Bonus points for sharing what you learned/key takeaways from the experience.
Lesson #1: Don't code when you're distracted.
Some hours later, the problem manifested. The queue workers came down, and AR (which is totally dependent on them for its core functionality) immediately stopped doing the thing customers pay me money to do. My monitoring system picked up on this and attempted to call me -- which would have worked great, except my cell phone was in a box that wasn't unpacked yet.
Lesson #2a: If you're running something mission critical, and your only way to recover from failure means you have to wake up when the phone rings, make sure that phone stays on and by you.
Later that evening I felt a feeling of vague unease about my change earlier and checked my email from my iPad. My inbox was full of furious customers who were observing, correctly, that I was 8 hours into an outage. Oh dear. I ssh'ed in from the iPad, reverted my last commit, and restarted the queue workers. Queues quickly went down to zero. Problem solved right?
Lesson #3: If at all possible, avoid having to resolve problems when exhausted/distracted. If you absolutely must do it, spend ten extra minutes to make sure you actually understand what went wrong, what your recovery plan is, and how that recovery plan will interact with what went wrong first.
AR didn't use idempotent queues (Lesson #4: Always use idempotent queues), so during the outage, every 5 minutes on a cron job every person who was supposed to be contacted that day got one reminder added to the queue. Fortuitously, AR didn't have all that many customers at the time, so only 15 or so people were affected. Less than fortuitously, those 15 folks had 10 to 100 messages queued, each. As soon as I pressed queues.restart() AR delivered all of those phone calls, text messages, and emails. At once.
Very few residential phone systems or cell phones respond in a customer-pleasing manner to 40 simultaneous telephone calls. It was a total DDOS on my customers' customers.
I got that news at 3 AM in the morning Japan time, at my new apartment, which didn't have Internet sufficient to run my laptop and development environment to see e.g. whose phones I had just blown up. Ogaki has neither Internet cafes nor taxis available at 3 AM in the morning. As a result, I had to put my laptop in a bag and walk across town, in the freezing rain, to get back to my old apartment, which still had a working Internet connection.
By the time I had completed the walk of shame I was drenched, miserable, and had magnified the likely impact that this had on customers' customers in my own mind. Then I got to my old apartment and checked email. The first one was, as you might expect, rather irate. And I just lost it. Broke down in tears. Cried for a good ten minutes. Called my father to explain what had happened, because I knew that I had to start making apology calls and wasn't sure prior to talking to him that I'd be able to do it without my voice breaking.
The end result? Lost two customers, regained one because he was impressed by my apology. The end users were mostly satisfied with my apologies. (It took me about two hours on the phone, as many of them had turned off their phones when they blew up.)
You'd need a magnifying glass to detect it ever happened, looking on any chart of interest to me. The software got modestly better after I spent a solid two weeks on improved fault tolerance and monitoring.
Lesson the last: It's just a job/business. The bad days are usually a lot less important in hindsight than they seem in the moment.
Lesson #1: Don't push code on Friday afternoon.
Lesson #2: Beer, Code and Commit is totally fine. Just don't push! Wait until next day to review and push/deploy it..
As a general rule, the best time to launch is first thing on a Tuesday. Why Tuesday morning? We are also well aware of "Monday morning madness" too - pre-planning stress for Monday morning is akin to masochism.
The technical staff all talk about the dangers of Friday evening deployments.
The company has ignored/overridden these warnings in the past, and sometimes experienced catastrophic failures because of it.
STILL, sometimes the marketing staff will have weekend promotions that are considered more important than these concerns.
In particular, a weekend promotion will be implemented, tested on Wednesday, and ready to go. And only after it's active, is a bug found in it on Friday at 5pm. And the promo is more important than the risk of failure.
I'm not saying this as a "bitter techie"... I'm just explaining that this is how it happens.
I think that the "it's more important" argument is often correct. Not running the promo costs the company $X of opportunity cost, and X is often very well known, from past promotions. Balance "losing $X" against "well, there's only a small chance of something going wrong", and that's why the risk is often taken.
Take the risk enough times, and eventually something goes wrong.
We exclusively push code on Fridays.
Funnily enough, change control actually runs a lot more smoothly in terms of getting past senior non-technical managers when you include elements like that
continuous deployment to a dev/test environment is the easiest for most organisations to move to. Due to the live environment being mission-critical, they can't afford the risk of any degradation of service. So you push regularly (after passing test suites) to a test environment, get a small subset of users working, and at some point then push out to live from that. But I suspect that isn't what you are referring to, as this is too similar to typical change management.
the alternative definition of continuous deployment is that of constantly pushing to live, initially for a subset of users then rolling it out from there gradually, but always on the live environment. In many large, often 'cloud-y' solutions, that subset might be all users. Except you can't have a public transport ticketing system fail at peak times. A hospital patient record system must stay available for staff, and give sufficient notice before any possible impact to service to allow manual processes to be used. Payroll, accounts, HR systems... all of these have failure modes that have financial penalties at best.
Hence continuous deployment only makes sense when you can afford to risk service, possibly with significant impact.
Excuse me for caviling at your post, but "fortuitously" is a synonym for "accidentally", not "fortunately".
Thanks for the comment though - I personally always appreciate a gracious correction.
Who ships a product release immediately after a commit?
Turns out? Nope.
I also find that the arguments I had with my bosses come around again, from my employees.
I still don't really understand how business works - but I've learned enough to understand that I don't understand how business works.
My question is: Who never makes mistakes? I certainly do not belong in that set.
And I like Patrick's candid business anecdotes. Snarky comments like this might disencourage people from writing useful advice here. Which would be very unfortunate.
this is the primary difference I see between programmers and sysadmins, development and operations. I know programmers who don't own cellphones at all, while I know some sysadmins who take tertiary backup communication devices on vacation.
It's a difference in focus.
Of course, most programming jobs have /some/ operational responsibilities, and most sysadmin jobs have /some/ development responsibilities, but most people see themselves as primarily one or the other, and act accordingly.
> Don't code when you're distracted.
Same story here, I can't remember the exact scenario but I was concurrently acting under all three of my titles (Developer, Architect, Escalation Engineer/Critical Debugger). The customer (who was 7 hours different to us) had been battling for 3 months and we were getting nowhere (all thanks to a, since fixed, bug in WinDBG which essentially came down to broken stack traces in certain scenarios), for those 3 months I had been working 20 hour shifts (development by day, support by night).
Under that strain I eventually made a screw up with the dev and it cost QA time. The MD of my region had a sit down with me and I
> cried literal tears
Needless to say they were impressed that I someone cared so much and sent me home to sleep. The next day I came in and decided to go through the 800kloc codebase line-by-line and see what could be causing the issue - I found it in a few hours.
Lesson #4: Get sleep.
She found out about it pretty quickly due to having syslog be a constant presence in one of her gnu screen windows and gave me a look. She quickly reverted what I did, updated our config management tool, tested it, then deployed it, while explaining why this was the right way to do things. I slowly came around to doing things the right way and haven't thought much about the initial incident until we found her personal logs that she archived and left on our public network share for future reference.
In the entries for the day that I started, we saw the following two lines:
[*] 2007/09/09 09:58 - yan started. gave sudo privs and initial hire forms.
[*] 2007/09/09 10:45 - revoked yan's sudo privs.I'm amazed that this is possible. How would I set something like that up? A realtime log of only the most significant events of a remote system?
In fact, I'd like to take this opportunity of ignorance-admitting to ask the community for general linux/bsd sysadmin resources. What books should I read, or what topics should I study? I want to become an expert at modern sysops. Modern deployment, hardening, backup, managing dozens of boxen, etc.
I've been thinking of going through any MIT OCW on the subject, but it seems like hard-earned experience might not necessarily translate well to an academic setting. What would you recommend I do?
General devops / sysops/ sysadmin knowledge can be had through a variety of means - I got most of my knowledge from simply reading the FreeBSD manual and making a lot of mistakes with my own servers.
This way, a complete admin newbie would learn about digging through the systems by working out the kinks of practical everyday problems. Remember, this is only the most basic instruction, nowhere near enterprise-grade.
If there was a "prospective admin" who had never before run Linux, I'd tell them to install and use Ubuntu/Mint. (Those guys whould usually only be trained to be a helping hand for a "senior" admin.)
If he'd already used Ubuntu at home, I'd tell them to start using Debian and work out how to set up an SSH server and set up their home machine so they could access it remotely.
If they had dabbled with Debian, Fedora, SuSe or something similar, I'd tell them to install Arch and set up some "interesting things", like a mail server or a nis server.
If they were using Arch or Gentoo at home, I'd just personally show them the important things about our system and have them wingman with me for a few days.
If you are already an advanced Linux or BSD user, my approach is of course not appropriate. Instead I'd recommend to pick skills that you want to learn (iptables? Exim?) and set that up. Read Manuals! Read RFCs!
Best of luck.
Accurately assessing one's own competence is difficult and makes for boring reading, but since it's probably necessary here, I'll give some background.
If he'd already used Ubuntu at home, I'd tell them to start using Debian and work out how to set up an SSH server and set up their home machine so they could access it remotely.
If they had dabbled with Debian, Fedora, SuSe or something similar, I'd tell them to install Arch and set up some "interesting things", like a mail server or a nis server.
If they were using Arch or Gentoo at home, I'd just personally show them the important things about our system and have them wingman with me for a few days.
I'd say my current skill level is a mixture of those three. For example, I don't know how to deploy a web service which can send out email for users to e.g. reset passwords. So I don't know anything about email. On the other hand, I've been trying to hone my skills by hardening a Debian server using iptables. On my third hand, while I could set up a box at home that can be SSH'd remotely, I'm not yet confident I know all the best practices. I think the best SSH practices are: change the default SSH port, disable root login, and disable password-based login (use a password-protected keyfile instead).
Beyond that, what is interesting to me is being able to set up dozens or hundreds of systems. Doing this by hand is fraught with error, so it seems like I should learn about virtualization + deployment systems. I've heard good things about Ansible and Salt, but I've also heard Salt considered security an afterthought, which didn't sound good.
It's sounding like my best bet is just to try things, but I want to set things up correctly from a security perspective.
I should also enhance my knowledge of networking... perhaps by spending a few weeks on OCW material regarding the networking stack. How packets are routed, the details of TCP, that sort of thing.
Thanks so much for your insight!
If they are using gentoo, you should be finding someone else. Gentoo users are typically the most dangerous combination of profoundly ignorant, yet absurdly overconfident in their abilities. Seeing a bunch of autotools and gcc output scroll by does not teach you anything. But the mistaken reputation as an "advanced" distro makes people think that by using gentoo, they are therefore "advanced".
In numbers, I mean, not form factor.
If you go though the handbook properly (and potentially enough times until you don't need it to install), then the amount of inherent linux usage and admin knowledge you can pick up is just phenomenal -for example I love the xkcd[1] even if it stopped applying when I started using gentoo.
I would expect a gentoo user to be comfortable on the command line, which doesn't hold true for a lot of other desktop users. That said, isn't it immense desktop linux has gotten to the point where the barriers to entry are grandma level low :)
It's also probably fair to say that every userbase has it's vocal idiots... [1] http://www.xkcd.com/1168/
There are any number of log tailers. One I've recently discovered and love is multitail by Folkert vanHeusden:
http://www.vanheusden.com/multitail/
Incidentally, a related utility (and one which works really well with multitail) is rsstail:
http://www.vanheusden.com/rsstail/
One option is to get an RSS/Atom feed of critical system events.
Are there any MIT OCW courses on this? I haven't come across any.
Sys Admin Body of Knowledge
Nobody was killed, but we had a few injured. Thankfully the brunt of it hit the MRAP in front of us. If it hit my vehicle (HMMWV, flat bottom) instead I probably wouldn't be here.
That was the first major operation on my first deployment, too. Hello, world!
My takeaway? Shit just got real.
We ended up stranded that night after the 3rd IED strike (our "rescuers" said it was too dangerous to get us). It was the scariest day of my life, but in similar future situations it was different. I still felt fear and the reality of the existential threat, but I accepted it. It was almost liberating. Strange.
I deployed for another year after that (to Afghanistan that time). After Afghanistan I left the Corps and started my company. Because if it fails, what's the worst that can happen? Lulz.
Depending on the industry, that might not be the case though. Thanks for your service.
I was a little sleepy one morning and accidentally connected to prod instead of testing. I thought, "That's weird, this UPDATE shouldn't have taken so long-oh shit." I'd managed to clear all allergy and malignant hyperthermia fields. For all I knew, some anesthesiologist would kill a patient because of my mistake. I was shaking. I immediately found the technical lead, pulled him from a meeting, and told him what happened. He'd been smart enough to set up hourly DB snapshots and query logs. It only took five minutes to restore from a snapshot and replay all the logs, not including my UPDATE.
Afterwards, my access to prod was not revoked. We both agreed I'd learned a valuable lesson, and that I was unlikely to repeat that mistake. The tech lead explained the incident to the higher-ups, who decided to avoid mentioning anything to the affected hospitals.
If it's any consolation, the company is no longer in business.
Just remember when you screw things up: Your mistake probably won't get anyone killed, so don't panic too much.
Blaming yourself here is like blaming yourself for being hurt after being told to drive a car with no seatbelt or brakes.
What system would you put in place to prevent this? The issue was that I connected to prod when I thought I was connecting to a test DB. We each had different credentials for prod vs everything else, but the SQL client remembered my username and password. Anyone with prod access could have made the same mistake.
Also, sort of related, I'm using MacOS, and in the back of my head I've wanted to create a tool that will change the color of the menu bar (at the top of the screen) to, say, bright yellow, when I'm connected to the VPN so that I don't accidentally visit a porn site while still connected to work.
That said, neither of these systems is even close to fail-proof :)
Another option would be to configure your routes. At a previous job, I set up my home router to connect to the VPN and route 10.* to the VPN interface. Setting this up isn't easy, but it's oh-so-convenient. Reading http://wiki.openwrt.org/doc/howto/vpn.client.pptp will start you on the right track.
Be careful though. This gives anyone on your home network access to work. It almost certainly violates security policy. I only did this because I knew I'd just be chastised if I got caught. (Same goes for running rogue APs at work.)
* Keep the prod DBs in an isolated VPN that requires a separate login outside of the SQL client. Stay logged out of that except when you explicitly need production access. This keeps you from casually messing with production.
* Don't save production credentials in your SQL client - uncheck the box or whatever you have to do. Probably a good idea for security anyway.
* Some clients will let you change UI for each DB. I know that SQL Server Management Studio will let you change tab and editor background color. So maybe make prod all red (or pink, or something else annoying).
* Only give a few people production logins and require them to audit everything before they run it. Actually, I'm surprised this wasn't already the case for a company dealing with health info.
I'd have an emergency procedure, sure, one where in some dire circumstance somebody could poke a hole in the firewall, change the database configuration, open a sealed envelope, and then look at/change the real data.
But in normal circumstances, anybody who really needed to see prod data would look at a read-only copy. (Or better, would look at an identity-scrambled version of it.) Any anybody who needed to change it would write a bit of code to do the work and take it through the normal review and push process.
P.S. You poor soul! I feel your pain.
Regardless, I am very glad they are now out of business. :)
We were able to restore the data pretty quickly, but we had to interrupt warehouse workflow for a few minutes. They were surprisingly accommodating, almost amused by my mistake.
The story reminds me of the day I was introduced to "BEGIN TRANS", "COMMIT" and "ROLLBACK" when someone upgraded the Sybase console and helpfully changed the default setting so we didn't need those pesky semi-colons to finish a query any more. The result was:
DELETE * FROM TABLE x
131054 rows deleted
WHERE a = "foo"
>> Malformed query <<
This stuff is way too easy for us noobs. Thank goodness that with modern technology we've found ways to make sure it doesn't happen any more... :-)
That right there is one of the worst feelings in the world. I imagine that everyone on HN just felt it with you as they were reading it!
Totally agree that the higher-ups were responsible for not putting better roadblocks in place to prevent this type of thing from happening.
When I worked at Subway, the bread dough came frozen, but you would put loaves in a proofer, proof it for a certain amount of time, and then bake it. My first shift, however, got busy and I left several trays in the proofer for a very, very long time. Consequently, they rose to roughly the size of loaves of bread, as opposed to the usual buns.
It was my very first shift alone at any job in my life, so I did the most logical thing I could think of and put the massive buns in the oven. They cooked up nicely enough and I thought I was saved. Until I tried to cut into one.
Back in that day, Subway used to cut those silly u-shaped gouges out of their buns. In retrospect, I think this was most likely a bizarre HR technique designed to weed out the real dummies, but at the time I was oblivious (likely because I was one of the dummies they should have weeded out). When I ran out of the normal bread, I grabbed one of my monstrosities, tried to cut into it, and discovered that it was not only rock hard, but the loaf broke apart as I tried to cut it.
That night, my severe shyness and social awkwardness had their first run-in with beasts known as angry customers. I was scared I would get fired, so I promptly made new buns, but spent the rest of my shift trying to get rid of my blunder. I discovered some really interesting things about people that night. First, you'd be surprised how incredibly nice customers are if you are straight up with them. Some customers I never met before met the big, crumbly buns as an adventure and, in doing so, helped me sell all the ruined buns.
In the end, I came clean (and didn't get fired). That horrible night was a huge event in the dismantling of my shell. It taught me an awful lot about ethics. And frankly, that brief experience in food service forever changed how I deal with staff in similar types of jobs.
Surprisingly it all seemed to work well. These disaster recovery steps weren't heavily tested before. Brilliant! I went to shut down the AWS instances. Kill DB group. Wait. Wait... The DB group? Wasn't it DB-test group...
I'd just killed all the production databases. And the streaming replicas. And... everything... All at the busiest time of day for our site.
Panic arose in my chest. Eyes glazed over. It's one thing to test disaster recovery when it doesn't matter, but when it suddenly does matter... I turned to the disaster recovery code I'd just been testing. I was reasonably sure it all worked... Reasonably...
Less than five minutes later, I'd spun up a brand new database cluster. The only loss was a minute or two of user transactions, which for our site wasn't too problematic.
My friends joked later that at least we now knew for sure that disaster recovery worked in production...
Lesson: When testing disaster recovery, ensure you're not actually creating a disaster in production.
'hyper-committed disaster recovery testing'
A relatively minor bug in the software that I wrote caused the safety curtain to stop triggering when a certain condition was met. We discovered this bug after an operator was injured by one of these machines. Her hand needed something like 14 stitches.
Lessons learnt:
1. Event-driven code is hard.
2. There's no difference between a 'relatively minor' bug and a major one. The damage is still the same.
I was hired by my college to build a grade management system in my second-to-last year there. I was in a hurry due to a lunch meeting with other IT staff at the University, forgot to add the where clause, and suddenly every single student was a Computational-Science major (mine).
Funny part of the story was that the moment it happened I uttered "oh shit." My boss, who sat beside me, said "what'd you do?", and about 15 IT staff from other departments walked into the office to go out for lunch. I'm sure I was an interesting shade of red.
I had to explain what I did in front of all these people. My boss laughed out loud, brought the system offline, and simply said: "well, after lunch we get to test our backup process." We went for lunch.
Two valuable lessons I learned...
People make mistakes, that isn't a problem, it's how they respond that's important.
Don't try and solve hard problems when emotions are running high. If shit is going down in production, the most important thing to do is to breathe, and get a glass of water. That little bit of time helps a lot.
> UPDATE
> UPDATE SET url='new_url' WHERE source_id IN (etc)
> UPDATE sources SET url='new_url' WHERE source_id IN (etc)
Why doesn't MySQL have a version control baked in? Even if it preserves just the last n hours of state..
https://dev.mysql.com/doc/refman/5.0/en/binary-log.html
https://dev.mysql.com/doc/refman/5.0/en/point-in-time-recove...
[1] http://dev.mysql.com/doc/refman/5.6/en/mysql-tips.html#safe-...
SELECT * FROM table AS OF TIMESTAMP some_timestamp;
Of course, you should probably do every update of production data in a transaction, check the result and then commit, and if you want to be sure, you can do UPDATE ... RETURNING to check what's changing. Autocommit on manual access to production is pretty crazy. But still, flashback is useful.
I did this in a production database (thought it was a QA environment) and brought trading on the mortgage desk of an investment bank to a grinding halt on September 14th, 2008.
The DBAs saved my 23 year old ass that day. I make it a point to send them beer on 9/14 every year.
I had had some test tables sitting around in the database for a while and decided to clean them up. I stupidly forgot to check the status of my backups; because of an earlier error, they were not being correctly saved.
So, I had a bunch of tables with similar names:
users_1024
users_1025
users_1026
Guess what got deleted along with them? The actual users table (which I've since renamed to something that does not even contain "users" in it).
So, how do you recover a users table when you've just deleted it and your backup has failed?
Well, I happened to have all of my users' email addresses stored in a separate mailing list table, but that table did not store their associated user IDs.
So I sent them all an email, prompting them to visit a password reset page.
When they visited the page, if their user ID was stored in a cookie -- and for most of them, it was -- I was able to re-associate their user ID with their email address, prompt them to select a new password, and essentially restore their account activity.
There was a small subset of users who did not have their user IDs stored in a cookie, though.
Here's how I tackled that problem:
Because the bulk of a user's activity on the site involves answering poll questions, I prompted them to select some poll questions that they had answered previously, and that they were certain they could answer again in the same way. I was then able to compare their answers to the list of previous responses and narrow down the possibilities. Once I had narrowed it down to a single user, I prompted them to answer a few more "challenge" questions from that user's history, to make sure that the match was correct. (Of course, that type of strategy would not work for a website where you have to be 100% sure, rather than, say, 98% sure, that you've matched the correct person to the account.)
Now, if you go to a CNET site and view source, there's a <!-- Chewie loves you --> comment. I like to think of that as an homage to my original fuckup.
Because of me, one December, everyone in the country who went to the cinema got to watch anywhere between 30 and 45 minutes of ads before the main presentation started.
Lesson learned: write more tests, monitor everything.
My story (though I wasn't directly responsible): we were delivering our software to an obscure government agency. Based on our recommendation, they had ordered a couple of SGI boxes. I wrote the installation script, which copied stuff off the CD, etc. Being a tcsh afficianado, I decided to write it in tcsh with the shebang line
#!/usr/local/bin/tcsh
In the script, the first few lines were something like:
set HOME = "/some/location"
/bin/rm -rf $HOME/*
In the rm line of the snippet above, "/some/location". Magic variables in bash tend to lose their magic once set.
In Bash, it doesn't do anything useful.
> In Bash, it doesn't do anything useful.
on IRIX, $HOME for root is almost certainly root. So the resulting command is rm -rf /
What I said about un-magic-ing is true for sure. "HOME=foo; echo $HOME" will print "foo".
Interestingly, it's propagated. "HOME=foo sh -c 'echo $HOME'" will also print "foo".
We were storing payment details sent from a PHP system into a Ruby system, I was responsible for the sending and receiving endpoints. Everything was heavily tested on the Ruby end but the PHP end was a legacy system with no testing framework. Since the details were encrypted on the Ruby end, I didn't do a full test from end to end AND unencrypt the stored results.
Turns out for two months we were storing the string '[Array]' as peoples payment details.
Takeaway: If you're doing an end to end test, make sure you go all the way to the end.
~ 2007, working in a large bioinformatics group with our own very powerful cluster, mainly used for protein folding. Example job: fold every protein from a predicted coding region in a given genome. I was mostly doing graph analysis on metabolic and genetic networks though, and writing everything in Perl.
I had a research deadline coming up in a month, but I was also about to go on a hunting trip and be incommunicado for two weeks. I had to kick off a large job (about 75,000 total tasks) but I figured spread over our 8,000 node cluster it would be okay (GPFS storage, set up for us by IBM). I kicked off the jobs as I walked out the door for the woods.
Except I had been doing all my testing of those jobs locally, and my Perl environment was configured slightly differently on the cluster, so while I was running through billions of iterations on each node I was writing the same warning to STDOUT, over and over. It filled up the disks everywhere and caused an epic I/O traffic jam that crashed every single long-running protein folding job. The disk space issues caused some interesting edge cases and it was basically a few days before the cluster would function properly and not lose data or crash jobs. The best part was that I was totally unreachable and thus no one could vent their ire, causing me to return happy and well-rested to an overworked office brimming with fermented ill-will. And I didn't get my own calculations done either, causing me to miss a deadline.
Lessons learned:
1) PRODUCTION != DEVELOPMENT ever ever ever ever 2) Big jobs should be proceeded by small but qualitatively identical test jobs 3) Don't launch any multi-day builds on a Friday 4) Know what your resource consumption will mean for your colleagues in the best and worst cases 5) Make sure any bad code you've written has been aired out before you go on vacation 6) Don't use Perl when what you really needed was Hadoop
I woke up the next morning to several inbox screens' worth of messages from angry people I didn't know, demanding explanations for what I did to their jobs and their cluster. I don't think I have ever biked to the lab faster.
After multiple rounds of palm-drenching emails with the cluster sysadmins and the computational mathematics group PI (and my own boss agonizingly cc'ed), we determined the cause. The cluster sysadmins, lacking imagination for the destructive naivete of their users, had not foreseen that anyone would want to submit more than 10^4 jobs at once. That broke the scheduler, preventing other people from running jobs and me from canceling them. Meanwhile the blast jobs blew past the disk quota, leading to a Hellerian impasse where I somehow lacked the space to delete files so I could create space. I still don't fully understand it.
I believe it took a full day to get the cluster back online.
Second web related job at an insurance company, I was 20 years old at the time. We were heavy into online advertising, mostly banners at the time (this was right around when adwords started to get big). The company just bought out all of the MSN finance section of their site for the day-- it was a pretty big campaign ($100,000). We drove all the traffic to a landing page I had created with a short form to "Get a quote".
IT had given me permissions to push things live for quick fixes and such, I made a last minute design tweak and, you guessed it, broke something. I was checking click traffic and inbound leads and realized traffic was through the roof but leads were non-existent. This was about 45 minutes after the campaign was turned on. I jumped on the page and tested it out and got an error on submit. FUCK. I literally started to perspiration INSTANTLY.
Jumped into my form and quickly found the bug, can't recall what it was but something small and stupid, then pushed it live without telling a soul. Tested, worked, re-tested, worked. Ran some quick numbers to get a ballpark estimate on the damage I caused... several thousand.
Stood up and walked over to the two IT guys, mentioned I borked things and that I had fixed it... what should I do? I can still see the look on their faces. Shock, then smiles. Walked back to my desk and about 10 minutes later my two bosses show up (I worked for both dev & marketing managers).
They said thanks for catching the problem, not to worry. I did good for finding it myself, fixing it, and pushing it live. I was still sweating and shaking. They walk off and later that day marketing manager informs me MSN will refund us for the 45 minutes of clicks.
It took about a month before I felt competent enough to touch our forms again.
After the test was complete, I forgot to turn off the Adwords. (Such a silly mistake...) Nobody notices until our bill arrives from Google, and it's substantially higher than normal. When my coworker came to ask me about it, "are these your campaigns?!?" I just sank in my chair.
I think it cost the company $30k. I suppose it's not that much money in the grand scheme of things, but I felt very bad.
Queue me 7mos later reviewing the system. Realizing that critical jobs were no longer running and that our users were all essentially receiving 100% free hosting for however much storage they wanted. SOOOO i turned the jobs back on.
The lead engineer before me left no documentation of what the jobs did other than that they should be run. In my stupor i did not review the code. The jobs sent out a blast of emails warning that files would be deleted if not cleaned up or maintained. Then seconds later deleted said files...
We nuked around 70GB worth of files before we realized what happened. WELL GET THE TAPES! Turns out our lead engineer ALSO forgot to follow up w/ system engineers and the backups were pointed at the wrong storage.
No jobs lost, thankfully the manager at the time was a word smith of the highest degree and can play political baseball like a GOD.
# cd /etc
# emacs inetd.conf
# ls
...
... inetd.conf
... inetd.conf~
...
# rm * ~
# ls
# ls $ ls
$ sleep 5
$ ls1) after few months working in a bank, I was doing some simple admin check task via RDP to a Windows 2003 (no, maybe 2000) server, when I right-clicked the network icon and instead of clicking the properties options i clicked "disable". Just the time to say "oh sh!t" and to realise that it was the production Trading On Line machine, on a remote datacenter, during market hours, and to discover couple of minutes later that the KVM over IP was crappy and was not working. We had to call the datacenter operators to go back to the local KVM and re-enable the NIC.
Lesson 1: Better move slowly when you're on a production machine (and also have plan B and C to reach your machines is a good idea)
2) same bank, one or two years later, I was doing some testing on a new mail system that integrated also VoIP (SIP). Mail/SIP System running in a VM (I think Vmware Server at that time) in the same remote datacenter as above. So, I enable the SIP feature and after few seconds, bum, we lose the whole (production) datacenter and the connection between the local server room and the datacenter. Panic, I look at my colleague, WTF in stereo, everything come back for few sec, bum again down. Long story short, the issue was that that version of Netscreen firewall ScreenOS had a buggy ALG implementation for SIP that lead to core dumps. The fun thing is that we had two of those in HA, same version of course, so they were bouncing between core dumping, rebooting slave becoming master and then core dumping again etc.. We had to ask a datacenter operator to reach the rack, disconnect one of the cables from the firewall (the one that was managing the traffic of the DMZ where that machine was hosted) and then reach the virtual host to kill the machine.
Lesson 2: you can segment your network but if everything is connected through the same device(s), sh!t can still hit the fan...
On a Solaris box.
Hilarity ensued when we next rebooted it.
It wasn't a but an hour before I lost sysadmin privileges.
Never "experiment" with a production system - ever.
I was commissioning a new control system at a power plant's water treatment facility. I was fairly new to the industry and had mostly looked over the guy who did the bulk of the work's shoulder as on the job training.
This particular day the guy was out sick and we had to finalize a couple of things before we ran through the final tests.
There was an instruction to open a valve to fill a tank and it had the wrong variable linked to it. The problem was to maintain the naming standards I had to do a download to the processor to make the change. When I had been doing work in the office this was not a big deal, download the program to the processor, it stops running for a moment while it loads the new logic into memory and starts back up.
Not thinking through the implications of the processor shutting down while the process was up and running I made the code changes, hit download and about 30 seconds later an operator came running over looking like he had seen a ghost and he was pissed.
While I was making my code changes the operator was hooking up a hose to drain a rail car of some chemicals. The way the valves were configured before I made my changes was correct and would have had no consequence it I didn't touch anything. The way the valves were configured when the processor restarted would have routed the rail car's contents to the wrong tank resulting in a reaction which would have created a huge plume of highly toxic gas. The way the wind was blowing this plume would have blown directly to the largest town in the area and could have killed a ton of people.
The operator heard the valves in question changing position before he opened the valve on his hose to empty the rail car and figured something was up. When he saw the whole process had shut down he got really angry because I had ignored the protocol in place to avoid such a disaster.
I got chewed out and kicked off the site. My boss attributed my mistake to inexperience and I had to give a safety presentation on what I did wrong.
Lessons learned: Be sure you are aware of any implications your actions have. If you are unsure or guessing about something stop what you are doing and go ask someone first.
Don't give people mission critical work on their first project and have them work unsupervised. Training is important.
Always be aware of safety requirements, especially when you are working with machinery, automated processes, chemicals or anything else that can hurt, maim or kill you.
Me: No we don't. We have 121 bad orders.
Boss: There are thousands of them!
Me: No there aren't. There are exactly 121 of them. I'm sure.
Boss: I'm not going to argue with you!
Me: Good. Because you'd lose.
I fixed 121 orders that night. The next day my login & password wouldn't work.
In theory you are supposed to be able to tell your boss he is wrong (and a good boss will appreciate this) in practice it depends on the boss.
That's how they "notified" people. You got to play a lucky jackpot game of logging into your desktop after the meeting to see if you still had a job. Straight out of Dilbert.
I was doing HVAC work while I was in college and we were removing an old air handler from underneath a house. Just inside the crawl space, under the access door was a water pipe. My boss told me to make sure I held it down while we slid the air handler out through the hole. I lost my grip on the pipe and the air handle snapped it in two, at which point gallons of water began to gush into the crawl space.
I ran for all I was worth to the road, which in this case was about 600 feet away, to turn off the water at the water meter. I ran up and down the road in front of the house and never found the water meter. So I ran back to the house and inside and told the homeowner who promptly informed me that they used well water. She called her husband and he told us where to turn off the well pump.
It wasn't really that bad in the grand scheme of things but letting the homeowner's water gush under the house for about 15 minutes does not bode well when you are supposed to be there to fix problems not create them.
The next morning, the bunker was full to ground level and the automatic power cutoff had failed, as the float switch was directly under the cable duct and the water pressure of the deluge and kept the float depressed. By the time the water stopped flowing the float was under a foot of mud. The powered circuits were undergoing electrolysis and eating themselves away, made worse the the site managers refusing to drain the bunker or turn off the power until a week long arse-covering evaluation had been completed.
A few hundred million dollars of front line radar was out of action for several months.
Being a naive newly graduated engineer, I wrote a completely honest report and analysis. My boss said it was one of the best reports he had read and there was no impact on my career (if anything it got me noticed by the upper echelons of the organisation).
Lessons:
1. If you tell the truth you will be respected, even if it is incriminating.
2. If there is a way for something to go wrong it can do so (slight variation of Murphy's Law). Even if it's judged to be uneconomic to take preventative action, be aware of the possibilities, so you can make a conscious decision about the risk.
Reaction was standard: mostly to point out I did my best in unfamiliar territory and things should be sorted soon.
Take aways were: (1) less support calls than expected - users put up with things. (2) you learn when you fail (3) always have a backup
They kept me on at that job but I left pretty soon anyway as I got a 'real' (as in creative) job hacking perl-powered VPN modules for those Cobalt Raq/Qube devices, and building a Linux-related online retail venture for the same employer ... that worked great, but failed commercially.
The PM brought the issue to the CTO, but somehow I didn't get fired. Ended up apologizing (obviously a poor choice of words :)) and moved on. Never made that infrastructure change.
Key takeaway: if you're going to talk shit, don't do so in writing. ;)
Key takeaway: always check the cam.
This sounds like a simple task, but it gets complicated by the variety of pipe fittings and adapters available. Our sensors are a particular thread type, and we have to find a free slot to install them, and come up with any pipe connection converters necessary to install them there. Another tricky part is that the rig workers who actually know about all of this stuff are often not particularly eager to help out.
So on one particular job, the only free slot to install the sensor is a male pipe fitting, capped with some sort of female plug. Our sensors are male in that pipe size, so I need a female-female adapter to install it. I go looking around and come up with one, not paying too much attention to it. I install it, and everything seems to go more or less smoothly. We go on drilling with this installed for like a week or two.
One day, the rig manager comes to find me and ask me about this adapter that I used. He tells me that it is meant for drinking water lines, and is only rated to 200 psi. And had been installed on a 2000 psi line for weeks. My jaw dropped in shock - I have no idea how that adapter didn't fail, and it's entirely possible it could have hurt or killed somebody if it did.
They sent one of their guys to find an adapter that was actually rated for the pressure and replace it, and never said much else of it. No telling how much trouble I could have been in there if anything else had happened. It did make me a lot more safety-conscious.
It wouldn't be a big deal, wasn't for the fact it was an EC2 instance, and back then halting the instance was equivalent to deleting it permanently. We then spent the night at the office recovering and testing the server. I think we left 3:00 AM that day.
Lesson #1: it's never a good idea to "shutdown -h now" on a shell. any shell.
Lesson #2: have the process to spin up a new production server fully automated and tested
http://manpages.ubuntu.com/manpages/trusty/man8/molly-guard....
Personally, I'd think that training this is a lot more portable.
# hostname
<foo>
# shutdown -h nowThe lesson from this is pretty obvious. Backup. Make sure your backup is good and safe.
My worst work-related mistake was getting into business with a friend. It cost me the friendship, a very valuable client, and a good portion of my retirement savings. I'm not sure how related it was, but a few years later my (former) friend killed himself.
And the lesson here is not to go into business with friends. Or at least to set up the business as if you're not friends.
Most servers had those hot swap drive bays for convenient access from the front while the server was running. You only had to make sure no write operation occurred while you pulled the drive out of the bay.
So, I had to exchange a backup disk on a database server running quite a few rather large forums. The server had two disk bays: One for the live hard disk and one for the backup disk. I was absolutely sure at that time which one was the backup disk so I didn't bother to shut down the database server and incur a minimal downtime. Of course, I was wrong and blithely yanked the live disk from the drive bay.
I spent the rest of the night and most of the following day running various MySQL database table repair magic. It worked out surprisingly well but having to admit this error to our forum users was embarrassing, nonetheless.
Lesson: Appropriately label your servers and devices.
The billing specs kept changing, as did the specs for the show itself. New price points, more plans, change the show interface, add another option here, etc. The plan had been to do a free preview show the day before to work out the kinks. That didn't happen.
The time leading up to show start was pretty tense, lots of updates, even a few last minute changes! Then the show actually started, brief relief. The chat system built in started deleting messages, one of those last minute feature changes had screwed up automatic old-message deletion. We had a fix though, update the JS, and bounce everyone out of the show and back in so the JS updates. Fixed!
Then the CEO pointed out that the quality just kept getting worse. Turns out that while the video player had both a numeric value and a string description for the different quality levels, it assumed they were in ascending order. So once it confirmed it could stream well at a given level, it automatically tried the next, which worked! Poor quality for everyone. Fixed, and another bounce.
Then it was over, time to go home. Back in the next day to finish off the billing code. I decided to approach it like a time card system. Traverse the logs in order, recording punch in time, when someone punches out, look up their punch-in times and set that user's time spent to the difference. Remove punch-in and out from the current record so they're not used again.
Now two facts from above added up to a pretty serious bug. 1) I _set_ the time spent to the difference between the two times. Not added, set. 2) We bounced everyone from the show twice to update their JS, and video player. So everyone had multiple join/parts.
I under-billed customers by tens of thousands of dollars.
Things I learned:
- Don't just argue that you need a trial run, make sure management understands the benefits. Why, not What.
- Duplicate billing code. After that a co-worker and I wrote two separate billing parsers for things, 1 designed to be different, not efficient.
- Give yourself ways to fix problems after they crop up. The bounce killed my billing code, but not doing it would have damaged the actual product (which later became a regular feature). Wish that thing had been my idea.
I list the content of my home directory trying to understand which folder was so big. Then I see it. A folder usually empty. Empty because I use it as generic mount point. A mount point that the day before was attached via sshfs to the production server...
I had a strange feeling, like if I was seeing myself from behind, something crumbling inside me. And at that moment someone start to ask "what's happened to <hostname>"?
I take my courage and I say "I know it"...
That was really hard. The worst day at work in years, and during the last day too. Luckily we had a good enough backup strategy and the damage was mostly solved in a couple hours.
There I realized how much of an idiot I was to have mounted the production server on my home and I grow a little.
I lost $7 million dollars in minutes by being short $700 million of US 2yr notes when the levees failed during the hurricane Katrina disaster.
Although my bet that the 2y point would be under pressure in the intermediate term turned out to be true, I got carried out by fund flows as folks spazzed out to cut risk by rolling into short duration high quality paper.
To his credit, my boss, who sat across from me, said only: "wouldn't want to be short 2 years." He let me make the call, which I did, and I covered my position. (Ouch.)
My book was up considerably on the year already, but this was a huge hit, and nearing year-end. I dialed back the risk of my portfolio and traded mostly convex instruments (options) for the remainder of the year.
yup. that really happened. it was 4-5am in the morning and I'd been working all night. I was on the server trying to set something up and was trying to blow away a folder ... I did a normal rm and that didn't work (obviously) because there was crap in the folder. So I pulled out my nuclear weapon to nuke the folder but left off the preceding ./ (which still wasn't that smart anyway) ... I sat there for a second wondering why the deletion was taking so long ... then another 30 then a minute ... then I looked at what I'd just typed again ... then I realized what had happened.
ctrl-c'ed (or d, can't remember now) out of it. then tried to find root folders
cd /etc => folder not found
cd /var => folder not found
I'm from a third world country where we laugh at Americans (sorry) for throwing up when they're nervous or having panic attacks, but at that moment, I had a full blown panic attack. I'll never forget it.
The work was a subcontract for a client who was doing work for Nike, and it was a decently sized project that was critical to the success of the firm, and I'd just blown away their live production server ...
Afer freaking out and almost crying for 5 minutes. I decided to call media temple support (we were using one of their vps servers) ... and by the biggest absolute stroke of luck they'd just backed up the entire server ... not even 2 hours prior to my madness. $100 for a full restore (I don't recall why) and would I like to do that?
HECK YES I WOULD!
so they restored the server for me. I wrote an email to the head of the small company I was doing all the work for, explaining what I had happened and telling him I'd sent over a check for $100 to cover the backup because it was my fault. He was obviously very relieved and never cashed the check I sent.
I still get chills thinking about that exact moment when I thought I'd fucked up my career and reputation for good.
They had ASIC design runs for research purposes once every three months, yielding your design on Silicon as ten 6" wafers. It gives enough parts for testing the first revision of your design. The person was carrying the wafers to a vendor for cutting into separate ICs and packaging or something. Gets to the parking lot, and where are the keys. Puts the wafers on the top of the car, finds the keys in his pockets and starts driving. Boom, the box of wafers was still on the top of the car, now on the ground. All broken. Some $100K in wafers + three months lost + bad face before the customer + ... Lesson: Don't put stuff on the top of the car!
"Shit. Well, as I have just demonstrated, it becomes possible to wipe out a million user login credentials at the touch of a button. So now we'll be needing to restore that from the backups which we don't have." Luckily, and ONLY BY CHANCE, I happened to have a copy of that table exported for other reasons from a few days back.
Lessons learned: Never press enter.
This wasn't a mistake; just a hilariously successful penetration test!
I now never execute ! commands as root. Actually, nowadays I simply use CTRL-r.
Users were mapped into specific silos to separate out each level of the stack from CDN to storage to db. There was a bit of code executed at the beginning of each request that figured out if a request was on the proper subdomain for the resource being requested.
This was a feature that was always tricky to test, and when I joined the codebase didn't have any real automated tests at all. We were on a deploy schedule of every morning, first thing (or earlier, sometimes as early as 4am local time).
By the time the code made it out to all the servers, the ops team was calling frantically saying the power load on the strips and at the distribution point was near critical.
What happened: the code caused every user (well upwards of millions daily) to enter an infinite redirect, very quickly DoSing our servers. It took a second to realize where the problem was, but I quickly committed the fix and the issue was resolved.
Why it happened: a pretty simple string comparison was being done improperly, the fix was at most 1 line (I can't remember the exact fix). There was no automation, and testing it was difficult enough that we just didn't test it.
What I learned: If its complicated enough to not want to test using a browser, at least always build automation to test your assumptions. Or have some damn tests period. We built a procedure for testing those silos with a real browser as well.
I got a good bit of teasing for nearly burning down the datacenter on my very first code deploy, but ever since, its been assumed that if its your first deploy, you're going to break something. Its a rite of passage.
Within about 2 minutes CTO strolls in asking about the flood of exception emails due to each request being unable to connect to the database.
Thankfully, I was able to apt-get install mysql-server, all the data was still there, and things were back to normal within 5 minutes.
1. I passed the resume and chat portion
2. I passed the telephone questionnaire and got along great with the interviewer
3. (Fail) I scheduled my interview on a Friday at 4:30pm and there is a 30 min travel time. I left 1hr early...still it was Memorial Day weekend, so I thought the streets would be quicker than the freeway since it was at a stand still. I was so stressed that I literally had an anxiety attack and couldn't even find the address. Never happened to me before, so I'll never forget it.
So one of the first things I wanted to do was setup a development db for which I exported the structure from their prod db. I then proceeded to change the name of the create database statement at the top to the new dev db I wanted and ran the script.
Unfortunately the prod db name was still pretended to every drop and create table command in the script so I had just replaced their whole prod db with an empty one.
Owning up to that was one of the most embarrassing moments of my career. It was such a rookie mistake I just wanted to die. Luckily they had daily backups so I only cost their 4 man business about half a day of work but... it was enough for me to be a much more careful developer from that day forward!
me: "unix definitely won't just let me cat /dev/urandom > /dev/sda"
other: "sure it will"
me: <presses enter>
what I learned? unix will absolutely let you hang yourself. 1998, production server for a fortune 5 company.
Linux will indeed let you hang yourself. I maintain that every experienced Linux user has seriously messed up their system at least once. More often personal / dev / test environment than production though...
I was off Friday, so I come in Monday morning to see that ~20k customers have been getting free stuff since Thursday lunchtime.
Lost something like $200k because of two nullable columns :(
edit: This was after I asked for permission to do this.
Lesson learned: Don't EVER use Coldfusion as a web server.
We had the tech support contract for the city's Mexican consulate. One of the things we were doing was patching and updating their server and installing a tape drive backup system. Server was NT4.
I'm in there doing work after 5pm, and wrongly assume that everyone's gone home for the day. Install some patches and the server asks me if I want to reboot. I say yes. Few moments later, a guy sticks his head into the server room and asks if I'd shut down or rebooted the server. Oh, whoops, someone's here. Yeah, I just installed some patches. Oh, OK, see ya.
Next day? Turns out he had been doing some work in their database where they track and manage visa applications. That database got corrupted when I did the server reboot while he was doing his work. That night, the backup process then overwrote the previous good copy database on the tape drive with the newly corrupted database. We had not yet started rolling over multiple tapes to prevent backups of corrupt data, though we were going to purchase some tapes for that purpose shortly.
Summer was ending, and I quit a week later to return to school. Horrible timing in terms of quitting! No idea what happened after that, as I was spending the summer in a city that was not my own. I do know that the original database developer contractor was on vacation at the time and so they couldn't reach him. I think the consulate was SOL. I regret rebooting that server without checking if anyone was working to this day.
Lesson learned? Don't assume anything when doing anything. Carried that lesson with me for the rest of my life. And find a boss who knows how to guide you if you don't have much experience in your area. I guess for founding startups, at least get an advisor.
edit: spelling
Unfortunately I got my selection criteria wrong and pulled out all of one cluster and half of a second, halting a few thousand operations.
Luckily the monitoring system was very quick to alert me of this and using the same (wrong) selection criteria it was a fairly simple process to stop the update and put them all back in the cluster.
Takeaways? The age old cliche of "With great power comes great responsibility". Oh and have good monitoring!
I'm still not sure how this bug slipped past the bank's tough app certification process, though.
I left that job about 3 years later when the metaphorical train stopped at a nicer place. My name is still known in certain circles for this ["Oh bah, how could I forget?" one former manager recently stated], but I don't plan to go back there at this time.
I learned that life's too short for assholes and working in an environment you don't like. If you don't screw up, your soul will die and you'll become that former coworker you hated so much and who hated you in return. It's worth picking and choosing where you work.
Thankfully, someone stopped me before I turned it on.
A friend had referred me for a sysadmin job opening at a web hosting company in Florida. After a brief interview I got the job for a pretty decent salary and was told when I could start. What they hadn't told me was that my schedule would be tuesday to saturday. I had informed the hiring manager of my preferred schedule (monday-friday), but I guess nobody mentioned it to the manager of the group.
When I got there they told me my schedule and I immediately told them that's not what I signed up for. So they asked me to sit for a while so they could figure out what to do next. I took a tour of the NOC, and saw one of their tier 1 technicians was chatting and watching a movie. I walked up and asked him "Heyya! Workin' hard, or hardly workin'?" and smiled. He did not smile back. So I went back to the desk I was assigned to, which was already logged in - with the credentials of the previous admin.
While I waited I decided to see what other trouble I could get into. Sure enough, all the old passwords were saved in the old admin's browser with no master password. I couldn't copy-paste the list, so I took a screenshot and began to find a way to print the list out to post on my cube wall. Before I could finish I was asked to leave for the day while they figured out my schedule changes. I should have gotten the hint when they asked me to leave the badge there.
Later I got a voicemail telling me they'd pay me for the time I spent there (about three hours) and they'd no longer require my services. Luckily I got hired soon after to a different company, which was also hiring away all the talented people from the place that had let me go, and the web hosting company eventually went under. So it turned out to be a good thing in the end.
I was working at a startup that was trying to create an affordable 3D printer. We had two working prototypes that were used for everything - demos, print testing, software testing, PR shoots, everything. Each prototype had cost hundreds of man hours to build and debug and quite a bit of cash as well.
Among other things I had done all the work on the thermal control system for the printer, it kept the print heads and build chamber at the correct temperature. One night while working on one of the printers I hit an edge case that my control code didn't handle well and the printer turned all of the heaters on full-bore. Half an hour all the plastics in the prototype had either melted or burned and I was left with a room full of smoke and a pile of scrap aluminum.
1. First day at a job. I need to get familiar with a legacy system and get a SQL dump from it to create a local copy of the database. After some SSHing and MySQLing, I confuse my two split terminal panes and end up importing my local dump to production server. Of course database names and users were the same so I end up dropping the database. No biggie. Backups were available from previous day.
2. Similar story to the first one. I got a new shiny Zend Studio IDE. Want to set up sync with remote server (just a static company website with no version control). Fill all the settings, press the sync button - and what happens? Zend Studio somehow figured that I want to force sync my local folder, which is empty, to the remote site, and it just deletes everything on the web root and uploads my empty folder. Wat. Should have read the settings twice.
We were rearranging the layout of the office. Coworker was moving in to his new space, setting up his desk. He boots up his computer, wonders why he has no network. Looks around, discovers the ethernet cable isn't plugged in. Plugs it in to the wall, still has no network.
A few minutes pass, and the entire office is running around wondering why the hell the network isn't working. Maybe an hour passes, the network guys are losing their shit trying to hunt down what is wrong. I'll give you a hint: the router was lit up like a Christmas tree, and the aforementioned coworker had both ends of his ethernet cable plugged in--but neither end was attached to his computer.
I had to quickly get a patch in for the improper code and had to maintain that buggy implementation. In addition, the "standard" itself got a rather scathing write up from Peter Gutmann, which is completely valid:
https://www.cs.auckland.ac.nz/~pgut001/pubs/pfx.html
This is a critique on the "standard" itself, the process was just as ugly.
Luckily, there was no slap on the wrist or anything, the store manager knew that after doing thousands of these cards this was only one of a few slip ups I've made so they just brushed it off and moved on.
9 hours later I wake up to check my inbox has 800+ emails. Django by default sends out email when an error occurs and a tiny mistake of not installing a package led to a lot of frustrated customers and well a huge pile of email in my inbox !
Moral of the story: Put that pip freeze > requirements.txt and pip install -r requirements.txt into your deployment flow.
Some how in settings, we had flag "Store Credit Card Info" as "Plain Text" enabled. The Admin/Staff of that client could have use this information to make transactions (As in Backend it would show Full CC info into order details)
We didnt realized untill we worked on it again for some bug fixes and adding new features.
Lesson Learned :- When transitioning from DEV to PROD env, make sure to check all these critical flags and correctly set
Luckily, the client didnt had any idea about what was wrong in backend.
Lesson: Keep the code that touches production databases as simple as possible so it's easy to verify exactly what it does. I was using a framework's database tooling incorrectly because I never dreamed what I used would touch the databases's counters.
(Not my worst mistake in terms of people affected, but it's the only mistake that was literally laser etched in metal forever.)
All in, took 4 days and a new server where the hard drive had stored bad pages on the DB. We lost 2 days of orders (they were processed through to the internal systems though so not really lost)
Lesson learned, validate backups and check page integrity when backing up
Every database alias I have now has the MySQL --i-am-a-dummy flag appended. This has been a career-saver in my eyes.
One day I was doing a change control, I was scheduled to change some settings on the modulator of a satellite system providing internet access to a portion of the Middle East. I called the satellite operator and told them I wanted to do the scheduled work, as they would have to confirm I was outputting new configuration within the constraints of the contract.
I entered the change and re-initiated the modulation, the operative said he was seeing nothing. Now, because this signal was going to the middle east I couldn't see it in Europe and without substantial plumbing I couldn't tap in to the antenna. My heart started going, I was checking amplifiers, up-converters, everything and I couldn't see anything wrong. After a few minutes the client called panicking because the action should have taken moments, not minutes. After more confirming with the satellite company that I wasn't transmitting I checked back through my steps and eventually saw I had missed one crucial thing: when you changed certain parameters it muted the output! A quick few button presses and the patient man on the phone said "there it is!" and I can relax again.
Lessons learnt: 1) I should have noticed a critical LED on the modulator was not lit! 2) This is the reason change controls say "working period 10 minutes" (time taken 9m50s). 3) A good boss will defend you if you recognise your mistakes and don't f*ck up too badly. 4) Don't go for a quick drink with your brother before a night shift.
In the broadcast industry they say you aren't a real engineer until you've taken a TV network off air. Lets say I am very experienced, but my employers have never had a problem with me. This is probably because I have also seen people meltdown under the pressure of delivering live services to millions or even billions of viewers, but if you can keep cool you can deal with it.
Thoughtful pause "Why is this taking so long!?"
"OH FUCK"
In production.
I'm the DBA.
Poof. Equipment electronics fried and useless.
I was chewed out. Could have been way worse.
Follow your safety procedures.
Two months later, the certs were expiring soon and we changed our configuration to something Android liked by default. The bad news was that our production Android app rejected the new configuration and only wanted to accept the current certs.
We ended up quickly shipping a hotfix that accepted the current and upcoming configuration a few days before the certs expired. There technically wasn't any 'downtime' as long as users updated the app, but this all took place right before 'holiday vacations', and the QA team had to test the fix while all the devs were away.
EDIT: I proposed a new password of: @$tevezA$$ignedPwD@# (Steve's Assigned Password)
He said no to that one.
The screw-up is the boss in this one.
Possible outcomes of unplanned system haults include plugged machinery that would need to be manually cleared, mixed products which would become immediate net losses for the company and damaged motors.
Thankfully no product was being run at the time. I have also implemented changes across the board to our client sites that prevent this type of shit from ever happening again. You know when you look at a system and go "this is going to bite us in the ass eventually?" This was one of those systems, they just needed a new hire to give them the push.
learning from this experience: never do an "rm" on the log file, instead do "truncate -s 0" on the log file.
now on to my tasks.. had some files to print out. Where did they g...... FUCK.
I found a box of tapes and some sunos manuals. Spent the next several hours figuring out how tar and tape drives worked. Got everything back. Never told a soul.
1992. I've never done anything so careless since.
Luckily we had backups from that morning so we only lost any address updates people would have done that day, but it made for some interesting customer service calls for awhile...
Oh yea, I run a proprietary trading firm (still at the same spot), as a result of that bug we went down and lost about $250k over the next few hours. Testing is important in automated trading :)
I have no idea why they didn't use UPS, but it took many critical servers offline and caused a few hours of headaches for everyone.
Come to think of it, that was the last time I was allowed in the server room.
Lessons learned - don't let developers in the server room.
After a couple hours of swearing, instead of working from a root shell in my own account, I just logged into the GUI as root. And there was a pretty interface showing the disks. I could just click on one and format it. Hooray!
Well either the GUI was buggy or I clicked on the wrong disk, because as the format was going, I realized the external drive wasn't doing anything. I was formatting the internal boot hard drive. And since nobody but me gave a crap about this weird free box somebody had given them, they had repurposed it. As a file server. For the home directories of a bunch of my colleagues. Who were now collecting around me wondering what was going on. Oops.
No problem, says I. I'll just restore from backups. But this thing used a weird magneto-optical drive [2]. The only boot media we had was on an MO disk. The backups were on another. And there was only one of these drives, probably only one in the whole state. The drives were, of course, incredibly slow, especially if you needed to swap disks. Which, I eventually discovered, I would have to do about a million times to have a hope of recovery.
Long story short, I spent 28 hours in a row in that chair. It was my immersion baptism [2] in the ways of being a sysadmin. The things I learned:
Fear the root shell. It should be treated with as much caution as a live snake.
Have backups. People will do dumb things; be ready.
A backup plan where you have never tried restoring anything may lead to more excitement than you want.
Be suspicious of GUI admin tools. Avoid new GUI admin tools if at all possible. Let somebody else be the one to discover the dangerous flaws.
If you were smart enough to break something, you're smart enough to fix it. Don't give up.
When some young idiot fucks up, check to make sure that they are sufficiently freaked out. If they are, no need to yell at them. Instead support them in solving the problem.
Seriously, my colleagues were awesome about this. I went on to become an actual paid sysadmin, and spent many years enjoying the work. The experience taught me fear, and a level of care that sticks with me today. I'm sure at the time I was wishing somebody would wave a magic wand and make it the problems go away, but working through it gave me a level of comfort in apparent disasters that has been helpful many times since.
[1] http://en.wikipedia.org/wiki/NeXTcube [2] http://en.wikipedia.org/wiki/Magneto-optical_drive [3] http://en.wikipedia.org/wiki/Immersion_baptism
My first job was as a Marine avionics technician and I had the misfortune to work for bosses who reassured themselves by blaming you at full volume and making you feel like shit. Usually with the 'you could have killed somebody you useless idiot' which is actually not that often true. Not a great environment.
Last week I'm just starting my second year at this company as a real proper electrical engineer and I neglected to run all the steps of the pre-tests I said I ran and I therefore missed a (in retrospect, really cool) bug that showed up in front of the certifying examiner on a Friday afternoon. Said examiner costs a lot of money per hour, it's a government agency, etc. etc. I spent a weekend and Monday finding the bug and squashing it. It took til Friday before the new code was released and all the paperwork up to the certifying agency was fixed.
The whole time my boss was there and clearly frustrated but also very clearly keeping quiet. I definitely went and thanked him on Friday afternoon when it was all over for being so chill about it. He said it was no problem as he could see I had sufficiently embarrassed myself and was working hard to fix it. It's so true. I was keeping my head down, being calm but intent on the problem, and I don't know how I could've solved anything if he'd lost his shit.
Fear and care are definitely with me now. As is proper testing and better logging of my own day-to-day dev process. (Note to self: power cycle in the test when it tells you to power cycle. You're also testing the non-obvious functionality.)
Gratitude is demonstrated through actions, not vague verbal commitments.
But this was one of my first. Years ago, making boot floppies for a physics lab where I was reinstalling all the servers:
I meant: dd if=/dev/zero of=/dev/fd0
I did: dd if=/dev/zero of=/def/hda
Oops. Bye, partition table.
(Always double-check everything you type as root.)
Somehow I managed to salvage the situation using the man pages for the filesystem, the C compiler, and making sure I did not reboot. Really don't remember much about that late night...
Said person entered the number of metric tons of concrete 3 magnitudes higher than it should have been. Imagine the cost difference between 1.0 * 10^6 and 1.0 * 10^9 metric tons... Our boss was not pleased, to say the least.
But imagine how easy it is to enter a few extra zeros in an excel data cell. Yikes!
I was a stock analyst, for a firm with dozens of institutional salesmen and thousands of retail brokers. Some of my recommendations were very, very wrong.
The right thing to do is stand up, take the heat, and explain what you now know as best you can. I learned that watching a colleague who I thought was otherwise an unserious ass.
Takeaway: Sometimes, it takes a disaster to realize you were in another disaster anyways.
As it turned out the only data that did go out was the single word "sheep" in the search index.
After the potatoes are peeled and washed they are run through a pipe with blades to slice the potatoes into french fries. These blades are sharpened with lasers and are insanely sharp because they need to cut a lot of potatoes before being changed.
One day they were shutdown and it was time to change the blades. The lady doing the change placed the new blades on the table and bumped the table when she turned to grab a wrench from her toolbox. The new blades started to fall and she instinctively reached out to grab them to prevent them from falling to the floor.
She ended up not grabbing anything because the blades sliced her fingers clean off. They took her to the hospital and due to the blades extreme sharpness, the cut was so clean reattachment was a pretty easy procedure. I don't know if she had any long-term negative effects from the incident.
Safety is important, be aware of your surroundings and don't instinctively grab things you shouldn't be touching in the first place.
Perhaps the only lesson is "slow down."
I inherited a mess of an architecture and am finally getting around to rewriting our deployment process. We buy VM services from a local outfit and the prices are basically an arm and a leg for rather small machines. Due to this my predecessor put in place an insane deployment script. It pulls the new version from github then reloads code on the running dynos, one after another. Reverting is out of the question with our current approach to VCS (something I am also fixing). Most of the time this is no problem, all we are changing really is some template code, or introducing new models and their views.
Thinking back I am quite happy we don't run into more problems than we do, but also happy that this type of insanity is soon in the rearview mirror.
The worst mistake was recently, cost us about 4 hours of downtime during the busiest time of the day.
A big feature on all news sites are lists of stories to present to the user to look at after they have read what you put in front of them at the moment. They may take the form of most viral, most read, most commented, sliced by time or category or many other factors. My predecessor had written all those lists statically, which made maintenance a nightmare and extension very fragile.
I made a function that was a generic list of items. You supply basic parameters, amongst them a QuerySet for what would construct the list and my function would check to see if it was cached and if it wasn't, generate it and cache it.
The framework I use (Django) generally uses lazy evaluation for all QuerySets and I rarely have to think about the size of the list I generate, I just take care to limit the query before I list() it. During development nothing showed up as a problem and I deployed this and all seemed to be good with the world.
A week passes by where I made at least 2 minor deploys (small changes to templates, minor tweaks to list filters) and all seemed to be good with the world.
Designer sends me a pull request, I look over the code, just some garden-variety template changes, nothing that should raise an eyebrow. Make the merge, plan to deploy and then go to lunch. Deployment done, all seems well for 2 minutes but then suddenly servers lit on fire. Pages spewed 404's and 500's like there was no tomorrow.
For 4 hours I tear my hair out, examine every piece of code I was deploying that day, call in the big gun support (the kind that costs more money than I care to think about). Everything I was looking at pointed to the caching agent not working. Too many pageviews requesting the database, too much load on the servers, reboots made them work fine for about a minute but then everything became bogged down.
The big gun support pointed something out finally that I had missed: Traffic from the database to the dynos was abnormally high. Made me take a look at code that had been there for a while and lo and behold: For some reason when you pass a QuerySet as a parameter, it seems to be evaluated for the receiving function! 2 lines of code added, one deploy, problem fixed.
I have no idea to this day how this code could be live for a week without causing problems but an unrelated change triggers the bad behavior. This is not be the first time I've seen strange behavior from code, having seen a Heisenbug in Java code.
There's a happy ending to this. I made a big mea culpa slideshow where I pointed out all the flaws and what we needed to do to prevent a re-occurence. I got support to make the changes needed and my new cluster goes live day after tomorrow. Now I can carefully change NEW dynos for a deployment, keeping the old one's around if the shit hits the fan. I got some changes instituted in how we approach VC, something that's hampered work for a while. And we save money in the long run because we will no longer be paying an arm and a leg for the VMs (AND I got to learn about clustering machines with HA, goodstuff with gravy).
On a Friday? :-)
Client, not happy.
What could possibly go wrong?
i was the one developing the macromedia director app running on the cd.
we were on-time.
we were ready to send them out the door.
it was awesome.
and then we tested the rom outside of our network...
in some far-off corner of code, i had baked in a hard reference to one of our file servers on our network for some streaming assets. the cd failed as soon as you put it in the drive due to that reference to the missing file.
by the time we discovered this, we'd already glass-mastered and stamped 30,000 discs to the tune of $40k or so. or, about $6k per employee. in a company that booked about $50k the previous year. where i worked for free for 9 months.
so, my line of code cost our little company the equivalent of almost all of our previous year's revenue -- not profit, but revenue.
we, of course, had to make the run again -- only this time at the emergency rush prices. and this time, we were running late.
we managed to book some time in the middle of the night at the stamping plant. it was 4am. i had a courier standing over my shoulder watching me run the final build again, this time without the dreaded line of code -- which broke other things i had to fix when i removed it -- before he could take it.
i finished testing. ejected the disc. handed it to the courier, who started running as he was placing it into its case. he drove like hell to make it to the airport where we counter-to-countered it on a 2-hour, 6am flight to vegas for stamping.
oh, and it almost got even worse from there. almost.
we didn't know if they would be able to stuff the cds into the packaging because this was an emergency run and they didn't have the people available.
so...
we were actually on our way to rent a uhaul which we calculated we could drive to vegas just in time for the stamping run to finish. from there, we would load the discs on their spindles, and 4 of us were going to sit in the back of the van, stuffing 30,000 discs while we drove the uhaul to palo alto. from vegas. yes, stuffing discs in the back of a traveling uhaul.
we even had the patio furniture from one of the employees yards already picked out to sit in while we were in the back of the truck.
luckily, the plant managed to squeeze in our packaging (at rush pricing, of course) and all we needed to do was have one of our guys take them as luggage on a later flight that day to the bay area instead.
as to a couple, big lessons learned?
1) i can honestly tell you, i've never, ever had a hard-coded, local network link in anything i've shipped since and never will again. always test off-network. especially these days with mobile apps and their on-off-network states.
2) a strong, non-finger-pointing team is where you need to be. i felt appropriately awful, but we handled it as a team and proceeded to grow that little company to about $40 million a year before a merger.
p.s. oh, and next time, remind me to tell you about the time i ran a database query on production that nuked the entire website for the publicly-traded software company which relied on -- wait for it -- the website to do all its commerce.
Learned: Learning on the job as you hack away on problems is great, but recognize that it's one part enthusiasm and one part risk management. Also learned to never try anything on the command line that wouldn't want to see pulled from my bash history and stuck on the breakroom fridge. Also learned to cope with humiliation well.
[root@importantServer]# reboot
My screwup was at my first "real" job, fresh out of college. I was asked to free up some space from the production server at $BIGCOMPANY, because it was already at 99% capacity (it managed to get to a 100% for a few minutes before I "solved" the problem). The thing is, at this $BIGCOMPANY, for some reason the budget for disk drives was non-existant, and this meant that whenever the disk usage was at or below 95%, we were happy because we still had free space... figure that.
So here I come, armed with the most dangerous tool a newbie can wield... root access and the drive to impress your boss. I said to myself, "I've used root at my home machines plenty of times and nothing bad happened because I've been using Linux for several years by now and I know I need to be careful... so I don't get why everyone says you should never log in as root". Oh boy, how I learned the hard way.
To continue my story, it turns out that the easiest/fastest way to free up some space was to delete the log files for pretty much everything(except the last 5 or 10 logs... because we were "careful", in case we ever needed them). We usually deleted things under certain directories known to hold "useless" logs. So here comes Mr. Newbie-guy-with-the-need-to-shine, and I thought to myself, "why keep deleting the logs from the same directories over and over if that only buys us about 1 or 2 percentage points, instead of cleaning as much logs as possible for the system and freeing up a lot more space?"
After thinking about it for like 10 seconds the most genius thought of my career materializes: do an rm -rf *.log on the topmost level directory of where we used to store everything (webserver, webservices, databases, etc). I happily pressed enter, and a couple of minutes later, hooray! I got the disk usage down to a whooping 90%! I was a hero! that meant we had bought enough time to keep on working without worrying about the disk space for at least another one or one and a half months. This was a clear victory and an testament to my superb sysadmin skills.
Fast forward 4 hours, and the phone starts ringing like crazy as every other employee (non-IT ones) started wondering and then calling us to try to figure out why was their data gone. They did not understand how come they have been working A-OK so far, and then suddenly ALL data from sales team, admin team, the bosses, etc was gone. And then a few minutes later... the whole intranet came down crashing and burning.... then a full stop... nothing was working.
So we went to the logs directory... oops... no logs there!. Ok, let's try to ping the DB. Dead. It's not running and it's responding with an unknown error. When I tried to connect to it would do so, but then some cryptic ORA-xxxxx error came up. No problem says I, I'll just google it and fix it.
Not so fast young grasshopper. That error meant that the DB was out of sync with its own files used for, ironically, data corruption prevention and rollback (or something like that... to this day I still don't fully undersand what those files were used for).
As far as I can remember those logs where a sort of pre-commit place, where all changes would be stored on those files and every X amount of hours the changes would get commited to the actual DB tables. It was some functionality that supposedly was used to correct corrupted entries and to recover (figure that.. ) and rollback data when lost, or something like that. And unfortunately bringing the system back in-sync was way out my league (did I failed to mention that I was by no means a DBA?).
However a struck of good luck came down on me, as the company had a support contract with Oracle and it was the Platinum-covered-diamonds level or something. That meant that after creating a support ticket at like 1AM, I got a call from one of the support guys like less than 20-30 minutes later. This guy seems calm and tells me I should not panick, it was just as easy as doing $crypticOracleStep1, $crypticOracleStep2, $crypticOracleStep3 and voilá! all would be good again. Except for the fact that I had NO IDEA what those steps actually required me to do. Almost in tears I ask the rep to pretty please SPELL every command I needed to execute, letter by letter. I did not want to screw up again.
So there I was, at close to 2AM, with my boss breathing down my neck asking me what every frigging letter of the command I was typing did (which I had no idea...), all the while trying to keep up with this supper friendly guy that was patient enough to spell everything two times.
After a couple of commands later, behold! the DB could be brought up again! oh boy, did I felt relieved. I was jumping up and down because I had fixed my stupid mistake... or so I thought. After almost causing the support guy to go deaf due to my loud cheering, he says "however...". wait... what? there's a "however"?!?. Then he continues saying, "since you deleted the pre-commit file of the last day, the DB is back in-sync... up to yesterday". My jaw dropped to the floor. That meant that the ENTIRE previous day was utterly lost.... sales data, contracts, customer's info, etc.
I thanked the guy for his help, hanged up the phone and turned to my boss telling him that I was ready to turn in my resignation letter just after helping capture what availabe data was actually there (in papers, by calling customers and asking them again, etc).
My boss then turns to me and says, don't worry. We've all been through this at least once in our careers. Even I made a mistake that is terribly similar... however when I brought down the database, it took us one full week instead of one day... and rest assured that as I learned my lesson, you did as well. And I need guys like you, that have the initiative to solve things... and the ability to learn from mistakes. So don't worry, you are not losing your job. However you can't go home until you help everyone get as much data back as you can.
Aw shoot... well.. I guess it could've been worse. So after having lunch with my boss and the other teammates at like 6-7 AM, I went to the sales dept and started asking around how could I help them get their data back.
Those were the longest 38 continous work hours I've ever had to resist. I did not go back home until more than a full day and a half later. I was tired as hell to say the least... but to this day I think it was a blessing that I got to learn such a hard lesson but being backed up by a boss that was very cool and progressive about it.
Lessons learned:
0) Never ever ever ever use root, especially for deleting files and ESPECIALLY with the -f flag.
1) Do not assume that something you know will hold. Confirm it in the particular system you are going to be working with. (i.e. do not assume .log files are always log files because in your laptop that holds true)
2) Be ready and willing to assume the consequences of your actions. Most of the time if you assume responsability for your mistakes, people will forgive you and even give you a piece of advice.
3) Never ever ever ever use root.
I run around like a headless chicken trying to find who knows the right backup to use and so forth, and I can't figure out why everyone is so calm and collected about it. Production was down/shit, I hope I still have a job. Turns out we had no active clients at the time - no-one was accessing the site. We'd finished one run and were in 'dead time' before the next. My next project involved implementing coloured prompts and I no longer leave production ssh sessions lying around when I've finished with them.
My CTO still has me listed as "database [vacri]" in his phone...
It failed anyway, but I wasn't around when it did and there would have been no "I told you so" credit even if I were.
One of those "big company" lessons, but probably applicable to startups (which have an even higher ego density).
Some companies are using internal prediction markets, where employees can speculate on various initiatives:
http://en.wikipedia.org/wiki/Prediction_market#Use_by_corpor...
Open allocation seems like the ultimate prediction market - people deciding on what initiatives to work on and invest their time in is a stronger signal than dysfunctional internal politics. And people are likely to be more motivated working on projects they think have value.
Raw unadulterated fear followed by panic.
A full reinstall.
Triple checked dd params ever since.
Those that did down-vote you are responding to the "Ruby-on-Fails" part of the sentence ... you were the failure, as you would have been with JavaEE, Django or Drupal. Your comment was needlessly offensive (to some at least) - if you're going to be critical, you should learn enough to contribute back too.