https://blog.cloudflare.com/durable-objects-alarms/ > Under the hood, Alarms are implemented by mak

https://blog.cloudflare.com/durable-objects-alarms/

Under the hood, Alarms are implemented by making reads and writes to the storage layer. This means Alarm
get
get
and
set
set
operations follow the same rules as any other storage operation

The DO alarm's blog post references the storage blog post:

https://blog.cloudflare.com/durable-objects-easy-fast-correct-choose-three/#part-2-output-gates

From what I understand of this blog post, the following would be how things would play out? Which doesn't seem to align with <any failure within the alarm handler causes all alarm calls in that call to be ignored and for the original to be retried>

alarm() {
  await setAlarm() // This would be committed (and therefore overwrite the initial alarm? what happens to retries?)
  await someOtherStorageOperation() // this too given no failure up until the first fetch
  await fetch(...external) // Output - gates mean we now wait for previous operations to succeed

  await setAnotherAlarm() // This would not be committed because of subsequent failure (write coalescing)
  await triggerSomeFailure()
}

alarm() {
  await setAlarm() // This would be committed (and therefore overwrite the initial alarm? what happens to retries?)
  await someOtherStorageOperation() // this too given no failure up until the first fetch
  await fetch(...external) // Output - gates mean we now wait for previous operations to succeed

  await setAnotherAlarm() // This would not be committed because of subsequent failure (write coalescing)
  await triggerSomeFailure()
}

alarm() {
  await setAlarm() // This would be committed (and therefore overwrite the initial alarm? what happens to retries?)
  await someOtherStorageOperation() // this too given no failure up until the first fetch
  await fetch(...external) // Output - gates mean we now wait for previous operations to succeed

  await setAnotherAlarm() // This would not be committed because of subsequent failure (write coalescing)
  await triggerSomeFailure()
}

alarm() {
  await setAlarm() // This would be committed (and therefore overwrite the initial alarm? what happens to retries?)
  await someOtherStorageOperation() // this too given no failure up until the first fetch
  await fetch(...external) // Output - gates mean we now wait for previous operations to succeed

  await setAnotherAlarm() // This would not be committed because of subsequent failure (write coalescing)
  await triggerSomeFailure()
}

Aalgads OK, let me make sure I understand this. If I set a DO to invoke an alarm, the a...

peachneoOP•2/1/25, 9:18 AM

It seems this is an ongoing point of confusion - would be great for one of the CF devs to provide some insight here

Ppeachneo https://blog.cloudflare.com/durable-objects-alarms/ > Under the hood, Alarms ar...

lambrospetrou•2/1/25, 10:14 AM

First of all await setAnotherAlarm()await setAnotherAlarm() does not set a new alarm, it overrides the alarm you set in the first line.
There is only 1 alarm set at a time.

Regarding the behavior of setting an alarm and throwing right after, I will check with the team and get back to you.

Ggeg Observed with single and multiple clients receiving mesages via websocket I sups...

Milan•2/1/25, 5:40 PM

Definitely drop an update here if it turns out to be a client issue! If ws hibernation is broken we'd want to fix it for y'all asap. FWIW I haven't seen other reports of this and we've tested this in the past, but not continuously testing this so it's certainly possible something has broken since we last tried it. I can't think of any changes to that code in quite some time though...

ps2000•2/1/25, 8:33 PM

Hello! I'm new to CF. Working on a DO with WSs. I'm wondering if it is possible to reassign a socket to a different DO after an ws.accept()? The reason: I'd like to "shard" the ws connections but I won't know the shard key until the first ws message. (first server message is a "connect" message with an auth key which the shard key is derived from).

And wondering how many ws connections can. be maintained per DO given the 128MB memory limit? I can estimate my own per ws state (eg: 3KB).

11984 Ford Laser You will need to do a `state.storage.deleteAll()` to remove an old instance. An ...

Eric•2/2/25, 6:08 AM

state.storage.deleteAll()state.storage.deleteAll() happens per-instance. Each user I have has an access to his/her own instance, but for some users, I no longer have access to user ID (forgot to track them...). Aren't they still going to incur a storage charge?

EEric `state.storage.deleteAll()` happens per-instance. Each user I have has an access...

森

森@VerseLink (RW)•2/2/25, 7:32 AM

yes, they will incur storage charge

森

森@VerseLink (RW)•2/2/25, 7:32 AM

if you forgot to track them, you can technically find a list of DOs here: https://developers.cloudflare.com/api/resources/durable_objects/subresources/namespaces/methods/list/

Cloudflare API | Durable Objects › Namespaces › List Namespaces

Interact with Cloudflare's products and services via the Cloudflare API

森

森@VerseLink (RW)•2/2/25, 7:33 AM

but it's using the cloudflare API and has stricter rate limitations

MMilan Definitely drop an update here if it turns out to be a client issue! If ws hiber...

geg•2/2/25, 10:03 AM

yep nvm, it's because I didn't send protocol level pings..

Ggeg yep nvm, it's because I didn't send protocol level pings.. 🤡

quan•2/3/25, 2:47 AM

hi i may have same issue, I was trying the default worker-chat-demo from github, is it just a blank msg send from client that you did to get over that?

Ggeg yep nvm, it's because I didn't send protocol level pings.. 🤡

1984 Ford Laser•2/3/25, 3:17 AM

not necessarily a question for you directly, but as someone who doesn't know websockets well, is there a suggested interval where we should be sending/acking pings to optimally keep clients connected?

DiamondDragon•2/3/25, 3:24 AM

I am playing around with drizzle's durable object integration. Got it working. How are people inspecting their DO storage? Is there any chance an integration with drizzle studio might be coming soon?

DiamondDragon•2/3/25, 3:27 AM

I'd also be curious to learn more about how people are building apps where you have a central database (say postgres) and then you utilize DOs. Are people persiting the DO namespace/ID in their main app DB and then splitting off other types of data to keep in the DO?

Like if you were to do any analytical querying, this data is scattered around all these DO's. Im just curious if there's been any writing anyone has done on these new ways of thinking when finding areas to implement a DO in an app, maybe progressively

DDiamondDragon I'd also be curious to learn more about how people are building apps where you h...

1984 Ford Laser•2/3/25, 5:57 AM

currently I'm just aggregating the different data from different DOs within a process then using it as necessary. If aggregated data doesn't change often, and gets accessed regularly, then it makes sense to save the aggregated data and disperse as necessary. That aggregated data could in theory be its own DO which handles itself as necessary. Service Bindings and RPC can make things easier to handle depending on how you need to handle the data

geg•2/3/25, 6:42 AM

@quan @1984 Ford Laser It was definitely because I didn't send pings when not sending any messages, It's a mistake on my part

I was so used to browser environment where that part is abstracted that I just forgot about it completely. I haven't tested how long the ping interval can be yet, would be nice if documented

elunesarrow•2/3/25, 9:31 AM

Hi, I need some clarifications regarding DO billing:

[1] "Each Durable Object stays alive as long as requests are being processed, and remains alive for several seconds after being idle before hibernating"
[2] "Once your Object finishes responding to all requests, it will stop incurring duration charges"
So assume if for my use-case, I only use RPC to query/migrate sqlite, does this mean I will get duration charges only for the RPC session, which is probably a few milliseconds, or also for the whole "alive duration" which includes "several seconds after being idle before hibernating"?

[1] https://developers.cloudflare.com/durable-objects/what-are-durable-objects/#durable-objects-highlights
[2] https://developers.cloudflare.com/durable-objects/platform/pricing

Eelunesarrow Hi, I need some clarifications regarding DO billing: - [1] "Each Durable Object ...

1984 Ford Laser•2/3/25, 9:49 AM

Only the few milliseconds is billed, the few seconds of aliveness is mostly a runtime consideration and doesn't really matter for most development architectures, unless you need to persist the DO state specifically between calls

jed•2/3/25, 12:22 PM

hey all, i've got a setup in which a worker serves an app (html / js) and an api (websocket). to prevent app deployment from resetting connections, i've got the app on a service binding. my question is this: what's the best way for the api to report app deployment events in realtime? if outgoing DO websockets could hibernate, i'd have the app server host a websocket that just reported its version on connection, but given that's not possible, what's the easiest way to achieve this? i'd rather not use the DO list API or open two websockets from the browser... would long polling from DO to app server be viable? i'd like the events to be as realtime as possible to allow instant reloads. thanks!

Ggeg @quan @1984 Ford Laser It was definitely because I didn't send pings when not se...

quan•2/3/25, 2:07 PM

oh, how long do you do the interval now? i am trying a timeout of 15 seconds to send a ws.send("") from react client to the websocket server, then server response back

i had more success with SSE before, but not sure why on ws it keeps disconnecting me from the server after around 30seconds

Qquan oh, how long do you do the interval now? i am trying a timeout of 15 seconds to ...

geg•2/3/25, 2:15 PM

If your client is in the browser this shouldn't be necessary. I was talking about websocket pings (not messages).

Rahul gangotri•2/3/25, 3:22 PM

I am running a browser rendering that goto a url and fetch text content from the page inside a durable object

i also reuse sessions and pick any free random session to decrease the load time

still many url's are taking upto 15 sec. i am waiting until networkidle0

can someone please help me optimize this?

Rahul gangotri•2/3/25, 3:23 PM

when i used to run playwright on my own machine and scrape using that, it takes 2-3 sec for page to load

Rahul gangotri•2/3/25, 3:23 PM

same page here is taking 10-15 sec.

DDiamondDragon I'd also be curious to learn more about how people are building apps where you h...

James•2/3/25, 4:06 PM

I read aggregate data a lot more than write transactional, so I usually aggregate to central DB on writes to source tables (in DO) in real time.

For example, rather than query all DOs for some cross DO action like, 'last updated TS', I'll sync up a central log of all 'last updated TS' mapped to the DO.

Llambrospetrou First of all `await setAnotherAlarm()` does not set a new alarm, it overrides th...

lambrospetrou•2/3/25, 7:00 PM

Circling back to this, if in your alarm handler you set a new alarm, and your handler throws AFTER, there won't be a retry since a new alarm has been successfully set.

Best practice for alarms, always set the new alarm AFTER your processing is done, so that exceptions are retried properly, if you need the retries.

RRahul gangotri I am running a browser rendering that goto a url and fetch text content from the...

Rahul gangotri•2/3/25, 10:56 PM

Anyone help on this?

Nnil wrote a util to help with maximum 6 concurrent network requests... has anyone do...

lambrospetrou•2/3/25, 11:53 PM

I have a similar, but simpler approach in my DO helper package: https://github.com/lambrospetrou/durable-utils/blob/main/src/do-sharding.ts#L154-L187

That's a bit specific to the specific class logic, but it does the same thing as yours, limit processing among N things with a given concurrency. Check it out in case it simplifies anything on your side.
I didn't know about the p-limitp-limit package you used though, so maybe that's enough already.

1984 Ford Laser•2/4/25, 5:45 AM

I wish that stubs would automatically proxy calls to things like get()get()/put()put() to be able to access instance storage from another worker/DO directly

Llambrospetrou I have a similar, but simpler approach in my DO helper package: https://github.c...

森

森@VerseLink (RW)•2/4/25, 6:57 AM

@nil p-limit should work, it's pretty much the same logic, it does what you want, I've used it before to queue up subrequests

Exception•2/4/25, 8:04 AM

I want to do some analysis on the values stored in a DO object. I current do, one DO per user session uniquely identified by an id. Is it wise to create another DO that keeps a list of all created sessions/DO and have it retrieve the data for all sessions? I'm also thinking I can forward the session data to a central DO that holds the data for all sessions. Or is there a better approach?

Nnil @lambrospetrou Does using a DO get around the simultaneous open connections of 6...

lambrospetrou•2/4/25, 9:37 AM

Probably, but you do pay all those DOs being active, so it depends on your use-case if it's better.

11984 Ford Laser I wish that stubs would automatically proxy calls to things like `get()`/`put()`...

lambrospetrou•2/4/25, 9:39 AM

You mean instead of adding the 2 lines yourself, or is there something else in this?

SnowBall•2/4/25, 10:07 AM

SnowBall•2/4/25, 10:08 AM

有中国的哥们吗

Llambrospetrou You mean instead of adding the 2 lines yourself, or is there something else in t...

1984 Ford Laser•2/4/25, 10:22 AM

I mean when you put it like that...

Nnil wrote a util to help with maximum 6 concurrent network requests... has anyone do...

sathoro•2/4/25, 11:05 AM

why do you need to do this? it will queue the requests automatically

Nnil Might have been a wrangler dev issue but I ran into a resource limit: "Response ...

Isaac McFadyen•2/4/25, 7:16 PM

Yeah I've gotten this as well in production

iLearn•2/4/25, 8:43 PM

Are the options "options: { expiration?: number, expirationTtl?: number, metadata?: object }" available for Workers-KV available for DO-KV storage ?

Checked the link "https://developers.cloudflare.com/durable-objects/api/storage-api/#supported-options-1" and could not find the aforesaid options and hence wanted to confirm.

Cloudflare Docs

Durable Object Storage · Cloudflare Durable Objects docs

The Durable Object Storage API allows Durable Objects to access transactional and strongly consistent storage. A Durable Object's attached storage is private to its unique instance and cannot be accessed by other objects.

IiLearn Are the options "options: { expiration?: number, expirationTtl?: number, metadat...

1984 Ford Laser•2/4/25, 8:47 PM

Nah I don't think so, Workers KV and DO KV are technically different. Like Workers KV will show a

list()

list()

as the keys, whereas DO KV returns the k:v pairs

11984 Ford Laser Nah I don't think so, Workers KV and DO KV are technically different. Like Worke...

iLearn•2/4/25, 8:50 PM

Thanks for sharing your understanding.

iLearn•2/4/25, 8:55 PM

Since, expirationTtl (option in Workers-KV) is not an option for DO-KV, what are the ways to achieve the same functionality where the Key need to be deleted after, say 10 minutes ?

iLearn•2/4/25, 8:56 PM

Since, expirationTtl (option in Workers-KV) is not an option for DO-KV, what are the ways to achieve the same functionality where the Key need to be deleted after, say 10 minutes ?

Original message was deleted

iLearn•2/4/25, 9:03 PM

Thanks for sharing your understanding.
now i am in catch-22 situation .. wanted to use DO-KV to use ready-made infra for race-condition-avoidance and all goodies of Workers-KV .. but, now with the missing option of expirationTtl .. considering to go back to Workers-KV ..

choice is either manage race-condition or manage auto delete of key using expirationTtl

ssandler•2/4/25, 9:44 PM

Has anyone run into troubles using otel-cf-workers for tracing with the latest durable objects sdk? workers-observability

https://blog.cloudflare.com/durable-objects-alarms/ > Under the hood, Alarms are implemented by mak

Similar Threads

Similar Threads

Similar Threads