Bit of a rambly story, but I swear it is relevant.

So previously I worked as a consultant for a company that manufactured a relatively small number of high value (tens of thousands of dollars each) Gizmos in a lightly regulated industry - the requirements weren’t too crazy, basically that everything has a serial number and they can prove that any given serial passed the full range of tests before it left the factory. Pretty much the sort of thing you’d want to have if you gave a crap about quality products anyway.

Initially they were using Excel to keep track of this - they manufactured 10 units a week, it worked well enough. Eventually, they got more successful and needed to scale up to 50 units a week, and it was decided that they needed A System to keep track of testing and manufacturing. Their head of manufacturing “looked around and couldn’t find anything off the shelf that was suitable” (ie, cost $0, and perfectly matched his aesthetic tastes; mistake #1), so they decided to build their own system.

They had a few in house developers, but they were focused on building new features (things that drive sales, unlike maintaining their reputation for delivering reliable products), so head of manufacturing decided to get one of the production line techs (who was “good with computers” by virtue of having built the Excel system, but was not a software developer mistake #2) to do it.

Eventually, they decided to use Microsoft PowerApps to build the new system - for those with the good fortune never to have seen PowerApps, it’s essentially a “no code required” drag and drop UI tool that you script using Excel formulas. Think Visual BASIC or Scratch, but Cloudy.

On the surface this made sense - the developer was proficient in Excel, so use what you know. Unfortunately, PowerApps is designed to rapidly build throwaway UIs over simple data models and lacks some of the things that actual software developers would have thought to ask about:

• It lacks real version control - you can “undo” a deploy, but there is no way to discover what changed between versions, or do branches, or code review
• Because you can’t effectively manage changes to the system, you can’t do pre-production releases
• Its native database system doesn’t do referential integrity
• There is no straightforward way to do any kind of locking - and because there is no referential integrity, it’s really easy for concurrent users to really mess up the data
• There is no way to do automated testing
• The development group could have actually documented how stuff worked, requirements, specs etc but didn’t, so any time there was any issue you had to play the game of “is this a bug or bad design?”

Eventually, these chickens came home to roost in the form of a defect that slipped through testing that they then couldn’t isolate to a particular batch because none of their testing data could be trusted. I was brought in to try and unpick this mess and advise on a replacement system, but between the cost to fix the issue and the lost sales from it they ended up in a pretty bad spot financially and ended up being acquired by an investment group.

Anyway, the takeaway from this is that you disregard experience and judgement at your own peril, up front savings generally don’t manifest in the long term and I expect there is going to be a thriving market of consultants brought in to point and laugh at companies that decided that a bunch of cheap, inexperienced developers and a magic talking parrot would build better software than cheap, inexperienced developers being guided and upskilled by an experienced senior developer

Isn’t this an admission that the US justice system is fundamentally unjust? Wouldn’t it be better if the judicial branch was functional and could update old unjust laws so there were zero people who needed their sentences commuting?

I was in the same place as you a few years ago - I liked swarm, and was a bit intimidated by kubernetes - so I’d encourage you to take a stab at kubernetes. Everything you like about swam kubernetes does better, and tools like k3s make it super simple to get set up. There _is& a learning curve, but I’d say it’s worth it. Swarm is more or less a dead end tech at this point, and there are a lot more resources about kubernetes out there.

thats_a_bold_move.gif

Trying to extort the federal government like that seems like a really quick way to end up with your face, phone number and home address in a press release, along with a note from the NSA that basically says “this guy has $33 million in Bitcoin, would be a shame if someone kicked in his door and beat him with a bat until he gave up the keys :)”

Especially when “tmpfiles” is an existing term of art with a very specific meaning

The early twenties intermediate dev on my team was explaining the other week that if you remember a time before smartphones and broadband, you are old

I personally am familiar with 2 organisations with millions of dollars in annual revenue that deploy critical line of business applications like this in 2024

They are, but I think the question was more “does the increased speed of an SSD make a practical difference in user experience for immich specifically”

I suspect that the biggest difference would be running the Postgres DB on an SSD where the fast random access is going to make queries significantly faster (unless you have enough ram that Postgres can keep the entire DB in memory where it makes less of a difference).

Putting the actual image storage on SSD might improve latency slightly, but your hard drive is probably already faster than your internet connection so unless you’ve got lots of concurrent users or other things accessing the hard drive a bunch it’ll probably be fast enough.

These are all Reckons without data to back it up, so maybe do some testing

Neurosyphilis

Weren’t the shit balloons retaliation for the south sending balloons to drop propaganda leaflets?

My 10 year prediction - Microsoft does a full transition to a services company:

• Basic Windows is free, even for OEMs
• Windows Professional becomes a subscription thing, maybe you get it as part of your Azure AD sub
• Things like Recall or not having ads are extra subscriptions

Pretty much - I try and time it so the dumps happen ~an hour before restic runs, but it’s not super critical

pg_dumpall on a schedule, then restic to backup the dumps. I’m running Zalando Postgres in kubernetes so scheduled tasks and intercontainer networking is a bit simpler, but should be able to run a sidecar container in your compose file

If you thought the bots were obnoxious now…

This is where I’d put my Framework laptop

IF THEYD SELL ME ONE

If you figure it out, I know several companies that would be more than willing to drop 7 figures a year to license the tech from you

Well well well, if it isn’t the consequences of my own actions

Dude wouldn’t know what to do with a condom.

My theory is it’s not dementia, it’s neurosyphilis. Party like it’s 1825!

git-annex maybe?

Yeah, they are mostly designed for classification and inference tasks; given a piece of input data, decide which of these categories it belongs to - the sort of things you are going to want to do in near real time, where it isn’t really practical to ship off to a data centre somewhere for processing.