I'm not seeing much in the way of errors on our side - I can see the "Container" (in this case the S

K

kianOP•10/14/25, 7:00 PM

Is it always reproducible for you when using that

writeFile

writeFile

writeFile

writeFile API?

Kkian I'm not seeing much in the way of errors on our side - I can see the "Container"...

C

christianklotz•10/14/25, 8:28 PM

Thank you for looking into it! Yeah, it's always

writeFile

writeFile

writeFile

writeFile. Interestingly I get this on

execute

execute

:

"[HTTP Client] Error executing in session:",
"Error: Session 'sandbox-default' not found"

"[HTTP Client] Error executing in session:",
"Error: Session 'sandbox-default' not found"

"[HTTP Client] Error executing in session:",
"Error: Session 'sandbox-default' not found"

"[HTTP Client] Error executing in session:",
"Error: Session 'sandbox-default' not found"

Not sure if related.

I am whipping up a more isolated implementation right now to see it still breaks

Cchristianklotz Thank you for looking into it! Yeah, it's always `writeFile`. Interestingly I ge...

C

christianklotz•10/14/25, 9:58 PM

After a bit more digging it feels like it's more an issue with the Sandbox SDK itself than containers. The basic setup works for me locally but makes the worker fail with

"Failed to create session"

"Failed to create session"

"Failed to create session"

"Failed to create session" once deployed.

A

Alex Dahl•10/15/25, 9:14 AM

Has anyone checked the network speed for R2 <=> containers? My workload is for semi-big files 1GB~ on R2, would be awesome if the containers had access to cloudflares internal network

C

CodingCoop | Co-Founder Nullshot•10/15/25, 9:21 AM

@kian | Containers any plans to set instance type dynamically or do i need a unique DO / Container class per instance type? I want users to choose their instance size.

Cchristianklotz After a bit more digging it feels like it's more an issue with the Sandbox SDK i...

C

christianklotz•10/15/25, 9:34 AM

Update on the basic setup: this works now which may have purely a matter of me having made requests before the instances to be provisioned. Still looking into why the Sandbox SDK fails

C

CodingCoop | Co-Founder Nullshot•10/15/25, 9:38 AM

You may need to create the sesion first @christianklotz - then run the command baesd on the session object returend by the createSession object. I haven't used the default session commands personally though ( the direct ones in the sandbox ) but ive been working out of the sandbox createSession with unique ids passed in which has worked well for past month or so.

C

christianklotz•10/15/25, 9:39 AM

Yeah, that's what I'm doing. And it works locally, for some reason it doesn't once deployed

C

CodingCoop | Co-Founder Nullshot•10/15/25, 9:48 AM

thats odd if its working locally, but not remotely.

I've heavily modified the sandbox sdk, but one thing I recall was that when i create session in multiple places, it would create a new session although it was the same id, so I created it so it woudl return the same session at the container level under isolation.ts

/**
 * Manages isolated sessions for command execution.
 * Each session maintains its own state (pwd, env vars, processes).
 */
export class SessionManager {
  private sessions = new Map<string, Session>();
  
  async createSession(options: SessionOptions): Promise<Session> {
    // Validate cwd if provided - must be absolute path
    if (options.cwd) {
      if (!options.cwd.startsWith('/')) {
        throw new Error(`cwd must be an absolute path starting with '/', got: ${options.cwd}`);
      }
    }
    
    // Return existing session if it already exists
    const existing = this.sessions.get(options.id);
    if (existing) {
      console.log(`[SessionManager] Returning existing session '${options.id}'`);
      return existing;
    }
    
    // Create new session only if it doesn't exist
    const session = new Session(options);
    await session.initialize();
    
    this.sessions.set(options.id, session);
    console.log(`[SessionManager] Created new session '${options.id}'`);
    return session;
  }

/**
 * Manages isolated sessions for command execution.
 * Each session maintains its own state (pwd, env vars, processes).
 */
export class SessionManager {
  private sessions = new Map<string, Session>();
  
  async createSession(options: SessionOptions): Promise<Session> {
    // Validate cwd if provided - must be absolute path
    if (options.cwd) {
      if (!options.cwd.startsWith('/')) {
        throw new Error(`cwd must be an absolute path starting with '/', got: ${options.cwd}`);
      }
    }
    
    // Return existing session if it already exists
    const existing = this.sessions.get(options.id);
    if (existing) {
      console.log(`[SessionManager] Returning existing session '${options.id}'`);
      return existing;
    }
    
    // Create new session only if it doesn't exist
    const session = new Session(options);
    await session.initialize();
    
    this.sessions.set(options.id, session);
    console.log(`[SessionManager] Created new session '${options.id}'`);
    return session;
  }

/**
 * Manages isolated sessions for command execution.
 * Each session maintains its own state (pwd, env vars, processes).
 */
export class SessionManager {
  private sessions = new Map<string, Session>();
  
  async createSession(options: SessionOptions): Promise<Session> {
    // Validate cwd if provided - must be absolute path
    if (options.cwd) {
      if (!options.cwd.startsWith('/')) {
        throw new Error(`cwd must be an absolute path starting with '/', got: ${options.cwd}`);
      }
    }
    
    // Return existing session if it already exists
    const existing = this.sessions.get(options.id);
    if (existing) {
      console.log(`[SessionManager] Returning existing session '${options.id}'`);
      return existing;
    }
    
    // Create new session only if it doesn't exist
    const session = new Session(options);
    await session.initialize();
    
    this.sessions.set(options.id, session);
    console.log(`[SessionManager] Created new session '${options.id}'`);
    return session;
  }

/**
 * Manages isolated sessions for command execution.
 * Each session maintains its own state (pwd, env vars, processes).
 */
export class SessionManager {
  private sessions = new Map<string, Session>();
  
  async createSession(options: SessionOptions): Promise<Session> {
    // Validate cwd if provided - must be absolute path
    if (options.cwd) {
      if (!options.cwd.startsWith('/')) {
        throw new Error(`cwd must be an absolute path starting with '/', got: ${options.cwd}`);
      }
    }
    
    // Return existing session if it already exists
    const existing = this.sessions.get(options.id);
    if (existing) {
      console.log(`[SessionManager] Returning existing session '${options.id}'`);
      return existing;
    }
    
    // Create new session only if it doesn't exist
    const session = new Session(options);
    await session.initialize();
    
    this.sessions.set(options.id, session);
    console.log(`[SessionManager] Created new session '${options.id}'`);
    return session;
  }

C

CodingCoop | Co-Founder Nullshot•10/15/25, 9:48 AM

But that wouldn't effect your local vs cloud...

C

christianklotz•10/15/25, 10:12 AM

A noticeable difference compared to the basic setup described in the Sandbox SDK docs is that the container state shown in the dashboard remains Active (instead of going back to Ready). Not sure if this is related tho

A

Allen•10/15/25, 12:06 PM

Hi, is there any way to avoid cold start? Because I have very high requirements for real-time performance.

P

pandalambda2•10/15/25, 12:12 PM

ping it to death - looks at cron triggers, but nothing will guarantee it 100%, can be evicted for different reasons all the way down to a node failure

A

Allen•10/15/25, 1:18 PM

Thanks for your suggestion. I have considered this method before, but it may not work because of the global node. The location I ping will be different every time, so this cannot solve the problem.

Ppandalambda2 ping it to death - looks at cron triggers, but nothing will guarantee it 100%, c...

A

Allen•10/15/25, 1:18 PM

@pandalambda2

P

pandalambda2•10/15/25, 1:20 PM

if you use the same sessionid it will

P

pandalambda2•10/15/25, 1:20 PM

since the worker which could be anywhere on the network, will route to the specific DO

P

pandalambda2•10/15/25, 1:20 PM

and DO is 1:1 with the container

T

Thomas•10/15/25, 4:05 PM

fyi the automatically-set runtime environment variables mentioned here (specifically the durable object id) is just set to 64 'x's in dev https://developers.cloudflare.com/containers/platform-details/environment-variables/

is that intentional? if so it'd be lovely if mentioned in the docs

Cloudflare Docs

Environment Variables

The container runtime automatically sets the following variables:

JJon M.Any plans to actually show rollout progress when you update a container?

K

kianOP•10/15/25, 5:12 PM

There is already a rollout bar in the dashboard, are you referring to something else?

TThomas fyi the automatically-set runtime environment variables mentioned here (specific...

K

kianOP•10/15/25, 5:13 PM

That's intentional, we'll add it in as a note to the docs. It usually represents the DO backing your Container but in local development it's just powered by Docker so there isn't one.

K

kianOP•10/15/25, 5:14 PM

Are your Containers gracefully shutting down? If you had a

max_instances

max_instances

max_instances

max_instances of 10 and rollout steps of

20, 100

20, 100

20, 100

20, 100 then we can't progress past 20% (aka 2 instances) until those 2 instances have stopped & we've brought them back up with the new settings.

K

kianOP•10/15/25, 5:14 PM

If they don't handle SIGTERM themselves then we give them 15 minutes before they're killed.

K

kianOP•10/15/25, 6:30 PM

Are you responding to a request and then starting compute like "hey, start this task -> yep, started"?

K

kianOP•10/15/25, 6:30 PM

The current lifecycle model of

sleepAfter

sleepAfter

sleepAfter

sleepAfter kinda has a web server focus where once you respond,

sleepAfter

sleepAfter

sleepAfter

sleepAfter starts and then we'll send a

SIGTERM

SIGTERM

SIGTERM

SIGTERM to your Container.

K

kianOP•10/15/25, 6:30 PM

There is an

onActivityExpired

onActivityExpired

onActivityExpired

onActivityExpired lifecycle hook in the Containers class where you can do your own logic to decide if

sleepAfter

sleepAfter

sleepAfter

sleepAfter should start counting down

K

kianOP•10/15/25, 6:46 PM

https://github.com/cloudflare/containers/

K

kianOP•10/15/25, 6:46 PM

Lots of the docs are in the README of the library right now but we are working on moving this library into the runtime so it just works as a normal binding

K

kianOP•10/15/25, 6:46 PM

onActivityExpired

onActivityExpired

onActivityExpired

onActivityExpired is basically your way to intercept the

sleepAfter

sleepAfter

sleepAfter

sleepAfter invocation for now

K

kianOP•10/15/25, 6:47 PM

There's also some other APIs on the way like setting a hard timeout (my Container should stop running after 20 minutes from being started)

D

dev•10/15/25, 6:58 PM

what's the difference between a sandbox and a container?

K

kianOP•10/15/25, 7:12 PM

Containers is the primitive where the code is running - the Sandbox SDK runs your commands inside Containers

K

kianOP•10/15/25, 7:13 PM

You could build your own Sandbox SDK with Containers for example. This just already has it all setup for you with Node.JS / Python / Git etc installed & setup

D

dev•10/15/25, 8:31 PM

gotcha, so it's a container + some goodies to make running code inside the container easier

N

nneaowwplane•10/16/25, 3:09 PM

Hi, I was talking to some of the cloudflair team and one of the comments was that a minecraft server can be run in a container now. Is there any more info on this? as this sounds like a fun side project to play around with.

JJon M.I keep getting SIGTERM shutdowns and I can't figure out why. So frustrating. Is ...

K

kianOP•10/16/25, 3:18 PM

The primary reason is usually that