Fly builders running out of disk space causing deploy to fail

Questions / Help Build debugging
1

Starting late yesterday evening our application began failing to deploy due to its build process failing halfway through from disk space exhaustion.

Snippet from the job:

#21 24.06 error: could not compile `crunchy` due to previous error
#21 24.06 warning: build failed, waiting for other jobs to finish...
#21 24.17 error: failed to write /usr/src/app/target/release/deps/rmetaNWeNef/lib.rmeta: No space left on device (os error 28)
#21 24.17 
#21 24.53 LLVM ERROR: IO failure on output stream: No space left on device
#21 24.55 error: build failed

Is there a published limit on the build environment resources that applications can use with remote builders?

2

Hello! We don’t really have a limit, but remote builders are created with 50gb volumes by default. There’s a few workarounds:

  • delete the remote builder app so you get a fresh one on the next build
  • ssh into the builder and run docker system prune to purge cache/dangling layers

And this might work, though I haven’t tried it:

  • create a new volume with the same name and a bigger size (eg flyctl volumes create <name> --size 100gb) then delete the old one. next start should use the bigger one

Expanding volumes is on our roadmap, but no timeframe.

1 Like
3

Is there a way to programmatically get at the builder app name? We’re looking to ssh in and run docker prune from github actions after a deploy.

4

this worked for me:

flyctl apps list | grep fly-builder | head -n1 | awk '{print $1;}' | xargs fly ssh console --command "docker system prune --force" --app

2 Likes
5

Glad to hear you have a solution!

You may have already seen either of these, but I wanted to highlight them in any case:

  • flyctl supports json output too, with the --json or -j flags
  • there’s also an api endpoint for your organization’s builders, as seen on GraphQL Playground – you can also grab the builder with something like
query { 
   personalOrganization { 
   remoteBuilderApp { 
    name 
    }
  }   
}
1 Like
6

Just to throw a +1 on here, I ran into this myself. I ended up just resizing the volume for now, but it would be nice for builders to have some first class way to manage the docker cache since I imagine the infinite docker cache growth will be a common problem for people.

4 Likes