Worst dev team failure I've experienced?

One of several.

Around 2012, a team of devs were tasked to convert a ASPX service to WCF that had one responsibility, returning product data (description, price, availability, etc...simple stuff)
No complex searching, just pass the ID, you get the response.

I was the original developer of the ASPX service, which API was an XML request and returned an XML response. The 'powers-that-be' decided anything XML was evil and had to be purged from the planet. If this thought bubble popped up over your head "Wait a sec...doesn't WCF transmit everything via SOAP, which is XML?", yes, but in their minds SOAP wasn't XML. That's not the worst WTF of this story.

The team, 3 developers, 2 DBAs, network administrators, several web developers, worked on the conversion for about 9 months using the Waterfall method (3~5 months was mostly in meetings and very basic prototyping) and using a test-first approach (their own flavor of TDD). The 'go live' day was to occur at 3:00AM and mandatory that nearly the entire department be on-sight (including the department VP) and available to help troubleshoot any system issues.

3:00AM - Teams start their deployments
3:05AM - Thousands and thousands of errors from all kinds of sources (web exceptions, database exceptions, server exceptions, etc), site goes down, teams roll everything back.

3:30AM - The primary developer remembered he made a last minute change to a stored procedure parameter that hadn't been pushed to production, which caused a side-affect across several layers of their stack.

4:00AM - The developer found his bug, but the manager decided it would be better if everyone went home and get a fresh look at the problem at 8:00AM (yes, he expected everyone to be back in the office at 8:00AM).

About a month later, the team scheduled another 3:00AM deployment (VP was present again), confident that introducing mocking into their testing pipeline would fix any database related errors.

3:00AM - Team starts their deployments.
3:30AM - No major errors, things seem to be going well. High fives, cheers..manager tells everyone to head home.
3:35AM - Site crashes, like white page, no response from the servers kind of crash. Resetting IIS on the servers works, but only for around 10 minutes or so.
4:00AM - Team rolls back, manager is clearly pissed at this point, "Nobody is going fucking home until we figure this out!!"
6:00AM - Diagnostics found the WCF client was causing the server to run out of resources, with a mix of clogging up server bandwidth, and a sprinkle of N+1 scaling problem. Manager lets everyone go home, but be back in the office at 8:00AM to develop a plan so this *never* happens again.

About 2 months later, a 'real' development+integration environment (previously, any+all integration tests were on the developer's machine) and the team scheduled a 6:00AM deployment, but at a much, much smaller scale with just the 3 development team members.

Why? Because the manager 'froze' changes to the ASPX service, the web team still needed various enhancements, so they bypassed the service (not using the ASPX service at all) and wrote their own SQL scripts that hit the database directly and utilized AppFabric/Velocity caching to allow the site to scale. There were only a couple client application using the ASPX service that needed to be converted, so deploying at 6:00AM gave everyone a couple of hours before users got into the office. Service deployed, worked like a champ.

A week later the VP schedules a celebration for the successful migration to WCF. Pizza, cake, the works. The 3 team members received awards (and a envelope, which probably equaled some $$$) and the entire team received a custom Benchmade pocket knife to remember this project's success. Myself and several others just stared at each other, not knowing what to say.

Later, my manager pulls several of us into a conference room
Me: "What the hell? This is one of the biggest failures I've been apart of. We got rewarded for thousands and thousands of dollars of wasted time."
<others expressed the same and expletive sediments>
Mgr: "I know..I know...but that's the story we have to stick with. If the company realizes what a fucking mess this is, we could all be fired."
Me: "What?!! All of us?!"
Mgr: "Well, shit rolls downhill. Dept-Mgr-John is ready to fire anyone he felt could make him look bad, which is why I pulled you guys in here. The other sheep out there will go along with anything he says and more than happy to throw you under the bus. Keep your head down until this blows over. Say nothing."

I've found and fixed any kind of "bad bug" I can think of over my career from allowing negative financial transfers to weird platform specific behaviour, here are a few of the more interesting ones that come to mind...

#1 - Most expensive lesson learned

Almost 10 years ago (while learning to code) I wrote a loyalty card system that ended up going national. Fast forward 2 years and by some miracle the system still worked and had services running on 500+ POS servers in large retail stores uploading thousands of transactions each second - due to this increased traffic to stay ahead of any trouble we decided to add a loadbalancer to our backend.

This was simply a matter of re-assigning the IP and would cause 10-15 minutes of downtime (for the first time ever), we made the switch and everything seemed perfect. Too perfect...

After 10 minutes every phone in the office started going beserk - calls where coming in about store servers irreparably crashing all over the country taking all the tills offline and forcing them to close doors midday. It was bad and we couldn't conceive how it could possibly be us or our software to blame.

Turns out we made the local service write any web service errors to a log file upon failure for debugging purposes before retrying - a perfectly sensible thing to do if I hadn't forgotten to check the size of or clear the log file. In about 15 minutes of downtime each stores error log proceeded to grow and consume every available byte of HD space before crashing windows.

#2 - Hardest to find

This was a true "Nessie" bug.. We had a single codebase powering a few hundred sites. Every now and then at some point the web server would spontaneously die and vommit a bunch of sql statements and sensitive data back to the user causing huge concern but I could never remotely replicate the behaviour - until 4 years later it happened to one of our support staff and I could pull out their network & session info.

Turns out years back when the server was first setup each domain was added as an individual "Site" on IIS but shared the same root directory and hence the same session path. It would have remained unnoticed if we had not grown but as our traffic increased ever so often 2 users of different sites would end up sharing a session id causing the server to promptly implode on itself.

#3 - Most elegant fix

Same bastard IIS server as #2. Codebase was the most unsecure unstable travesty I've ever worked with - sql injection vuns in EVERY URL, sql statements stored in COOKIES... this thing was irreparably fucked up but had to stay online until it could be replaced. Basically every other day it got hit by bots ended up sending bluepill spam or mining shitcoin and I would simply delete the instance and recreate it in a semi un-compromised state which was an acceptable solution for the business for uptime... until we we're DDOS'ed for 5 days straight.

My hands were tied and there was no way to mitigate it except for stopping individual sites as they came under attack and starting them after it subsided... (for some reason they seemed to be targeting by domain instead of ip). After 3 days of doing this manually I was given the go ahead to use any resources necessary to make it stop and especially since it was IIS6 I had no fucking clue where to start.

So I stuck to what I knew and deployed a $5 vm running an Nginx reverse proxy with heavy caching and rate limiting linked to a custom fail2ban plugin in in front of the insecure server. The attacks died instantly, the server sped up 10x and was never compromised by bots again (presumably since they got back a linux user agent). To this day I marvel at this miracle $5 fix.

GodDAMN, C# 7.0 is so ridiculously feature packed. I can pattern-match inside a predicate on an exception filter. Want to catch ONLY NHibernate's exceptions caused by a SQL timeout? Boom:

catch (GenericADOException adoEx) when (adoEx.InnerException is SqlException sqlEx && sqlEx.Number == -2) { return Failure("timeout"); }

At work, all errors within the site are logged into our database with a subject and error column. SQL errors are logged in the subject field while the traceback is put in the error column. However, a lot of SQL errors are really large and exceed the max character width of the subject field, causing yet another SQL error, and the cycle repeats. This recursive error has been the bane of my existence, because 1) it times my local dev instance out and 2) the error doesn't end up getting logged because the server both freezes and the error can't be inserted in the database. You can't even begin to imagine how many hours I've wasted trying to find what line I changed cause total and utter failure with absolutely 0 error logging. Next thing on my todo list is to fix this fucking issue since the head dev refuses to get it done.

So for those of you keeping track, I've become a bit of a data munger of late, something that is both interesting and somewhat frustrating.

I work with a variety of enterprise data sources. Those of you who have done enterprise work will know what I mean. Forget lovely Web APIs with proper authentication and JSON fed by well-known open source libraries. No, I've got the output from an AS/400 to deal with (For the youngsters amongst you, AS/400 is a 1980s IBM mainframe-ish operating system that oriiganlly ran on 48-bit computers). I've got EDIFACT to deal with (for the youngsters amongst you: EDIFACT is the 1980s precursor to XML. It's all cryptic codes, + delimited fields and ' delimited lines) and I've got legacy databases to massage into newer formats, all for what is laughably called my "data warehouse".

But of course, the one system that actually gives me serious problems is the most modern one. It's web-based, on internal servers. It's got all the late-naughties buzzowrds in web development, such as AJAX and JQuery. And it now has a "Web Service" interface at the request of the bosses, that I have to use.

The programmers of this system have based it on that very well-known database: Intersystems Caché. This is an Object Database, and doesn't have an SQL driver by default, so I'm basically required to use this "Web Service".

Let's put aside the poor security. I basically pass a hard-coded human readable string as password in a password field in the GET parameters. This is a step up from no security, to be fair, though not much.

It's the fact that the thing lies. All the files it spits out start with that fateful string: '<?xml version="1.0" encoding="ISO-8859-1"?>' and it lies.

It's all UTF-8, which has made some of my parsers choke, when they're expecting latin-1.

But no, the real lie is the fact that IT IS NOT WELL-FORMED XML. Let alone Valid.

THERE IS NO ROOT ELEMENT!

So now, I have to waste my time writing a proxy for this "web service" that rewrites the XML encoding string on these files, and adds a root element, just so I can spit it at an XML parser. This means added infrastructure for my data munging, and more potential bugs introduced or points of failure.

Let's just say that the developers of this system don't really cope with people wanting to integrate with them. It's amazing that they manage to integrate with third parties at all...

copying field data from one table to another,

wondering why the execution takes so long,

since your copying only data from three rows,

realising you forgot the WHERE clause in the SQL while working on a live database and overwriting all data with the last statement 😨