|
_Maxxx_ wrote: "Production" he croaked
Been there, done that.
That feeling of icy chill that goes down your back is one you never forget.
cheers
Chris Maunder
|
|
|
|
|
_Maxxx_ wrote: everything is being tested, tested, then tested again
In my experience, that's generally not worth the trouble. Something will always go wrong, fix it when it does and move on.
|
|
|
|
|
Ah Yes, different colors for different systems.
We had terminals and a switch box. I was working on Dev, ran out for a printout.
My tests were great, I entered the command to DELETE my test data.
The phones started ringing. I picked up, the ENTIRE list of GMAC Car loans were missing from Florida. Strange, I thought. That is the account I just deleted on Dev.
Let me turn my Switch box over to production Florida... Oh, strange, why is it ALREADY on Florida. OMG, run into bosses office, call emergency meeting. Contract operations, start the restore process from last night. Gather all work done for the day, and set aside to be re-processed. OMG, what a mistake.
Later, someone else killed Texas with the same command. I ultimately patched the operating systems so that command would FAIL on production, a strange version of the command would have to be used! Never happened again.
|
|
|
|
|
Yep, I trashed the Inspections system for a property management company once that way. And I had my systems color coded and everything.
I was at the tail end of a 32-hour shift, had just finished my God-only-knows-what-number cup of break-room coffee, and hit execute before I really had checked the command well. Only thing that saved me there was that I realized immediately what I had done and broke the backup-imaging process so I didn't lose the same tables in the backup system. Then copied the data back over and restored the process. Twelve minute outage, and enough adrenaline to light up a city block...
There are lies, there are damned lies and then there are statistics. But if you want to beat them all, you'll need to start running some aggregate queries...
|
|
|
|
|
It doesn't happened to me, but a mate from the office...
Those days we had no direct connection to customers so deployment involved going to the site physically...
We had a set of 18 1.44 floppy disks we were preparing for deployment.
My mate prepared the disks, took them and went home. At the morning he went on for a 3 hour driving to get to the customer, just to find, that the last floppy still in the PC at the office...
After some thinking over phone we send him to buy a modem (28.8 was the best back then), install it on one of the customers PCs and set up a connection to transfer the file...
After an other hour and half he had all the material at hand and with a 3 hours delay started the deployment...
And we got our ever first online customer!!!
Skipper: We'll fix it.
Alex: Fix it? How you gonna fix this?
Skipper: Grit, spit and a whole lotta duct tape.
|
|
|
|
|
What is until today still known as the "Ghent Crisis". A bug that creeped up after roll-out and eluded us on trying to reproduce it. There was a severe political deadline (and thus media attention upon failure). Me and my entire team of about 5 people had to go on site from 08:00-23:00 every day (including weekends) until the bug was fixed. My boss finally managed to reproduce the bug, but it still took us several days to fix it due to it's complexity.
At least we all got a small bonus for it. 
([EDIT]The entire crisis lasted for two weeks and we managed in not having a roll back.[/EDIT])
|
|
|
|
|
When you set up a retail banking system, it is imperative that you test ALL the posting possibilities of the General Ledger BEFORE deploying.
I was working at a rather large bank and we were going live with the new retail system. Friday night, unbeknown to us, the back office team in charge of config and data make an itsie teeny weeny change to one of the GL posting rules.
Data migration flew past over the weekend as it had been properly tested and we were good to go for Monday morning's go live.
Then.
Things.
Happened.
Luckily we had a couple of guys in branch to watch what was happening and we got a call that the first transactions where going through. So to be sure, we checked everything was going okay. Cash books - check, account balances - check, client [GL] accounts - check, bank's ledgers - FAARRRKKKK!!!!! what should be impossible - mismatched entries - where all over the GL. Feck, feck, feckity. We couldn't shut down to investigate, so a couple of poor sods from back office start manually entering corrections.
Then a bright spark asks "Where have the other postings gone? They have to be 'somewhere'?"
All the missing postings where found in a back drawer correction ledger and we had to manually back out all the earlier corrections and correct the original incorrect postings all while not knowing where the hell the mismatches where being generated from.
When the bad rule was found the guy who changed it was castigated some what. This then became a near standard feature of any roll out for them, at the eleventh hour someone would add in a little spice and it all go boom. 
|
|
|
|
|
Many years ago I worked for a company that had over a hundred Linux servers around the country running Progress DB and application. Each of these servers where connected by a dial up modem that we could also dial into (providing they were still plugged into the same phone line that we had installed them on). We needed to make a schema change to each of the DBs which were all identical.
We had no way of making such changes, although we did for code and data.
We came up with a script that could be deployed, pulled down by each server when it made a routine connection, and then run to make the schema change.
We tested it, it worked.
I released it at 6 o'clock on a Saturday morning when I was the only person in the office and sat back to watch the monitor software for signs of the change being successfully applied.
One by one the updates failed, then people started to phone up because they could no longer use their system.
I phoned up the big boss man, we had a quick chat, we worked out what had gone wrong and how to fix it.
The fix involved me dialing into each server, running a few commands against the DB, then starting it back up again.
In just over a hundred servers three applied the patch correctly.
By the time I had done fixing the rest I went to go home, locked up the warehouse I was in, set the alarm (I was the only person working in it on that Saturday), then went to leave the site. When I got to the gate I found them padlocked and the the site empty. I had to phone the big boss man back who made a few calls and found someone to come down with some bolt cutters to set me free.
There then followed a comprehensive review of the company's lone working procedures (of which there had been none before).
Some men are born mediocre, some men achieve mediocrity, and some men have mediocrity thrust upon them.
|
|
|
|
|
LOL, you can clock out anytime you want but you can never leave 
|
|
|
|
|
While it's not specifically related to software deployment, the first thought that came to my mind is that anyone in the business of launching satellites and people into space have truly the worst deployment stories.
Marc
|
|
|
|
|
Yeah. They win this game.
cheers
Chris Maunder
|
|
|
|
|
"Oh, you guys were using metric units?"
|
|
|
|
|
agolddog wrote: "Oh, you guys were using metric units?"
Yup. I remember that one.
Marc
|
|
|
|
|
In the last '80s, I was working for a company that was running two production systems on PDP 11/73's in a manufacturing environment. We had maxed out the memory and had code that would work in test, but not in production because of the memory issues.
The first system was product routing in the plant; the second was a storage system for receipt/delivery from/to the first system.
A project was created to rewrite the first system to run on a MicroVAX; similar technology, language, etc. The rewrite took a year with a team of people.
When implemented (and there was no go back, only go forward), it was discovered the code had not been complied with array bound check. It wasn't done in the old version because that would use to much memory and overlooked in the new system. Production ground to a halt for January and most of February. Changes had to be described on a paper form and signed off by on site support before being implemented.
The next year, the second system was rewritten with lessons learned. On implementation, around the clock support. Management asked when it was going to be installed; we said it was installed - two weeks ago. Much better.
I am a firm proponent to a post mortem on all projects; see what worked, what didn't and learn from it. No finger pointing, just learning.
Tim
|
|
|
|
|
Great Thread.
I love ghost stories.
|
|
|
|
|
But I dread failed Ghost stories
|
|
|
|
|
Oh, and not a deployment story, but...
There was a system I wrote that involved communication with a third-party product via a socket connection. I had a table in the database to contain all the messages to and from the socket. It quickly became rather large so I decided to trim it down a bit, something along the lines of DELETE FROM messages WHERE timestamp<amonthago in SSMS, and I sat there wondering when it would finish. Then the phone rang, it's the President of the Company, "the system is unresponsive, the call center is at a stand-still, are you doing anything that might be causing trouble?" Oops.  Try terminating the DELETE, no go.  Shut down my PC, reboot, get back into database, see that everything is working again.
I then wrote a feature that would delete a thousand messages then sleep and repeat until there were no more messages to delete. It took several days for the process to get the old messages cleaned out.
|
|
|
|
|
I've two that vie for worst ever:
First one was a server system upgrade. We were dropping in a new SAN, changing from the old switch to a iSCSI fabric and doing a major OS upgrade. All in the same night.
Why are you groaning?
My part was the (relatively) easy one; I had to make sure the databases were backed up and ready for shut down, wait for the rest of the team to do the physical changes in the colo facility, and then initiate the upgrade/rollout to 800 thin client machines throughout the building. The first part went just fine. The server techs pulled the plug on the switch...
And that is when the magic happened.
They hadn't actually bothered to shut down any of the servers before yanking the switch. We were in a 100% virtualized environment using vSphere and suddenly every server was basically disconnected simultaneously from the control systems. Our CPU on the servers red lined as the poor little VM's tried to figure out what had happened and fix it. Meanwhile, they're trying to dump stack traces and error logs to the database, to disk, to anywhere they can find a place... and nothing is responding. Of course, I have no idea this is happening, and the server techs are too busy trying to map a switch panel and replicate it in the fabric (I am not joking, they hadn't bothered to map this ahead of time) to notice.
Five hours later, the new fabric is in place and the team reconnects the servers. My monitoring systems are the first things back online and they suddenly flood with the best part of 400 GB of error data and stack dumps. It was like getting hit with the biggest DDoS I've ever seen. Everything was screaming for attention, the fabric was actually misconfigured (although we didn't know at the time) and was sending packets in a round-robin loop to non-existent addresses, and the databases, hit with error log write queries, slowed to a crawl.
It took 3 hours to sort it out. Reboot everything. Flush error logs to disk. Kill db threads. You name it. And all of this masked the fabric errors. So when our lead server tech left the colo, he headed straight for vacation. Two hours later, he was somewhere in the air with his cell phone off... and we found the misconfiguration issue when people started arriving for work.
The other one started as a relatively benign little program that ran questionnaires for our call center. Basically, it was a "ask these questions and route through the top 50 most common issues" type of program. Neat little thing. Anyway, we were asked to add a half-dozen additional scenarios to it, which we did. Tested. Retested. Deployed to the test bank. All good.
Our call center was 24-hour, but was split into three "banks," and usually only one would be in operation at a time. So, we planned the upgrade to banks 1 and 2 for 1830, and bank 3 at 0600 the next morning. Because of the nature of the business, all of these upgrades were getting pushed to the machines via central deployment script. We set it up. Sent it to the admins. They confirmed the script was ready. Everything good so far.
I asked my team to stay put for the 1830 rollout, because I've had deployments go pear-shaped in the past, but save for one brand-new (less than one year experience) developer, all of them had "other plans." I had no authority to force them to stay, so I watched them pack up and leave at 1700, leaving me and Newbie to watch the rollout. At 1805 I got a call from the call center supervisor. The system was down.
I logged on remotely to look at the problem. And then called the admin team. The dumb cluck that had set up the push schedule had put in 1800 and all three banks... And the version of the program that was sent for installation was the wrong one. The developer that had packaged it used the one on her local machine, NOT the one in source control. It was a debug version, and at least a week out of date. So, every system in the call center was locked out.
Newbie scrambled to repair the script with the source control version; I had to call the VP of operations to get permission to do an emergency deployment (yes, really), and we managed to rollout the new version with only an hour of downtime. The team meeting that next morning was not a happy one...
This one makes the list because when it happened, we had the big boss (company president) touring the call center. He was actually watching over the shoulder of one of the call center personnel when the system went down. Made for a whole new level of oops.
There are lies, there are damned lies and then there are statistics. But if you want to beat them all, you'll need to start running some aggregate queries...
|
|
|
|
|
My friend was on work experience from college with one of ireland's largest bank about a decade ago.
She was on night shift and was tired and deploying something and mistyped and accidentally shut down every ATM in the country.
Had to wait ages for them all to come back online again.
|
|
|
|
|
wasn't me. Thank goodness.
But at an organization I used to work at. A co-worker upgraded a SAN storage system. on a Tuesday. AFternoon,
Everything crashed.
Had he verified the backups nope, No backup. Everyone lost 3 months worth of work.
two weeks later SAN was finally up.
Not sure whatever became of co-worker. Never saw him again.
To err is human to really mess up you need a computer
|
|
|
|
|
I do a lot our deployments. I have a rule. Never trust anything built after 5pm, or after 3pm on Fridays.
|
|
|
|
|
Got a phone call in the middle of the night once -- system crashed, and since my tools had been used to convert it to the new OS, my name was on the on-call list. Unbeknownst to me, I was in the middle of an appendicitis attack in my sleep. I can't really describe how awful I was feeling while trying to help them with a problem that I had no clue about (turned out to be buggy filesystem code).
We can program with only 1's, but if all you've got are zeros, you've got nothing.
|
|
|
|
|
A few years back a company I worked for decided to develop an in-house system to handle debit card transactions directly with MasterCard instead of going through a middle-man.
Luckily, I wasn't on that team, but I worked in the same room and worked with those programmers on other projects, so I had a really good view of the whole thing, from development to deployment.
It turned out that certain key information about how data needed to be passed back-and-forth was never relayed to the programmers, who ended up coding to some slightly fishy specs.
It went live, and all debit card transactions started going through the new system. At that point there was no going back. And it blew the $%#@ up! Suddenly people all over the country started having their cards declined, and these cards were being used for prescription medications! That's a level of pissed-off you don't even want to know about.
Our clients started getting massive amounts of angry calls from their customers, and they tore into us. Those poor clients were caught in the middle between their irate customers and us, the cause of the problem, with nothing they could do except take angry phone calls all day and wait for us to fix it.
Once the programmers figured out what was wrong they were able to fix the system, but it took months to clean up the data mess and get everything sorted out. A lot of clients dropped us over it, it was nasty, by far the worst SNAFU I've ever seen. And I still thank my lucky stars that I wasn't on that team!
|
|
|
|
|
Gather 'round kiddies while I spin the tale of the deployment from hell that I had nothing to do with, but be an amused, detached, viewer.
The company I worked for published tax law books. They also published tax preparation software. Initially they had a mainframe version that was their big cash cow, but the advent of PC's had cut terribly into that business and so they started creating versions for the PC. They did versions for the US, Canada, and Australia.
The time, early 90's. The Australian tax season is in full swing. At a company that usually sacrifices a goat on the altar of schedules, daily, the deployment of the new version is late, but they get it out the door.
But there is one, teensey, weensey, problem. All the code was developed on 386's, when most of Australia's accountants had 286 based machines.
In their rush to get the product out the door, they never bothered to test it on a 286 based machine. If they had, they would have found performance was glacier.
The collective scream from Australian accountants was heard all the way to the corporate headquarters in the US.
The new VP of Technology was seen walking through the US offices asking which programmers were on staff that had 1) a valid passport, and 2) their vaccinations up to date. Anyone answering in the affirmative was on a plane the next day.
The ultimate solution ended up that the company gave the accountants that screamed the loudest, new 386 machines to do their work on. The agreement was that they could keep their shiny new machines, provided they promised to continue buying the future releases of the tax software.
Needless to say, there weren't any profits from that division that year and more than a few heads rolled.
Psychosis at 10
Film at 11
Those who do not remember the past, are doomed to repeat it.
Those who do not remember the past, cannot build upon it.
|
|
|
|
|
Ok you code monkeys.
Monday's CC batch didn't show up in our account ($100.00)
But all batches for Tuesday on have shown up without incident.
A call to our bank (1st Bank) revealed that Monday a new CC processor was switched to and there was a problem so all deposits for Monday won't land in accounts until Friday (Tomorrow). Needlness to say, $100.00 isn't a lot to go without but the lady at the bank was thankful to us that we were so nice about it. We said, "well it's only 100 bucks, I'll bet there were many more with much more to loose" - The lady at the bank said she has been fielding irate calls all week from customers mad as #$%^.
Ok what happened? Who did it?
Let's hear all about it.
|
|
|
|
|
General 
News 
Suggestion 
Question 
Bug 
Answer 
Joke 
Praise 
Rant 
Admin 
Submit an article or tip
