Show HN: Dead man's switch without reliance on your infra

[adamdecaf]

I wrote this Go project to implement a dead man’s switch that didn’t rely on cron jobs, timers, databases, etc on the infra it runs on. Deadcheck integrates to PagerDuty and keeps a long running incident snoozed until expected check-in times where it’ll alert unless a check-in occurs.

[remram]

If you call this a "dead man's switch" I'd expect it to turn my app off when I die. E.g. "switch" something when something happens to a "man".

Your own README links to this definition: "A dead man's switch is a switch that is designed to be activated or deactivated if the human operator becomes incapacitated".

This is a watchdog timer / monitor / heartbeat, setting off an alert if a timer elapses.

[aftbit]

The "original" dead man switch as I heard about it was a pedal on a train that would apply the brakes if the operator released it.

I've often wondered about how to reliably take software actions after my death or dishonor. After all, you can't really rely on me being able to pay my bills. I'm not looking to do something expensive, more like delete my accounts and send some messages.

[xnorswap]

> I've often wondered about how to reliably take software actions after my death

This is actually fairly simple and well understood: leave instructions in your will.

"Notify <Provider> to delete my account" is a perfectly valid instruction to leave for an executor.

You could leave behind a password cache with a master password left in your will, but I suspect much of this still runs on trust. I'd imagine (I haven't tried), that "X has died, please take action Y" is a fairly reliable social engineering vector if you have a convincing "proof" that X has died.

It's worth noting that the executor isn't hard forced to carry out your wishes, the legal recourse for them not doing so comes from other beneficiaries ability to take legal action against the executor. If those other beneficiaries don't care much for enforcement, then you might prefer technical methods such as the submission.

[ryandrake]

I keep a "death README" with all of my online and offline account credentials and phone unlock codes, PII that might be needed to authenticate w/ various companies' services, copies of wills, trusts, powers of attorney, health care proxies and so on, copies of all vital docs like marriage certificate, birth certificates, home router SSIDs and passwords, information about doctors, health insurance, life insurance, all financial accounts and brokerages, approximate balances, bills and how to pay them, tax returns and how to file them, a list of recurring expenses and how to pay them, property w/ approximate values, and so on. A hardcopy is kept in our house where next of kin can find it if needed without knowing a safe combination, but casual robbers wouldn't stumble across.

[fragmede]

Why without the safe combination? Not to out myself but mines in a safe. If I'm dead I expect my next of kin to crow bar it open but if I'm not, I'd rather not have anybody else access that.

[echoangle]

The „X has died, please take action Y“ thing also only works if the service reliably knows that the account belongs to X. My executor can’t delete my HN account because he can’t prove it’s actually my account (without getting the password).

[ssl-3]

Can it not be cryptographically proven?

Leave a public key in your HN bio.

And leave a matching private key and validation instructions in your will.

If the keys match along with a death certificate, then: The account owner is validated as being dead.

[PeterisP]

It's not about it being impossible to cryptographically prove/validate it, but rather about services choosing to not attempt to try to validate it. They generally don't provide such an option, because it's tricky, somewhat manual, has certain costs and risks, and no benefit to the service provider.

If some law prescribes that after following a certain verification process, the operator is required to delete the account, then that legally mandated process would work, but in the absence of such a law literally no process can be sufficient, because the operators can and will choose to ignore it, no matter how reliable it is.

[echoangle]

If I prepare for it, sure. At that point, I can just leave my password though. I was responding to the point that you don’t need to leave the password because a death certificate would be enough.

[edm0nd]

Would be cool to have some kind of "deadmans infra AI or bot" that would auto fund your server bills for X amount of additional months/years and then send out emails and post a notice you have died and your service EOL is estimated to be X or Y.

I also suppose you would have to also roll in some kind of automated patching and etc into it which would be rather difficult and break a lot of thing if went bad but some kind of "self healing" bot could perhaps also look after this part to fix anything should it break.

Also kinda opens up an entirely new attack vector. Threat actors could scan for these notices and go "hey this person is dead. lets hack their stuffs".

[LorenPechtel]

In the US we already have a system that could be used and would be hard to spoof but I am not aware of anyone implementing it. It has a delay of months built into it.

Specifically, the Social Security Death Index. Some months after you die your Social Security number gets listed. It's meant to prevent impersonating a dead person but could be used to shut down accounts etc.

[Muromec]

> After all, you can't really rely on me being able to pay my bills.

Bit that's exact trigger you wat.

Make something that keeps running while you pay bills and stops running after you no longer pay them. Pay those bills from your current account.

Make another something that periodically checks the status of the first system to be operational. After sufficiently long periode of failures activatie the cleanup crew.

Pay for the second system from a savings account, trust, llc or some other way that is not deactivated once you die

[DowagerDave]

>> after my death or dishonor

honest question: why do you care?

[fragmede]

I cared about these people when I was alive, why would that stop because I'm dead? If a meteor struck me dead in my bed, I'd want the survivors to be able to use my stuff to make their lives better.

I had a friend pass and his fiance was in need of money but he didn't leave the password to access the crypto, so it's lost forever. If I die unexpectedly, I'd at least want it to not be lost forever.

[vincnetas]

if you care about people that you leave behind...

[nurettin]

From experience, it is a huge inconvenience to people surviving the deceased leaving without any way to pay their debits. And if you don't care about what happens after you die, why did you even care when alive? Why not always be a dick? At least it is consistent.

[hotspot_one]

Under US law, the debts die with the person. You are under no obligation to repay your parent's debts. Now if the debt is tied to a house (mortgage) or a car (car loan), you might lose the house/car if you don't pay, but you do not have an obligation to pay. Likewise failure to pay will not impact your credit.

So if I die in debt up to my eyeballs, and if I am sole signatory on those debts, I have only hurt my creditors, not my family.

caveats-- if my family was counting on the house and I have an unaffordable mortgage, then yes I have caused them harm. Likewise other irresponsible debts.

-- at the end of the chain, creditors are also people. It is their job to loan money at risk, so their loss is their problem, but this assumes I was dealing in good faith when I took the loan.

[groby_b]

The debts very much don't die with the person - the estate is on the hook to pay your debts before distributing to heirs.

Obviously, with some "it depends" nuance - but if the difference between this and your world view would make a significant difference to your loved ones, you might want to talk to an attorney.

[LorenPechtel]

How to pay??

Two experiences with this. Companies don't particularly care who pays the bill. They send a bill, they get money, they don't care if the person who paid is the person. In fact, in my recent brush with such matters the company specifically knew the person was dead when they accepted the payment.

What's hard is getting access to their money to pay debts with.

[adamdecaf]

Yea the idea is that if your service doesn’t check-in then a preconfigured alert triggers.

[greiskul]

Google has an option for what to do with your account if you are inactive for a set period of time. So you can choose what to delete, and what to give access to someone you want. You can also have it send emails to up to 10 people, with whatever message you want.

[pests]

Facebook has "Legacy Contacts" which is a similar setup to give your account to someone after you pass.

I did the math years ago and even then FB has a large percentage of it's user base dying daily.

I think there were one of the first to properly allow an account to be "memorialized" and moderated by loved ones.

[IIAOPSW]

As a point of interest, trains still have these. Though it usually takes the form of a somewhat sporadic alarm and an acknowledgement buzzer. Reaction time to the buzzer somewhat influences the interval of the next one.

[BrandoElFollito]

What is "dishonor" in that context? (Sorry, not a native speaker of English)

[eep_social]

As a native speaker, I don’t think the phrasing is idiomatic but I read it to mean imprisoned or otherwise out of society without actually being dead.

[aftbit]

Yes, you got it 100%. I agree, it is not idiomatic, but was intended to call back the "death before dishonor" trope that your sibling commenter mentioned. I intended "dishonor" to cover a number of cases short of my death where I might be unable to continue to care for my obligations for a long period of time. For example, imprisoned, deplatformed, critically injured in a coma, lost in a serious mental health crisis, etc. In some ways, handling that is a harder challenge than handling death, as there are fewer well-worn paths to follow.

[eep_social]

I thought it was a good linguistic innovation.

[LorenPechtel]

(Native speaker) Wrong, but the intent was clearly that they were out of it for whatever reason. And since the true purpose of language is to communicate and it does that is it truly wrong?

[meindnoch]

It doesn't mean anything, because it is wrong. The correct idiom is "death before dishonor", which means that one would choose death instead of doing something disgraceful/shameful.

[superb_dev]

That doesn’t make sense in the context

[DoctorOetker]

it could also be read as in case a (bi- or uni-lateral) agreement is dishonored.

[asdfman123]

I thought this was going to be some way to exact retribution on your employer after being laid off

[hinkley]

That is the typical scenario in the context of software.

Not that it happens often, but that the incidents are so memorable.

[John_Cena]

Naw, in the modern world only people like Bench Simmons get guaranteed contracts. :|

[duggan]

Routinely surprised that there are adult human beings that do this, but so it goes.

[asdfman123]

I would never do it myself. Move on with your life!

I did know a guy who worked at a major corporation explain his dead man switch like it was the most normal thing in the world, though. Extremely cursed.

[hinkley]

I could see people in a somewhat toxic situation do things like this but it’s better to get out.

Especially if there’s some complex onerous task that needs to happen and you get no credit for doing it well. Leave, and let them find out what it’s like being one of the other characters in It’s a Wonderful Life in the worst timeline.

[asdfman123]

No points are awarded for contributing to a toxic situation

[vorpalhex]

Wire in pagerduty to whatever your action is. Ideally distributed on multiple infra pieces. This is just the switch side not the action side.

Normally deadmans switches can be compromised by disrupting the deadman switch hardware. This removes that attack vector and pushes it further up the chain (which may or may not help you).

It's certainly very clever.

[adamdecaf]

Thanks. We already rely on PD so preconfiguring the alert/snooze isn’t additional risk for us.

[Terretta]

> This is a watchdog timer / monitor / heartbeat, setting off an alert if a timer elapses.

Notice from the config file, this doesn't actually do checks, only has passive listeners with a remote switch to toggle. That makes this a dead man's switch, not a watchdog / angel / monitor / heartbeat check.

This doesn't do anything except sit there watching a clock not knowing if you're dead.

The service you want monitored, the service you want to be sure the heart is beating, has to ping this to demonstrate aliveness.

If this doesn't get a ping in time, it then acts by telling pagerduty to alert.

If your service (man) is dead (or your service otherwise can't reach this to check in in time), this switches PD to alert mode.

So, dead man's switch does seem a fair characterization.

[rozenmd]

The term of art you're looking for is: "heartbeat check"/"healthcheck", or most commonly: "cron job monitor".

[bravetraveler]

Seems very similar to a 'watchdog', just... reporting instead of doing anything about it.

[adamdecaf]

There’s no crontab for banking days. Deadcheck also requires checkins within a delta of the expected checking time.

Deadcheck also doesn’t rely on your infra to alert.

[herpderperator]

Yeah I thought "dead man's switch" was such a weird name for this.

[LorenPechtel]

It's something that happens if you fail to take action to prevent it at some specified frequency. Looks a lot like a dead man switch to me.

[vivzkestrel]

uptime monitor?

[rozenmd]

Uptime monitors ping your server, this seems to do the opposite.

[cduzz]

I guess I'm an oldster shaking my fist at clouds, but ...

Depending on where you live, and if it matters, give it to an attorney.

Maybe use this to trigger the notification to your attorney, with instructions on how to double check things before doing things.

[danbruc]

I agree, I think people underestimate the importance of actually checking why the heart beat was missed, double checking is important. You might think you can easily click some button once a week and it is no problem to take some drastic and maybe irreversible actions after a week without any event. And then you get into a serious accident and end up in hospital for weeks. Or there is some bug unintentionally triggering the actions.

[cuu508]

The configuration example in README has "Reports Finalized", so I assume this is more for IT infra monitoring, than for taking irreversible actions when you become incapacitated :-)

[9dev]

I’m not sure I understand what the application itself does; as far as I can see, it basically configures the external services to carry out the actual checks and ensures this configuration is up to date?

Wouldn’t it be better to do this as part of a Terraform script or something? Not to take away from the project, it sure looks neat—just wondering if I really want to deploy yet another, independent IaC tool.

[adamdecaf]

It sets up expected check in times for an app/job. The job is responsible for checking in, but the alert will fire regardless of your infra.

Doesn’t require a start command and your infra can completely fail and you’ll still get alerted.

[hiatus]

There is a pagerduty provider for terraform, too. https://registry.terraform.io/providers/PagerDuty/pagerduty/...

[pigeonhole123]

It doesn't look like PagerDuty supports sending out alerts on missing heart beats

[bryanlarsen]

AFAICT, that's a better description of the project for people like us. It's a way to add a missing heart beat alert to PagerDuty.

[adamdecaf]

The idea is an app checks in after scheduled jobs, which extends the snooze

[hiatus]

That's exactly how escalation in pagerduty works.

[focusedone]

I feel like posting a software project to HN and meeting criticism is some near-end-stage of programming mastery.

[arisAlexis]

Actually I got fried and very bad remarks on my project. Not true that hn crowd is friendly to show hn projects at all.

[divbzero]

Yes, it’s a positive signal when people take it seriously enough to criticize.

[wcallahan]

Was thinking the same!

[herpderperator]

So... like BetterStack's heartbeat monitor? [0]

[0] https://betterstack.com/docs/uptime/cron-and-heartbeat-monit...

[neilv]

Better Stack got on my disapproval list, for spamming me with a series of automated sales emails 15-20 minutes apart, which seemed designed that way. "One more thing.", etc.

They should've sent one welcome&upsell email, not several. And not abused contact info that was intended to reach me reliably and promptly.

For my immediate very simple need, Uptrends didn't spam me, and so far they've reliably and promptly notified me whenever my site is down even briefly (e.g., rebooting for kernel update), so they'll be getting money from me as I grow. https://www.uptrends.com/

[herpderperator]

Sorry to hear. I didn't have that experinece; it's been pleasant for me (there was a billing snafu while they changed their plan structure but it was resolved with a support email) and their monitoring has been very reliable since I signed up a year ago. Their service also seems to be better (4 checks/locations vs Uptrend's 2.)

Anyway, tl;dr is I never had an issue that resulted in not getting alerted about something that I needed to be alerted about, which, above all else, is the point of their offering.

[tnolet]

or Checkly's heartbeat check

https://www.checklyhq.com/docs/heartbeat-checks/

[adamdecaf]

I don't see how to configure anything but an interval (e.g. every N minutes) in checklyhq. Deadcheck allows you to expect check-ins at arbitrary times (e.g. 9am, 1:15pm, 3:15pm on weekdays).

https://developers.checklyhq.com/reference/postv1checksheart...

[siliconc0w]

A possible integration are messaging apps that support send later. I sometimes set these up if I'm going hiking.

[adamdecaf]

Yea, and you would delete the pending message. Good idea…

[christina97]

I sometimes think about a “dead man service”: you leave instructions (that you upload to the service), then when you pass away/etc, the service operators go and follow your instructions and do whatever you asked for. You’d pay some pre-agreed sum, possibly annuity-type subscription, and at the end we go and follow your wishes. Basically a technically competent will executor. It’s probably too much to expect your family to know how to operate your systems. Maybe you encrypt the instructions and give fractional keys to family etc.

[lancesells]

This is an attorney. No subscription, no tech, no bugs, no encryption, less hackable.

[mhink]

> Basically a technically competent will executor.

I don't think OP is saying that an attorney *can't* get these things done, but that it would make them feel more comfortable knowing that a technically competent person and/or service will be performing the actual actions.

I do think there's a place for an attorney here, in the sense that they could be the trusted individual responsible for notifying DeadManService, Inc. that a particular person has, indeed, passed on and wishes DeadManService to run their instructions.

[ufmace]

I'm not sure what value adding a DeadManService would have then. Probably simpler to just have the attorney's instructions say, "Hire a technical consultant to carry out the following:".

[INTPenis]

I've thought about this too but I think the type of business best suited to execute this is a law firm, or you know, the normal executor of people's wills. They just need the IT bit.

[INTPenis]

I did something similar years ago when I was working with observability. Prometheus alertmanager triggers a special alert constantly that calls a lambda (or any webhook), so when alertmanager dies or is unable to alert then the lambda will send an alert over a 3rd party push service to notify ops that alertmanager is down.

We called it a dead man's switch but it was really just a way to monitor alertmanager.

[adamdecaf]

Yea I’ve setup two alertmanagers that check each other before. It’s useful for multi site deployments.

[EGreg]

Speaking of not using anyone’s specific infra, we deployed software on blockchains to do this.

We had to implement a “heartbeat” and “succession” in our blockchain-based solutions for organizations to control things together.

It’s part of our “application suite for organizations” where each one is sort of this general-purpose LEGO block that could be used to build a custom solution.

In each case, you’d enter some parameters and create an instance from a Factory. We made it simple and secure for any organization to use this.

In this specific one, ControlFactory is used to create ControlContract instance that can be used to control an address together (to sign off things collectively, like transfer tokens or call an arbitrary method on a different address)

We had to handle what happens if the M of N people don’t show up for a while. And we said they have to call a heartbeat() method every so often. If they fail to call it then control temporarily passes to the next group in succession, until the OGs can finally call the heartbeat() method again.

Here is more info on the why: https://community.intercoin.app/t/intercoin-applications-con...

And here are the rest of the blockchain apps for organizations: https://community.intercoin.app/t/applications-of-intercoin-...

You can go ahead and use it, the factory it’s been deployed on many EVM blockchains, at the same address.

PS: fun fact, you can configure a ControlContract to also manage calling methods on itself, thereby creating custom “policies” for organizations when it comes to granting/recoving rights of other people to the quorum.

[turtlebits]

I'm not sure running an additional service plus a SaaS requirement is better than relying on infra. Standing up this service is going to rely on your infra. What if it goes down?

If you're going to have to write a scheduled check-in anyways, why not use something like cronitor and reduce the complexity? (they host the check-in endpoint)

[psnehanshu]

If the infra goes down, you fix it while alive, otherwise the snooze timeout on the PagerDuty incident will expire and that will trigger the configured tasks. If PagerDuty goes out of business before that, then that's a different discussion.

[turtlebits]

My point was - if you're going to rely on a SaaS (Pagerduty), why not just use one that includes health check monitoring (ie cronitor) and cut out this self hosted webhook service?

[adamdecaf]

Deadcheck handles calculating the snooze durations for you. It could be a library as well.

[adamdecaf]

Exactly. At Moov we rely on PD, so if they’re down we have bigger issues anyway. I plan to support additional integrations so a check-in could update multiple

[wil421]

Any support for something besides pager duty? I work for a telco and there are so many different heartbeats we rely, this project seems very useful for something like that. The config file looks much better than the python/perl/bash whatever scripts we have scheduled on our systems.

[adamdecaf]

Yea I want to support other vendors. What do you use?

[arisAlexis]

The problem with these kind of project is that they are no good for very serious situations. Say Navalny set it up. They kidnap someone, the make you delete the pagerduty service and then kill you. Immutable / undeletable tasks would be needed.

[roshan8]

Nice one! Do you have any plans to support something else apart from Pager duty. I have been using Squadcast.com for my hobby project monitoring. Integration with that would be nice. I'm open to create a PR if you are interested.

[adamdecaf]

Thanks, yea I’d like to support other vendors. Haven’t heard of Squadcast

I have https://github.com/adamdecaf/deadcheck/issues/12 tracking additional vendors

Edit: I'm open to a PR if you're willing. Curious, but would you use their delayed notification config to implement?

https://apidocs.squadcast.com/?version=latest#7742a9af-29fe-...

[v3ss0n]

Better have something that runs off the cloud. PagerDuty is not something under your countrol. A stuff as imporant as dead man switch should not rely on someone's service.

[PeterisP]

I'd argue entirely opposite, that a dead man's switch should 100% rely only on someone else's services and should avoid any dependencies on your stuff, as something happening to you is correlated to something happening to things you run or own, and the whole point of such a switch is that it should function properly when e.g. all your stuff burns down or gets confiscated or whatever.

[tatersolid]

Okay, so now your dead man’s switch depends on your home or colo’s power, internet connection, compute hardware, storage hardware, network hardware, software stack maintenance, plus the ability to pay bills for those on time?

Pre-Paying a cloud service pennies per month to do all that for you will likely be more reliable and much simpler.

[encoderer]

Cool trick using PagerDuty like that.

If you don’t want to use a service like Cronitor, you can self-host this without the usual fear that an outage will also take down your monitoring.

[adamdecaf]

Yep. I looked at Cronitor and thought about using it, but direct to PD removed a step for us at Moov.

[grahamj]

Neat. We use OpsGenie to monitor our alerting infra and it has deadman/heartbeat support so something like that is another way to go.

[adamdecaf]

How are they triggering the alert when the dead man’s switch isn’t tripped in time?

[grahamj]

It's basically an alerting platform so it's just what they do. It integrates with our alerting for things like metrics monitoring and it'll route the notifications based on on-call schedule etc. Deadman is just another way to trigger.

[lopkeny12ko]

[flagged]

[hooverd]

Your infra is relying on an ISP and colo and the power company and maybe even the Post Office. You're always relying on something that's not you. Personally I'd rely on pre-paid AWS. If all of us-east-1 goes down for days then we probably have bigger problems.

[hypeatei]

That's what I thought; this tool is "kicking the can" to whatever third party service you're using. So, it's essentially just a tool that will ping stuff for you in a slightly more convienent way?

[adamdecaf]

Systems always rely on something. PagerDuty has been very reliable for years and we use it for alerting, so relying on it more isn’t a big ask. I plan to support multiple integrations so you could get alerted from multiple streams.

[zeroq]

It's a cloudification of dead man's switch.

[ellyagg]

Now two isolated systems have to fail. Redundancy is fault tolerant. Nothing is fault proof.

[joshbetz]

Very clever. I have Alertmanager in a second region so it can check on my first Alertmanager, but this is much nicer.

[jrockway]

I ended up going this route: https://github.com/jrockway/alertmanager-status

[adamdecaf]

Yea I’ve setup multiple Alertmanagers that all check each other before. That setup is useful to detecting route failures between sites.

[hiatus]

Do the people on the receiving end of a pagerduty notification require pagerduty accounts?

[adamdecaf]

No, you can get text, email, slack, etc alerts from PagerDuty without installing their app.

[ForHackernews]

I don't know about the rest of the people here, but I aim to outlive PagerDuty.

[xyst]

I’ll use this for my next wormable vuln instead of hard coding DNS checks ;)

[jbverschoor]

Not to be confused with deadmanssnitch.com