member of the pair:

Name the server.

echo 'alice' > /etc/hostname⁴

Fix (as in “make permanent”) the networking in /etc/network/interfaces:

#vi /etc/network/interfaces

    # The loopback interface
    auto lo
    iface lo inet loopback

    # Configuration for eth0 and aliases

    # This line ensures that the interface will be brought up during boot.
    auto eth0 eth0:0

    # eth0 - This is the main IP address that will be used for most outbound connections.
    # The address, netmask and gateway are all necessary.
    iface eth0 inet static
     address pub.a.a.a
     netmask 255.255.255.0
     gateway gate.a.a.a

    # eth0:0
    # This is a private IP
    iface eth0:0 inet static
     address priv.a.a.a
     netmask 255.255.128.0

    # eth0 ipv6
    iface eth0 inet6 static
     address ipv6:pub:a
     netmask 64
     gateway fe80::1

Deal with some networking stuff that makes our life easier.

#vi /etc/hosts

    127.0.0.1       localhost.localdomain       localhost
    pub.a.a.a       alice.version2beta.com      # FQDN is public
    priv.a.a.a  alice               # short name is private
    pub.g.g.g   gertrude.version2beta.com   # FQDN is public
    priv.g.g.g  gertrude            # short name is private
    pub.f.f.f   ag.version2beta.com     # public floating IP
    priv.f.f.f  ag              # private floating IP

    #IPv6 addresses
    ipv6:a      ip6-alice
    ipv6:g      ip6-gertrude

    # The following lines are desirable for IPv6 capable hosts
    ::1     ip6-localhost ip6-loopback
    fe00::0 ip6-localnet
    ff00::0 ip6-mcastprefix
    ff02::1 ip6-allnodes
    ff02::2 ip6-allrouters

SSH Keys

# ssh-keygen -t rsa

Copy /etc/.ssh/id_rsa.pub into the other server’s /etc/.ssh/authorized_keys, which you’ll need to create. While you’re at it, put your key in there too.

Restart networking. (If there’s a mistake in the previous step, we’ll catch it either here or in the next few steps.)
```
# /etc/init.d/networking restart
```
Bring the package database, and installed packages, up to date.
```
# apt-get update && apt-get upgrade
```

Install and configure DR:BD

Install the tools

First, try:

# apt-get install drbd8-utils
# drbdadm -V

At the top of the options, it will give the version of the DRBD module and userland version. If these match, you’re golden. If they don’t (as is the case currently for me), you need to build userland tools from source.

# drbdadm -V

DRBD module version: 8.3.10
   userland version: 8.3.9
you should upgrade your drbd tools!

Building drbd8-utils from source:

# apt-get remove drbd8-utils
# apt-get install psmisc build-essential flex git xsltproc
# wget http://oss.linbit.com/drbd/8.3/drbd-8.3.10.tar.gz
# tar xvzf drbd-8.3.10.tar.gz
# cd drbd-8.3.10
# ./configure
# make
# make install

All good? Try the drbdadm -V again:

# drbdadm -V

...
Version: 8.3.10 (api:88)
GIT-hash: 5c0b0469666682443d4785d90a2c603378f9017b build by root@alice, 2011-08-06 22:35:11

Configure DR:BD global settings in /usr/local/etc/drbd.d/global_common.conf:

# vi /usr/local/etc/drbd.d/global_common.conf

    global {
        usage-count yes;
        # minor-count dialog-refresh disable-ip-verification
    }

    common {
        protocol C;

        handlers {
            pri-on-incon-degr "/usr/lib/drbd/notify-pri-on-incon-degr.sh; /usr/lib/drbd/notify-emergency-reboot.sh; echo b > /proc/sysrq-trigger ; reboot -f";
            pri-lost-after-sb "/usr/lib/drbd/notify-pri-lost-after-sb.sh; /usr/lib/drbd/notify-emergency-reboot.sh; echo b > /proc/sysrq-trigger ; reboot -f";
            local-io-error "/usr/lib/drbd/notify-io-error.sh; /usr/lib/drbd/notify-emergency-shutdown.sh; echo o > /proc/sysrq-trigger ; halt -f";
            # fence-peer "/usr/lib/drbd/crm-fence-peer.sh";
            split-brain "/usr/lib/drbd/notify-split-brain.sh root";
            # out-of-sync "/usr/lib/drbd/notify-out-of-sync.sh root";
            # before-resync-target "/usr/lib/drbd/snapshot-resync-target-lvm.sh -p 15 -- -c 16k";
            # after-resync-target /usr/lib/drbd/unsnapshot-resync-target-lvm.sh;
        }

        startup {
            # wfc-timeout degr-wfc-timeout outdated-wfc-timeout wait-after-sb
            wfc-timeout 15;
            degr-wfc-timeout 60;
        }

        disk {
            # on-io-error fencing use-bmbv no-disk-barrier no-disk-flushes
            # no-disk-drain no-md-flushes max-bio-bvecs
        }

        net {
            # sndbuf-size rcvbuf-size timeout connect-int ping-int ping-timeout max-buffers
            # max-epoch-size ko-count allow-two-primaries cram-hmac-alg shared-secret
            # after-sb-0pri after-sb-1pri after-sb-2pri data-integrity-alg no-tcp-cork
            cram-hmac-alg sha1;
        }

        syncer {
            # rate after al-extents use-rle cpu-mask verify-alg csums-alg
            rate 4M;
        }
    }

Configure DR:BD resource settings for mysql in /usr/local/etc/drbd.d/mysql.res:

# vi /usr/local/etc/drbd.d/mysql.res

    resource mysql {
        net {
            shared-secret "DONTTELL!";
        }
        on alice {
            device /dev/drbd0;
            disk /dev/xvdc;
            address ipv6 [ipv6:a]:7801;
            meta-disk internal;
        }
        on gertrude {
            device /dev/drbd0;
            disk /dev/xvdc;
            address ipv6 [ipv6:g]:7801;
            meta-disk internal;
        }
    }

And create resource settings for www in /usr/local/etc/drbd.d/www.res:

# vi /usr/local/etc/drbd.d/www.res

    resource www {
        net {
            shared-secret "DON'T TELL!";
        }
        on alice {
            device /dev/drbd1;
            disk /dev/xvdd;
            address ipv6 [ipv6:a]:7802;
            meta-disk internal;
        }
        on gertrude {
            device /dev/drbd1;
            disk /dev/xvdd;
            address ipv6 [ipv6:g]:7802;
            meta-disk internal;
        }
    }

On this setup, drbd expects configuration files to be in /usr/local/etc/, and won’t recognize them in /etc/. So we’ve put all our new stuff in the right place. Now get rid of the old. BUT, the cluster resource manager IS going to want them in /etc/, so we’ll create symbolic links.

# rm -rd /etc/drbd.*
# ln -s /usr/local/etc/drbd.conf /etc/drbd.conf
# ln -s /usr/local/etc/drbd.d /etc/drbd.d

Start up DR:BD

Create metadata for the devices (only need to do this on one host):

# drbdadm create-md mysql
# drbdadm create-md www

Modify the init script for drbd to accomodate a problem where IPv6 might not be ready when drbd starts:

# vi /etc/init.d/drbd # Add a sleep to the top of the script

    ### END INIT INFO
    sleep 5

On each server, at roughly the same time, start drbd:

# /etc/init.d/drbd start

On the server that will be primarily a database server:

# drbdadm -- --overwrite-data-of-peer primary mysql
# drbdadm disconnect mysql
# drbdadm connect mysql
# drbdadm disconnect www
# drbdadm connect www

On the server that will be primarily a web application server:

# drbdadm -- --overwrite-data-of-peer primary www
# drbdadm disconnect www
# drbdadm connect www
# drbdadm disconnect mysql
# drbdadm connect mysql

On either or both servers, check to make sure the primaries are primary, the secondaries are secondary, and the sync is syncing:

# cat /proc/drbd

version: 8.3.10 (api:88/proto:86-96)
built-in
 0: cs:SyncSource ro:Primary/Secondary ds:UpToDate/Inconsistent C r-----
    ns:3116484 nr:0 dw:0 dr:3593516 al:0 bm:219 lo:0 pe:2 ua:1 ap:0 ep:1 wo:f oos:1649980
    [============>.......] sync'ed: 65.5% (1608/4652)Mfinish: 0:05:08 speed: 5,328 (5,056) K/sec
 1: cs:SyncTarget ro:Secondary/Primary ds:Inconsistent/UpToDate C r-----
    ns:0 nr:2976256 dw:2976256 dr:0 al:0 bm:181 lo:0 pe:0 ua:0 ap:0 ep:1 wo:f oos:7509148
    [====>...............] sync'ed: 28.4% (7332/10236)Mfinish: 0:24:20 speed: 5,136 (5,060) want: 4,096 K/sec

Format the DR:BD devices, each on the server that is primary for that device. On the database server:

# apt-get install jfsutils
# mkfs.jfs /dev/drbd0

And on the web application server:

# apt-get install jfsutils
# mkfs.ext4 /dev/drbd1

On both servers, create the mount points:

# mkdir /var/www
# mkdir /var/mysql

# vi /etc/fstab // Add two lines

    /dev/drbd/by-res/www    /var/www        ext4    noauto,noatime
    /dev/drbd/by-res/mysql  /var/mysql      jfs     noauto

Mount the drives, each on it’s primary server:

# mount /dev/drbd/by-res/mysql /var/mysql/
# mount /dev/drbd/by-res/www /var/www

And there you have it - RAID 1 drives over a network. Failover comes later, after we install and configure the server software.

Services: MySQL and Lighttpd, plus a few other little things I like to have

Install stuff. Not too much stuff. Just the right stuff.

# apt-get install bc whois apg s3cmd lsof traceroute exim4 mailutils mutt \
    aspell-doc wamerican spellutils ispell mysql-server mysql-client lighttpd \
    lighttpd-doc lighttpd-mod-magnet lua5.1 php5-cli php5-cgi php5 php5-mcrypt \
    php5-mhash php5-gd php5-mysql php5-imagick php5-curl php5-memcache php5-ps \
    php5-pspell php5-tidy php5-xmlrpc php5-xsl libgd-tools libmagickcore3-extra \
    libmcrypt-dev mcrypt rrdtool

# dpkg-reconfigure exim4-config

Configure MySQL.

Edit /etc/mysql/my.cnf:

# vi /etc/mysql/my.cnf

    [client]
    port            = 3306
    socket          = /var/run/mysqld/mysqld.sock

    [mysqld_safe]
    socket          = /var/run/mysqld/mysqld.sock
    nice            = 0

    [mysqld]
    user            = mysql
    socket          = /var/run/mysqld/mysqld.sock
    port            = 3306
    basedir         = /usr
    datadir         = /var/mysql
    tmpdir          = /tmp
    skip-external-locking
    bind-address            = priv.f.f.f
    key_buffer              = 16M
    max_allowed_packet      = 16M
    thread_stack            = 192K
    thread_cache_size       = 8
    myisam-recover         = BACKUP
    #max_connections        = 100
    #table_cache            = 64
    #thread_concurrency     = 10
    query_cache_limit       = 1M
    query_cache_size        = 16M
    #general_log_file        = /var/log/mysql/mysql.log
    #general_log             = 1
    log_error                = /var/log/mysql/error.log
    #server-id              = 1
    #log_bin                        = /var/log/mysql/mysql-bin.log
    expire_logs_days        = 10
    max_binlog_size         = 100M
    #binlog_do_db           = include_database_name
    #binlog_ignore_db       = include_database_name
    # ssl-ca=/etc/mysql/cacert.pem
    # ssl-cert=/etc/mysql/server-cert.pem
    # ssl-key=/etc/mysql/server-key.pem

    [mysqldump]
    quick
    quote-names
    max_allowed_packet      = 16M

    [mysql]
    #no-auto-rehash # faster start of mysql but no tab completition

    [isamchk]
    key_buffer              = 16M

    !includedir /etc/mysql/conf.d/

We changed the default directory for MySQL, so we need to move it’s files into the correct directory.

# stop mysql
# chown mysql:mysql /var/mysql
# cp -Rp /var/lib/mysql/* /var/mysql/
# ifconfig eth0:1 priv.f.f.f netmask 255.255.128.0
# start mysql

Sync these settings with the other server (both need the same configuration so that either can run the service off the same data).

# scp -r /etc/mysql root@gertrude:/etc/

Configure Lighttpd. (Substitute other instructions for nginx, apache, etc.)

Edit /etc/lighttpd/lighttpd.conf on the web application server:

# vi /etc/lighttpd/lighttpd.conf

    server.modules = (
        "mod_access",
        "mod_alias",
        "mod_compress",
        "mod_redirect",
            "mod_rewrite",
    )

    server.document-root        = "/var/www"
    server.upload-dirs          = ( "/var/www/cache/uploads" )
    server.errorlog             = "/var/www/logs/error.log"
    server.pid-file             = "/var/run/lighttpd.pid"
    server.username             = "www-data"
    server.groupname            = "www-data"

    index-file.names            = ( "index.php", "index.html")

    url.access-deny             = ( "~", ".inc" )

    static-file.exclude-extensions = ( ".php", ".pl", ".fcgi" )

    ## Use ipv6 if available
    include_shell "/usr/share/lighttpd/use-ipv6.pl"

    dir-listing.encoding        = "utf-8"
    server.dir-listing          = "enable"

    compress.cache-dir          = "/var/www/cache/compress/"
    compress.filetype           = ( "application/x-javascript", "text/css", "text/html", "text/plain" )

    include_shell "/usr/share/lighttpd/create-mime.assign.pl"
    include_shell "/usr/share/lighttpd/include-conf-enabled.pl"

    magnet.attract-physical-path-to = ( "/etc/lighttpd/modx.lua" )

Enable some more modules:

# lighttpd-enable-mod fastcgi
# lighttpd-enable-mod simple-vhost
# lighttpd-enable-mod accesslog
# lighttpd-enable-mod magnet
# lighttpd-enable-mod status

Edit more config files:

# vi /etc/lighttpd/conf-enabled/10-accesslog.conf

    server.modules += ( "mod_accesslog" )
    accesslog.filename = "/var/www/logs/access.log"

# vi /etc/lighttpd/conf-enabled/10-simple-vhost.conf

    server.modules += ( "mod_simple_vhost" )
    simple-vhost.server-root         = "/var/www/servers/"
    simple-vhost.document-root       = "htdocs"
    simple-vhost.default-host        = "{hostname}"

# vi /etc/lighttpd/conf-enabled/10-fastcgi.conf

    server.modules += ( "mod_fastcgi" )
    fastcgi.server    = ( ".php" =>
            ((
                    "bin-path" => "/usr/bin/php-cgi",
                    "socket" => "/tmp/php.socket",
                    "max-procs" => 2,
                    "idle-timeout" => 20,
                    "bin-environment" => (
                            "PHP_FCGI_CHILDREN" => "4",
                            "PHP_FCGI_MAX_REQUESTS" => "10000"
                    ),
                    "bin-copy-environment" => (
                            "PATH", "SHELL", "USER"
                    ),
                    "broken-scriptfilename" => "enable"
            ))
    )

# vi /etc/lighttpd/modx.lua

    attr = lighty.stat(lighty.env["physical.doc-root"] .. "manager/includes/config.inc.php") -- Appears to be a ModX site
    if (attr) then
      attr = lighty.stat(lighty.env["physical.path"])
      if (not attr) then -- Requested resource doesn't exist in the file system
        path = "/index.php"
        uri = lighty.env["request.uri"]
        uri2 = string.gsub(lighty.env["request.uri"], "\?", "\&")
        -- print("Original request.uri: " .. uri .. " Replaced with: " .. uri2)
        lighty.env["uri.query"] = "q=" .. string.gsub(uri, "\?", "\&")
        lighty.env["uri.path"] = path
        lighty.env["request.uri"] = path .. "?" .. lighty.env["uri.query"]
        -- print("New request.uri: " .. lighty.env["request.uri"])
        lighty.env["physical.rel-path"] = path
        lighty.env["physical.path"] = lighty.env["physical.doc-root"] .. string.sub(lighty.env["physical.rel-path"], 2)
        -- print("New physical.path: " .. lighty.env["physical.path"])
      end
    end

# vi /etc/lighttpd/wp-rewrite.conf

    # Use when a site has a blog
    # Example:
    # $HTTP["host"] =~ "www\.example\.com" {
    #   var.wpdir = "/blog/"
    #   include "wp-rewrite.conf"
    # }

    url.rewrite-once = (
      "^" + wpdir + "(wp-.+).*/?" => "$0",
      "^" + wpdir + "(sitemap.xml)" => "$0",
      "^" + wpdir + "(xmlrpc.php)" => "$0",
      "^" + wpdir + "keyword/([A-Za-z_0-9-])/?$" => wpdir + "index.php?keyword=$1",
      "^" + wpdir + "(.+)/?$" => wpdir + "index.php/$1"
    )

Edit /etc/php5/cgi/php.ini so that these lines are correct:

display_errors = stderr
error_log = /var/www/logs/php_errors.log

Create some directories we just specified, but don’t yet have:

# mkdir /var/www/run /var/www/cache /var/www/cache/compress /var/www/cache/uploads /var/www/logs /var/www/servers /var/www/servers/{hostname} /var/www/servers/{hostname}/htdocs
# chown -R www-data:www-data /var/www/run /var/www/cache /var/www/logs/ /var/www/servers

Restart the local server and sync these settings with the other server (both need the same configuration so that either can run the service off the same data).

# /etc/init.d/lighttpd restart
# scp -r /etc/php5 /etc/lighttpd root@alice:/etc/

Test that either server can run either service.

On the database server:

# stop mysql
# ifconfig eth0:1 down
# umount /var/mysql/
# drbdadm secondary mysql

On the web application server:

# drbdadm primary mysql
# mount /dev/drbd0 /var/mysql/
# ifconfig eth0:1 priv.f.f.f netmask 255.255.128.0
# start mysql
# mysql -p
# stop mysql
# ifconfig eth0:1 down
# umount /var/mysql/
# drbdadm secondary mysql
# /etc/init.d/lighttpd stop
# umount /var/www
# drbdadm secondary www

On the database server:

# drbdadm primary mysql
# drbdadm primary www
# mount /dev/drbd0 /var/mysql/
# mount /dev/drbd1 /var/www/
# /etc/init.d/lighttpd start
# ps ax | grep light
# /etc/init.d/lighttpd stop
# umount /var/www
# drbdadm secondary www
# ifconfig eth0:1 priv.f.f.f 255.255.128.0
# start mysql

On the web application server:

# drbdadm primary www
# mount /dev/drbd1 /var/www
# /etc/init.d/lighttpd start

Configure a failover cluster

Just a note here before we start. I would have liked to run corosync⁵ rather than heartbeat⁶ based solely on my understanding of pacemaker⁷, how and why it split from heartbeat. However, corosync seems to mostly require multicast networking to work⁸ and after some struggles, I’ve learned that Linode doesn’t support multicast.

Set up heartbeat and pacemaker

Install and configure software on both servers

# apt-get install heartbeat pacemaker

Edit /usr/lib/ocf/resource.d/heartbeat/Filesystem to comment out this whole block. This will fix a problem with the script’s handling of jfs filesystems. Besides, they say right in the code, “Why should a filesystem resource agent magically load a kernel module?” I agree. Lemme handle that part and just mount the drive please.

# vi /usr/lib/ocf/resource.d/heartbeat/Filesystem

    #       if [ "X${HOSTOS}" != "XOpenBSD" ];then
    #               # Insert SCSI module
    #               # TODO: This probably should go away. Why should the filesystem
    #               # RA magically load a kernel module?
    #               $MODPROBE scsi_hostadapter >/dev/null 
    #
    #               if [ -z "$FSTYPE" -o "$FSTYPE" = none ]; then
    #                       : No FSTYPE specified, rely on the system has the right file-system support already 
    #               else
    #                       # Insert Filesystem module
    #                       $MODPROBE $FSTYPE >/dev/null 
    #                       grep -e "$FSTYPE"'$' /proc/filesystems >/dev/null
    #                       if [ $? -ne 0 ] ; then
    #                               ocf_log err "Couldn't find filesystem $FSTYPE in /proc/filesystems"
    #                               return $OCF_ERR_INSTALLED
    #                       fi
    #               fi
    #       fi

Configuration files:

# vi /etc/ha.d/authkeys

    auth 1
    1 sha1 Don'tTell

# chmod 600 /etc/heartbeat/authkeys

# vi /etc/ha.d/ha.cf

    autojoin none
    logfacility daemon
    keepalive 2
    deadtime 15
    warntime 5
    initdead 120
    udpport 694
    ucast eth0 priv.a.a.a
    ucast eth0 priv.g.g.g
    node alice
    node gertrude
    auto_failback on
    use_logd yes
    crm respawn

Propogate the configuration and restart each server

# scp -r /etc/ha.d/* root@gertrude:/etc/ha.d/

# /etc/init.d/heartbeat restart

Disable automatic start of Lighttpd (using LSB) and MySQL (using Upstart)

# update-rc.d lighttpd disable
# vi /etc/init/mysql.conf

    # Comment out the startup
    #start on (net-device-up
    #          and local-filesystems
    #         and runlevel [2345])

Configure the resources

# vi /usr/lib/ocf/resource.d/linbit/drbd

    OCF_RESKEY_drbdconf_default="/usr/local/etc/drbd.conf"

# crm configure

    primitive db_alert ocf:heartbeat:MailTo \
        params email="root" subject="(db)"
    primitive db_drbd ocf:linbit:drbd \
        params drbd_resource="mysql" \
        op start interval="0" timeout="240" \
        op stop interval="0" timeout="100"
    primitive db_fs ocf:heartbeat:Filesystem \
        params device="/dev/drbd0" directory="/var/mysql" fstype="jfs" \
        op start interval="0" timeout="60" \
        op stop interval="0" timeout="120"
    primitive db_ip ocf:heartbeat:IPaddr \
        params ip="priv.f.f.f" cidr_netmask="24"
    primitive db_mysql lsb:mysql \
        op monitor interval="0" enabled="false"

    primitive www_alert ocf:heartbeat:MailTo \
        params email="root" subject="(www)"
    primitive www_drbd ocf:linbit:drbd \
        params drbd_resource="www" \
        op start interval="0" timeout="240" \
        op stop interval="0" timeout="100"
    primitive www_fs ocf:heartbeat:Filesystem \
        params device="/dev/drbd1" directory="/var/www" fstype="ext4" \
        op start interval="0" timeout="60" \
        op stop interval="0" timeout="120"
    primitive www_ip ocf:heartbeat:IPaddr \
        params ip="pub.f.f.f" cidr_netmask="24"
    primitive www_lighty lsb:lighttpd \
        op monitor interval="0" enabled="false"

    group db db_ip db_fs db_mysql db_alert \
        meta target-role="Started"

    group www www_fs www_ip www_lighty www_alert \
        meta target-role="Started"

    ms ms_db_drbd db_drbd \
        meta master-max="1" master-node-max="1" clone-max="2" clone-node-max="1" notify="true"

    ms ms_www_drbd www_drbd \
        meta master-max="1" master-node-max="1" clone-max="2" clone-node-max="1" notify="true"

    location db_gertrude db 10: gertrude
    location db_alice db 100: alice
    location www_gertrude www 100: gertrude
    location www_alice www 10: alice

    colocation db_on_drbd inf: db ms_db_drbd:Master
    colocation www_on_drbd inf: www ms_www_drbd:Master

    order db_after_drbd inf: ms_db_drbd:promote db:start
    order www_after_drbd inf: ms_www_drbd:promote www:start

    property cluster-infrastructure="Heartbeat" \
        stonith-enabled="false" \
        no-quorum-policy="ignore" \
        default-resource-stickiness="5"

A new OpenERP product and license

2011-06-27T20:03:00Z

Last Friday (June 24, 2011) under an unassuming headline ‘Improved OpenERP Website’,¹ OpenERP announced the new Enterprise edition of the OpenERP software. The response among community members was swift. Two topics attracted most of the conversation: the new OpenERP AGPL + Private Use license² available only to Enterprise edition customers, and apparent policy changes regarding security alerts.

The OpenERP AGPL + Private Use license

Essentially, OpenERP SA has introduced a dual licensing model for the OpenERP program.

Download the community version and the code is strictly licensed under the GNU Affero Public License³, which is much like the GNU General Public License⁴ except that it closes what is sometimes called the “Application Service Provider Loophole”.
Subscribers of the OpenERP Enterprise Edition receive much the same code as the community users, but they are conveyed the right to create “private modules”. These modules are meant for companies that need to include confidential data and/or business practices in their customizations to the program. This second, private license grants such users an exception to the AGPL so that they are not required to distribute that sensitive code to users of the system, but only under very limited circumstances.

I have no doubt this additional license addresses a need or meets a demand among some users of OpenERP. Certainly I can see how it might desirable, or even necessary in some applications. Nonetheless, members of the OpenERP community raised serious concerns about this announcement.

The right to license

OpenERP SA owns the rights to a significant portion of the source code for the OpenERP system. However, portions of the core code for OpenERP, and the vast majority of the 1,200 modules for OpenERP, were written by developers who have not handed their rights over to OpenERP SA. As such, these developers have the right to release their code as they choose. In fact, they have exercised that right in releasing their code under the GNU AGPL license, which is why their modules can be distributed along with the rest of the OpenERP system.

For clarity I should note that all OpenERP modules must be licensed under the AGPL, as they are derivative works of OpenERP, which is released under the AGPL license. (There is an exception for GPL licensed code; more below.) This isn’t to say that anyone anywhere should have access to the source code for any custom OpenERP module ever written, but because OpenERP itself is released under the AGPL, the users of any given OpenERP system have the right to the source code for the entire system including the custom modules.

It is unclear at this point exactly what code OpenERP SA will be offering under this new dual license. It could be as described above – only including core code and modules to which OpenERP owns all the rights. It’s my opinion (and I am not a lawyer) that they could license a system comprised only of OpenERP SA code under any license they wish. But some community members fear it will include work they created, relicensed without their permission.

One basis for this concern is that OpenERP SA code, without modules and base code contributed by community developers, may be insufficient for building an effective ERP system. If this is the case, every Enterprise edition installation will include community code.

Another basis for this concern might be drawn from the new dual license itself. In reviewing the text of their OpenERP AGPL + Private Use license, some of the terms seem to include code created by the community.

Under section 0, Additional Definitions, the license defines Program to include all components without limitation (emphasis is mine):

In the context of this License, the Program refers to OpenERP including all its components: client programs, server system, modules, etc. “OpenERP modules” are extensions to the Program that can be activated separately.

Under section 1, Exceptions to section 13 of the GNU AGPL, describing the conditions required for the exception, the license implies that OpenERP SA is the original licensor of the program defined to include all components in section 0 (again, emphasis is mine):

a. You have received written permission to do so from the original licensor of the Program, OpenERP S.A.

b. You do not convey the covered work.

One could read this as OpenERP SA claiming licensing rights to code written by the community, and several developers in the community interpreted the announcement in exactly this way.

Compatible licenses

Under the terms of the dual license, nearly all of the code distributed to an Enterprise edition subscriber is licensed AGPL. The subscriber then has the option of creating a derived work (in the form of a module that extends the functionality of the system) that is not open source. Under the GPL this is clearly allowed. The Free Software Foundation has made this clear in the Frequently Asked Questions about the GNU Licenses:⁵ ”The GPL does not require you to release your modified version, or any part of it. You are free to make modifications and use them privately, without ever releasing them.” In fact, OpenERP’s language even parallels that of the Free Software Foundation in terming the license “AGPL + Private Use”.

If OpenERP were released under the GPL (as it used to be), private use of modified source code could cover Application Service Providers offering the program as a service over a network, and then the provider would not be required to offer their customers the source code for the program. OpenERP base code and all modules, however, are released under the Affero GPL license that plugs the Application Service Provider loophole. Users of the system, even those accessing it as an online service, must be prominently offered the opportunity to download all corresponding source code.

I mentioned briefly above that there is the option to combine another (possibly non-derivative) work licensed under the GPL with OpenERP licensed under the AGPL. The terms of the AGPL suggest the GPL portion of the combined work would be covered under the GPL license. However, Section 13 of the GPL clearly states that “the special requirements of the GNU Affero General Public License, section 13, concerning interaction through a network will apply to the combination as such.” So even combined with the explicitly compatible GPL license, the more restrictive (from the publishers perspective) requirement to prominently offer source code applies, even to the GPL code.

Why it matters

In many cases, an OpenERP installation with base and community-developed modules will be sufficient all by itself. No additional modules are needed, let alone custom modules containing hard-coded sensitive business intelligence. In cases where custom modules are required, much or all of the sensitive business intelligence can or should be integrated as data, not program code. (Pricelists, for instance, are data and under normal circumstances would never need to be distributed with the source code for a program.) In the rare circumstances where business methods are implemented by a module, the source code must be available to the company that performed (or hired) the modifications, and their users.

Those final three words, “and their users”, contain the key to understanding the change. The GPL was created in large part to protect users by ensuring they have access to the source code of the programs on which they depend, and some companies would not wish employees and other system users to have access to the source code.

The AGPL extends that protection to include another class of users, those that access the system across a network. In this case, that will include users who subscribe to OpenERP under a Software as a Service (SaaS)⁶ model. Under the AGPL, no company (including OpenERP SA) could offer OpenERP as a hosted service without offering links to download their service’s source code, including custom modules. Under the dual license, only the code which is distributed (sold or given away to other companies) must include the option to download source.

SaaS is a powerful and lucrative force in the market. OpenERP SA itself is promoting their online service at 39€ per month per user⁷. In the US, a company of merely twenty employees would be paying $11,760.00 a year for the service. Under the AGPL, OpenERP SA is required to allow me (assuming I am an online user) to download the source code, including the source code for any modules they’ve written exclusively for their online service, which I may then use to create a competing service. Under the dual license, for some amount of money starting at 1,950€ per year, I can create a custom module for SaaS and not offer the source code to my users. Or, if I’m OpenERP, perhaps I can just do it for free.

Community Relations

In the two years I’ve been working with OpenERP, I’ve watched OpenERP SA sink in the estimation of many community members and developers. The release of the Enterprise edition with a dual license may have created a new low.

(Update 23 August 2011: This tweet has been removed from Fabien Pinckaer’s Twitter account, so if you click on the image, you’ll see a Twitter 404 page.)

I first heard of the dual license possibility from the transcript of a Certified Training Partner (CTP) meeting held on IRC in May. (I am not a CTP, but a link to the transcript was provided by an attendee.) Some partners in the meeting expressed considerable opposition to the idea at that time, and Fabien Pinckaers made what many would consider an unequivocable statement on May 26th that there would be no dual-license. Nonetheless, the June 24 announcement indicates Enterprise subscribers will have access to the new dual license.

It could be argued that Fabien’s tweet has not been violated:

The new license is a modified AGPL, granting additional permissions in limited circumstances.
Only undistributed custom modules qualify for the private use license. For the rest of the software, the license is modified from AGPL only enough to allow for those private modules to be installed with the system.
Fabien didn’t actually say they wouldn’t dual license, only that they didn’t have to. (Okay, this one feels like a stretch.)

I haven’t heard these arguments presented by anyone in the community, but my impression is they are not reassuring. My own employer tweeted last Friday:

Security Alerts

On the OpenERP ‘Buy’ page⁸ showing the options for purchase and download, the presence of ‘Security Alerts’ under OpenERP Enterprise, and only under Enterprise, attracted conversation.

@SharoonThomas:

Acc to @openerp community users will not get security alerts. Not needed since community has always found security holes, not openerp SA

@SharoonThomas:

@openerp how dare u talk about security alerts when u dont know why software needs encrypted passwords & not plain text?? http://j.mp/lffPGi

@LaGeekitude:

Security alerts should be public announcements. All users deserve notice. @OpenERP @fpopenerp

Upon closer review, it’s apparent that OpenERP only promises that Enterprise customers will receive notices in advance of a public disclosure. This is detailed on the multiple product pages describing OpenERP Enterprise, such as the page describing the 70 to 150 user license⁹ (emphasis is mine):

If we detect a security issue on OpenERP we develop the patch and warn our customers three weeks before a public announcement. We also provide you with the explanation on how to migrate or apply the security patch. After three weeks, we do a public communication about the security hole for the rest of the community. We strongly advice you to apply the patch to your server within these three weeks to avoid any security trouble.

Under the OpenERP Publisher’s Warranty, the promise regarding security alerts was even less favorable to community users (again, emphasis is mine):

As OpenERP manages critical data for your company, it is important to be very proactive in case of a vulnerability issue detection in the OpenERP software. Once a security bug is detected by OpenERP sa or by the community, we provide you with a patch to fix the issue one month before the public announce of the security fix. This allows you to update your installation before the public announce and release of the patch.

Since the new Enterprise offerings replace the OpenERP Publisher’s Warranty, it seems likely that the Enterprise policy on security alerts will become the current practice. In effect, this is an improvement over the previous practice.

OpenERP SA and the commercial open source business model

I can certainly relate to the community of developers who are concerned and frustrated by OpenERP SA’s announcement. Some of the developers have devoted literally months of their time to make OpenERP a better product, and have reason to feel used, even abused, by liberties that may be (or have been) taken with their hard work. Because of this, I’ve tried to discuss the situation objectively, and I hope I’ve demonstrated that the situation is not quite as bad as it seemed at first. There are unanswered questions, and I trust OpenERP SA will answer those through further communication and their future actions.

It is important to consider that OpenERP SA is working to cement their commercial open source business model. We know what to expect: a difficult transition, especially for early adopters and contributors as more of the product is brought under the company’s control. They are master editor of the software, selling services to end users as well as companies offering services related to their software. With persistence and planning, OpenERP SA and many of their partners will enjoy a profitable future.

One freedom often ascribed to the GPL is the freedom to not use the software. Don’t like the fact you have to release your source code? Don’t use it. Don’t like the direction the project is going? Fork it.

Please don’t hesitate to do your own investigation about the new OpenERP products and licenses. Visit their website¹⁰, especially their FAQ on the changes¹¹ and community section¹² for information on all of their communication channels.

In related news…

Nicolas Évrard of B2CK¹³ posted to the Tryton mailing list¹⁴ last Saturday a call for comments on the creation of the Tryton Software Foundation¹⁵. B2CK proposes this foundation as a consensus-based community formed to promote and protect the freedom of Tryton. As Tryton and OpenERP share a common lineage (both can be traced back to TinyERP, each with improvements in different areas), OpenERP users and contributors frustrated by OpenERP’s business model may find a more favorable environment among Tryton’s community open source model.

I have been writing a series of articles on Tryton on this blog; consult them or visit Tryton’s website¹⁶ for more information.

Updated 28 June 2011

There has been remarkable conversation on Twitter regarding this post, and I’m grateful for all the feedback. I’d especially like to thank Olivier Dony (@odony on Twitter), Community Manager at OpenERP SA, for his clarifications on some of the points I’ve raised.

The biggest comes from this tweet, which @ravlyi described as “the missing OpenERP re-licensing FAQ”:

This certainly seems to make clear that no Enterprise edition customer using private modules may install any AGPL-only modules. It doesn’t answer concerns about what happens if an Enterprise edition subscriber installs a community module, or about community contributions to OpenERP core code, such as the netsvc.py¹⁷ file @cedrickrier tweeted from OpenERP-server trunk, as excerpted here:

…  
5 \# OpenERP, Open Source Management Solution  
6 \# Copyright (C) 2004-2009 Tiny SPRL (). All Rights Reserved  
7 \# The refactoring about the OpenSSL support come from Tryton  
8 \# Copyright (C) 2007-2009 Cédric Krier.  
9 \# Copyright (C) 2007-2009 Bertrand Chenal.  
10 \# Copyright (C) 2008 B2CK SPRL.  
…

Also, one aspect I did not address in the original post was OpenERP’s original transition from GPL to AGPL. Fabien Pinckaers discussed it on his Blogspot blog²⁰ and reviewing this clears any question about his thoughts at that time on SaaS and OpenERP:

“Our goal is to better promote and maintain the free and open source nature of OpenERP. We think this licence will better protect the community to ‘evil’ SaaS offers that do not want to release their source code to the community.”

It’s my opinion (and again, I am not a lawyer) that the new AGPL+Private enables the ‘evil SaaS offers’ Fabien warns about, so long as the SaaS provider is either a paying OpenERP Enterprise subscriber, or OpenERP SA themselves. I am in no way saying it is OpenERP SA’s intention to support a secondary SaaS market based on private modules, or to create private modules for their own SaaS product. I merely point at the license and say I think it could be done under these terms.

Updated 23 August 2011

I was doing maintenance on this post (moving it to the version2beta.com domain with the rest of my blog) and came across some discrepencies between linked documents when I wrote the post, and how those documents appear today. For example, the netsvc.py¹⁷ discussion above: Cédric Krier’s copyright no longer appears in the header in the trunk source code. I researched the revision and found this:¹⁸

Committer: Olivier Dony
Date: 2011-06-29 09:51:38
Revision ID: odo@openerp.com-20110629095138-hmo91mvk1m8u4643
[FIX] netsvc: corrected copyright header

Corrected license to AGPLv3, and removed obsolete copyright lines
as confirmed by Cedric Krier by email on 2011-06-29.
These copyright lines referred to a refactoring copied from Tryton
in revision rev 1439 - stephane@tinyerp.com-20081218235433-cnsc5a4iyfzqvbdx
All this code has disappeared since then, specifically after
a major rewrite by xrg, merged at revision 1900 
(mga@tinyerp.com-20091126113653-0ym60nvtjinvp82h).

This was reassuring, that the copyright notice was in error. I believe it is better that my research into the issue exposed an oversight rather than a disagreement over who owns the code. According to Oliver Dony’s forum post two weeks later, all such issues are resolved:¹⁹

…

OpenERP SA does own the copyright to the server, clients, and all modules whose author is OpenERP/Tiny. And we are exclusively offering the AGPL+PrivateUse option for those pieces of code, and nothing else. If someone believes they own copyright on a contribution within that specific set of source code lines, and are against the AGPL+PrivateUse license, they can contact us, and we’ll find a solution. We’ve only seen one such claim so far (from C.Krier), and it was about code that was long gone from our source, so there was no problem. So if you are such a person, please contact us.

…

Again, this claim is reassuring. However, I understand these are not the only items in question or the only copyright that has been contested. Here are three more sections of code with conflicted copyright.

In the OpenERP client, widget/view/list.py:²¹

…
4 #    OpenERP, Open Source Management Solution
5 #    Copyright (C) 2004-2010 Tiny SPRL (<http://tiny.be>). All Rights Reserved
6 #    Copyright (c) 2008-2009 B2CK, Bertrand Chenal, Cedric Krier (D&D in lists)
…

In the OpenERP base_vat module:²²

…
4 #    OpenERP, Open Source Management Solution
5 #    Copyright (C) 2004-2010 Tiny SPRL (<http://tiny.be>). All Rights Reserved
6 #    Copyright (C) 2008-2009 B2CK, Cedric Krier, Bertrand Chenal (the methods "check_vat_[a-z]{2}"
…

And in the webdav module:²³

…
4 #    OpenERP, Open Source Management Solution
5 #    Copyright (C) 2004-2010 Tiny SPRL (<http://tiny.be>).
6 #    Copyright (c) 1999 Christian Scholz (ruebe@aachen.heimat.de)
…

Without a complete audit of the source code, all such copyright discrepencies may not be recognized, and it would seem to me that such an audit should occur before the product is relicensed.

One last note: Several people have raised the question whether modules in OpenERP (or, in two cases, themes in Wordpress - a similar but not identical question) are in fact derivative works that must be licensed AGPL. This is a big question, of course, and deserves more than merely a mention in an addendum to a months-old blog post. Here I’d like to point out that I make the statement in my original post that modules must be licensed AGPL, and that this statement is debatable. I hope to question that statement in a future blog post.

Tryton as an e-commerce back-end - Stalking the Tryton community

2011-06-17T12:58:00Z

This is part two in a series exploring how Tryton might fair as the heavy-lifting portion of an e-commerce package. This part discusses some of my experiences with members of the Tryton community.

The whole series, at least as planned:

Part 1: About Tryton
Part 2: Stalking the Tryton community
Part 3: The crux of the matter

The Tryton community

My first experience with Tryton was two years ago, when we were evaluating open source ERP packages for a business systems integration project. I came, I saw, and I moved on to OpenERP, which offered more features and centralized commercial support. Now that we have considerably more experience with OpenERP, I’ve come back for a closer look at Tryton.

I started with the community, and I suspect that is a very valid approach. I’m looking for a long term relationship. I wanted to meet the family. So naturally I hid behind some bushes and watched them for a month or so. Some of the things I noticed:

They don’t bicker very much. Like, hardly at all. Most of the communication I’ve seen is in non-native English, and they speak carefully so as not to be misunderstood.
Mostly they are proactive, identifying potential problems and solutions before people become invested in various outcomes.
They air their laundry (bugs, issues, etc) as well as most open source projects. It seems they don’t have as much laundry to air, though.

The Tryton family has some key members. I’ve identified some of them from looking at the services page on the website¹, some from the Google groups for Tryton² and Tryton-dev³, and some on IRC⁴. Many of these people I also recognized from the OpenERP community, as there’s considerable overlap. I’m reluctant to point names and name fingers (with a nod to Jack Sparrow⁵) as I’m far too likely to miss someone’s most excellent contribution, and that would be an unfortunate oversight. Kudos and gratitude to everyone who has worked on Tryton. Despite my reluctance, I still want to point out a few people.

There is no doubt a reason why the Tryton.org website⁶ looks so much like B2CK Software Development’s website⁷. I’ve read hundreds of emails on the mailing lists from Cédric Krier, Nicolas Évrard, and others in this company. Clearly these are key people in Tryton development and I cannot say enough nice things. Cédric is a programming god: targeted, responsive, attentive, and sharp. I greatly enjoyed the night Udo Spallek recommended two accessibility improvements to the Tryton GUI and Cédric replied 52 minutes later with the one word “Done”.

I already had a healthy respect (perhaps even a geek-crush) on Sharoon Thomas⁸, CEO of Openlabs in the UK and India⁹. Sharoon and Openlabs have written about using Tryton as a Python module in Django¹⁰, integrated UPS international shipping with OpenERP¹¹, and built Callisto e-commerce for OpenERP¹² as well as Nereid for Tryton¹³. Sharoon has clearly already considered the question of Tryton and e-commerce, as indicated in his email list discussion on building a Magento Connector for Tryton¹⁴. (My favorite part: “An integration with Magento does not add ‘The best e-commerce possibility for Tryton’ but does make ‘A great ERP backend for Magento’.”)

Again I want to add, there are others. I can just barely read French, German, and Spanish, and do not know the developers who spend most of their time in these channels. Even among the participants in the English language forums, I’m not representing everyone here, but I do wish to include them in my appreciation.

What else is there?

Part 1: About Tryton. An overview of Tryton – architecture, business model, features, license.
Part 2: Stalking the Tryton community. There are some impressive people working on this project. I’m pleased to get to know them.
Part 3: The crux of the matter. Tryton needs a web framework so we don’t have to do things twice.

Freaky shadow

2011-06-09T13:21:00Z

The webserver logs showed I was being followed on my path through a website by a computer at an IP address in Texas. It trailed two seconds behind me, on every page.

It was a dark and stormy day, and a strange thing happened while I was debugging some PHP code.

I was using [error_log][] to pass some variables and values to the standard Apache log, and tailing it to make sure PHP was seeing what I wanted it to see. That was when I saw something I didn’t expect to see.

[Thu Jun 09 13:37:54 2011] [error] [client 24.xxx.xx.xxx] Carrier: ...
[Thu Jun 09 13:37:54 2011] [error] [client 24.xxx.xx.xxx] Returning ...
[Thu Jun 09 13:37:56 2011] [error] [client 75.xxx.xx.xxx] Carrier: ...
[Thu Jun 09 13:37:56 2011] [error] [client 75.xxx.xx.xxx] Returning ...

Each time I hit the page, my code logged the access. Then, a second or two later, another entry from a different IP address was also logged.

Everything I did was being shadowed by a server in Texas.

It took a few minutes of research to discover that the IP address belongs to one of ClickTale’s servers, and that the page I was modifying runs their analytic code. Freakiness abated.

Tryton as an e-commerce back-end - About Tryton

2011-06-05T15:24:00Z

This is part one in a series exploring how Tryton might fair as the heavy-lifting portion of an e-commerce package. This first part describes Tryton - architecture, features, community, license.

Upcoming parts (as planned, anyway):

About Tryton

I want to call Tryton¹ an open source ERP² system, and it is, but I like their description of the project better: A three-tier³ high-level general purpose application platform released under the GPL-3⁴ license, written in Python⁵, using PostgreSQL⁶ as its database engine.

That’s a lot of descriptors, but basically what it says to me is that it’s a platform for building stuff, made from tech I like, and licensed under the GPLv3. And when you look at the base modules that come with Tryton, you can see it has an emphasis on ERP: accounting, invoicing, sales management, purchasing management, inventory management.

[Note to past self: In the future, you’re going to get excited about this stuff. No, not anxious. I said excited.]

[Note from past selves: Did you forget your first professional programming project - 1985 - was calculating economic order quantities? And your first industrial automation job - 1991 - counted feet of magnetic wire coming off 50 machines for integration with their business systems? Loser. We’re ready. You’re the anxious one.]

Architecture

So Tryton is made up of three main components (tiers): A client, a database, and some code stuck in the middle to make things go smoothly.

On the client side, we have a Python GTK+ program that runs on a bunch of desktop operating systems (unfortunately not yet including mobile.) But there’s lots of ways to look at what the client application is. In some ways, my e-commerce websites will be the client, too.
Opposite the client, we have PostgreSQL, a badass database powerhouse. ‘Nuff said.
In the middle comes a bunch of Python programming that does all sorts of cool stuff. From my perspective, it manages a bunch of objects (as in Object Oriented Programming⁷) like ‘Orders’ and ‘Persons’, and keeps most of them tucked safely away in the database for future reference. This is where the magic happens. It’s also where the growth happens, because Tryton is modular, which means you can create new objects, introduce Tryton to them, and wham-bam you get new features. For example, instead of blogging about Tryton right now I’m supposed to be building two new modules for Tryton’s cousin, OpenERP⁸, adding increased functionality for interfacing with UPS⁹ for shipping and CyberSource¹⁰ for payments.

That middle layer between the client and the database is also the reason Tryton isn’t just an ERP system, because really, you can use it to manage any data-intensive application. There’s nothing about Tryton that intrinsically says “You must have an Certified Public Accountant involved in this project” – Tryton’s design lets it manage schools and hospitals and music festivals and motorcycles and people and emails and whatnots just as easily procurement orders or P&L statements.

While the middle tier provides the magic, there’s an important consideration in the client, too. The client doesn’t have to be human. If I had my druthers, I’d no longer develop any web applications that don’t provide an API, or Application Programming Interface¹¹. Tryton’s server, known as trytond or “the Tryton daemon”, connects to the regular old client using a protocol known as XML-RPC¹², or “eXtensible Markup Language – Remote Procedure Call”. trytond uses XML-RPC to tell the client what to show, and the client responds in XML-RPC to tell the server what to do. Well, that means that any program that knows the secret handshake and how to speak XML-RPC can be a Tryton client. (For accuracy’s sake, I’d like to point out that Trytond is like, trilingual. It’s fluent in XML-RPC¹³, JSON-RPC¹⁴, and Net-RPC¹⁵. It’s also gets by with WebDAV¹⁶, psycopg¹⁷, and even some OpenOffice¹⁸ – you know, enough to ask where the bathroom is, how to hail a cab, what’s safe to order at a restaurant.) The upshot of this, of course, is that I can create an e-commerce site running on MODX¹⁹ Revolution²⁰ or Django²¹ that interfaces completely with Tryton because my website is the client.

Features

I feel like I should mention something about what Tryton actually does, which is an interesting concept. I almost think of it like I would a programming language – not what it does, but what you can do with it. Given enough time.

Despite my inclination to see multiple uses for it, Tryton focuses on business systems. Install the standard modules and it will handle the business of a catalog retailer, for example:

Manage contacts: customers, vendors, companies, employees, addresses, etc.
Workflow management
Issue RFQs (requests for quotation) and purchase orders to suppliers, keep track of current and historical costs associated with products
Receive stock, track inventory by location, move stock around and out the door, get the value of inventory by location, set orderpoints
Process invoices for accounts payable, reconcile the invoices against received product, make payments
Track leads, schedule meetings and events
Create customer quotations, accept purchase orders, schedule delivery
Issue invoices, accept payments, reconcile accounts
Manage projects, including billing for time on projects
Review finances, generate reports, close out financial periods
Track specialized business intelligence
Generate standard and customized reports

There are certainly some things that Tryton doesn’t do, that a typical ERP system might (or should, or maybe I wish it would) do:

Manage production (work orders, repair orders, schedule resources, etc.)
Comprehensive CRM (track emails, phone calls, social media; manage issues or complaints; delegate accounts and tasks, etc.)
Better localization support for the United States
Support mobile clients

That’s not to say Tryton can’t do these things. It can – some assembly required, please allow six to eight months for delivery.

Looking at my target application, however – performing as the backend for a large-scale e-commerce system – Tryton has much of what I need:

Account management
Catalog management
Stock management and product availability
Multiple pricelists and multi-tiered pricelists
Sales quotations (often called “shopping carts” and “wish lists”)
Sales orders (“Complete your order”)
Order management
Customer support
Process integration (“track your order”)

Alas, there are a few things I’d want to add to that list:

Online payment support through Authorize.net, PayPal, Google Checkout, etc.
Shipping integration with UPS, USPS, and FedEx
WYSIWYG editing of product data
A decent point of sale terminal interface for e-commerce stores with a brick-and-mortar presence

While I might not expect it to be part of Tryton, there’s always the rest of the e-commerce site too – content management, templates, dynamic content outside of the store, etc. I think it could be interesting to see a merger between these elements, but I am not convinced it’s a good marriage. Should an ERP system provide internet and intranet, online document management, institutional knowledge management? Sounds like a question for another blog post. Meanwhile there’s MODx and Django.

Community

I’ll discuss community more in the next post, but Tryton seems to be developed under what’s called a Community Open Source model. If you ask me, that’s the old school way of doing it. You know, the right way, before the new fangled methods came along and perverted the process. (I’m not opinionated or anything.)

Tryton is developed by a federation of companies and individual developers. There seem to be at least five main companies involved, all of which also provide commercial services for the system as well. Of course, there is a strong argument that these companies are most expert in the software. This expertise translates into demand, and revenue.

In contrast, some open source projects (examples include OpenERP²² and Magento²³) follow what’s called a Commercial Open Source model. In this model, one company typically calls the shots (or “edits”) the software. They often do most of the development, especially early on, but they also draw upon work done by users and developers of the system. If the editor likes your code, they pull it into the main branch. If they like your idea but not your code, they might implement your idea themselves. If they don’t want either your idea or your code, it sits out there for others to find and use. Often, the commercial open source model matures into a dual-release model, where the package is released in a community (open source) version, and an enterprise (proprietary) version.

License

I’ve developed in OpenERP and already had that uncomfortable discussion with the client where you teach them how to read a specification, especially the section defining how the code is licensed. OpenERP is, curiously, licensed under the GNU Affero General Public License version 3.0²⁴. It’s just like the (in)famous GNU General Public License version 3.0²⁵, except it adds a bit of text closing what’s called the “Application Service Provider Loophole”. That is, the user of a system licensed under the AGPL has a right to the source code for the system, even if they’re accessing the system over a network. Even if it’s been customized. Even if it’s got trade secrets in it. In practice (and I Am Not A Lawyer) this seems to mean that the people using the OpenERP client to connect to an OpenERP server over a network should be able to retrieve the source code for custom modules. Generally speaking, this would be the customer who paid a developer to create the modules, and of course they should get the code. Likewise, the person who owns a website that connects to an OpenERP server should be able to get any custom OpenERP module code that supports the website connecting to it. Phew. But I don’t think it goes farther than that – I don’t believe this means that the website needs to be licensed AGPL, or that visitors to the website should be able to download code for the website, or any custom OpenERP modules.

Tryton, on the other hand, is licensed under the regular old GNU General Public License version 3.0, complete with ASP loophole. That tells me that the source code for any customization made to the Tryton code (a custom module, for instance) used by a client over a network does not need that custom code offered to the client. I don’t have a problem playing nicely with others. I like to share. I appreciate those subtle shades of difference between the AGPL and GPL. But when it comes to splitting the hairs of freedom, it seems that distributing custom code to lowly users causes palpitations among some clients, typically those with legal departments.

What else is there?

Part 1: About Tryton. An overview of Tryton – architecture, business model, features, license.
Part 2: Stalking the Tryton community. There are some impressive people working on this project. I’m pleased to get to know them.
Part 3: The crux of the matter. Tryton needs a web framework so we don’t have to do things twice.

Instantiate Blog

2011-05-09T20:15:00Z

The story of how and why I ended up blogging. Again.

Stranger in a strange land.

This is somewhere around blogging attempt number 12 for me. I feel like the cheerleader on Heroes¹ announcing yet another failed try at offing myself. And here I go again.

Actually, I haven’t done horribly. Some of my blogs had a few readers. 2 or 3 a month even. Some high points:

grokd.org – an automated blog pulling news clips on autism. Hundreds of readers a month, probably because I didn’t write hardly any of the posts. I shut it down when the b2/cafelog² developers announced they were shifting focus to some new project they called “WordPress”.
closew.org – discussion of community and cooperative living, back when I still thought consensus was the greatest political tool since pork on sliced bread. Now I favor benevolent dictators³ (so long as he is me) and anarchy⁴, especially anarcho-primitivism⁵.
woundedpig.com – an attempt to satirize scary conservatives. Took me forever to write each post. Turns out I’m not Jon Stewart.

Look. And feel.

What can I add beyond what I’ve said on the about page? As of this writing, the blog has an extremely basic theme. One could call it naked, except that would suggest you could see what lies underneath. Let’s say it’s unadorned. I’m using the Roots theme⁶ for these features:

HTML5 Boilerplate (http://html5boilerplate.com/) makes rock-solid HTML5 sites
960.gs (http://960.gs/) makes 960px grid-based design easy
Cleaner URLs with a custom .htaccess
jQuery Cycle and FancyBox – my two favorite jQuery plugins

Of course, Roots is just a starter theme. Ideally the developer will still style it, you know, choose those cleaner and more trustworthy fonts, maybe add a logo, etc.? Take a look at this page. If you’re seeing what I’m seeing (quite possibly not, since I’m at a different point in the space-time continuum than you are), then the developer (a/k/a me, version2beta, rob) has not gotten around to it.

Updated 8/22/11: I’ve moved the blog onto my version2beta.com domain and now I’m using Hyde⁷ to generate the site, so I’m not actually using Wordpress, Roots theme, or Apache, or even HTML5Boilerplate. Or FancyBox, jQuery Cycle, or jQuery anything. It’s all rustic. I’ve gone native. (But I’ll probably get around to making it look more designed eventually.)

Poke me.

Look around at what I’ve written. Or maybe what I’m going to write, since so far all I’ve written is this and the about page. Or maybe, what I think I’m going to write. I’ll probably leave comments on, but really, don’t bother. If I’ve inspired a response from you (probably the type of response that deserves an apology), mention @version2beta on Twitter and I’ll see it there. Or, if you can’t comment in less than 126 characters (my handle takes 14 characters with a space), blog about it and tweet that. I have my doubts about blogging (especially me blogging), but one thing it seems good for is inter-blogal discussions.

version2beta archive

Ramada Plaza Milwaukee kicks my 70 year old mother out at midnight

A Response to Tom Henrich's Document The Why

MODX can beat up your content management system

A quick note on MongoDB replica sets

My 18-monitor setup

Drupal, on GlusterFS, on Rackspace Cloud

Elavon Phishing Scam

Tryton as an e-commerce back-end - Web Framework

First a review

What’s e-commerce anyway?

E-commerce is the ERP interface

ERP is the e-commerce interface

Wow, so Tryton can do all this?

What does Tryton need?

A Tryton web framework

What else is there?

DjangoCon, part four

DjangoCon agenda

Keynote: Glyph Lefkowitz on Why does Django hate Python

Advanced django forms

Teaching Django to comrades

Deployment, daemons, and datacenters

Content ain’t a thing: web scraping

Cache rules everything around me

DjangoCon, part three

DjangoCon agenda

Lightning talks

Keynote: Brad Fitzpatrick

Restful API’s, promises and lies

A Little South Sanity

Building APIs with TastyPie

Stop Tilting at Windmills: Spotting Bottlenecks

Safely Deploying on the Cutting Edge

Benevolent Designer For Life’s Keynote - Designers Make It Go to Eleven

DjangoCon, part two

DjangoCon agenda

David Eaves’ keynote on the science of community

From designer to Django’er in six weeks

Confessions of Joe Developer

The story and tech of Read the Docs

Testing - The developer strikes back

Secrets of PostgreSQL performance

DjangoCon, part one

DjangoCon agenda

Tuesday

Alex Gaynor, Summer in the wild

Confessions of Joe Developer

The story and tech of Read the Docs

Testing - The developer strikes back

Secrets of PostgreSQL performance

Wednesday

RESTful APIs - promises and lies

Real-world Django deployment using Chef

Building APIs in Django with Tastypie

Making the Django ORM multilingual

Stop tilting at windmills - Spotting bottlenecks

Safely deploying on the cutting edge

Thursday

Advanced Django Form usage

Pinax after three years: lessons learnt and the way forward

Advanced security topics

Write the friendly manual

Content ain’t a thing: Web scraping with our favorite Python libraries

Cache rules everything around me

Find me?

jQuery.multiup plugin for multiple files

blog reset

High Availability Linode Pairs

Initial configuration, from within the Linode1 dashboard

Some notes before we get into it

Initial configuration, operating system

Install and configure DR:BD

Services: MySQL and Lighttpd, plus a few other little things I like to have

Configure a failover cluster

A new OpenERP product and license

The OpenERP AGPL + Private Use license

The right to license

Compatible licenses

Why it matters

Initial configuration, from within the Linode¹ dashboard