#v0.19.9: rpc error: code = NotFound
1 messages Β· Page 1 of 1 (latest)
Tried to fully scrap and relaunch a new EC2 runner and still getting the same error.
Downgrading...
Downgrade seems to work fine
Dang, now getting this on v0.19.8 too all over the place. Nothing else has changed. The only thing I've changed was going from v0.19.8 to v0.19.9 and then back. But in between the entire EC2 was relaunched from scratch, so there's nothing remaining. The PR itself that is failing to build is extremely tiny, just upgrading a GitHub action "checkout" from v4 to v6.
@storm otter are you re-using the state between dagger versions by any chance? Like re-attached EBS storage or something?
@vague lodge Caught another one
@vague lodge is that a known issue ?
i'm having the same issue after upgrading
it's very weird. I have a file passed to a module function. if i leave it untouched. The function completes successfully. If i do a change to the file (as simple as adding a whitespace) I encounter this error
! failed to receive stat message: rpc error: code = NotFound desc = get full root path: rpc error: code = NotFound desc = eval symlinks: lstat /source: no such file or directory```
if remove the whitespace it runs fine again
can replicate on two different workstation on 0.19.9
Are you sure you're not hitting a cache hit ? π
cache should be invalidated as the file changed right ?
@eternal gyro yes it's a known issue
(scoped this back a bit from the original more general goal of multiple cache keys per result, there was more stuff to figure out but this ended up being needed ASAP to fix a problem a user was hit...
is the workaround to revert back to 0.19.8 ?
Fix will go out in the next release
Not necessarily, when removing the space again, your file could be in the same state as it was in your first run, resulting to the same content addressed hash and thus not even recompute. As it was successful once, it would work, whilst it might have failed on the other case (due to above bug).
But glad to see that we fixed it ahahah
i see. i'll revert back to 0.19.8 until 0.19.10 is out π
hmmm getting the same issue on v0.19.8 now. Help π€£
I think this issue goes way back to 19.6 or something around that IIRC
what version were you using before?
0.19.8
are you using persistent runners or something?
just checking if you might have to prune the cache
nah it's on my local
ok.. it's strange that it's failing in v0.19.8 if it was working before then π€
when downgrading locally the cache gets cleared automatically
can you please try: dagger core engine local-cache prune
- 1 st run post pruning worked OK
- changed file content
- 2nd run failed
On v0.19.8, weeeird π
yep
do you mind nuking your dagger volume ?
sure thing
do you want commands ? or is it ok ?
all good i nuked the docker engine container and delete the associated docker volume
i'm running the steps now just takes a bit of time (building an app)
same as before
- 1 st run post pruning worked OK
- changed file content
- 2nd run failed
wanna go on dev-audio ?
From syncing with Julien, it seems that this is not fixed by main. It's very suspicious and probably another bug, trying to repro on the side
I repro'd it locally, ty both for the info 
We are gonna do a release tomorrow to get some other emergent fixes needed out, I will see if I can sneak in a fix for this one too but hard to say until I start looking into the cause
@twin cradle Guillaume mentioned that it was only when using external module (git )
so that's part of the reason we couldnt replicate locally
yeah I had to push a dependency module to a tmp git branch to trigger it
ftr (also in case I lose this), repro is
dependency module code (pushed to a git repo):
package main
import (
"context"
"dagger/dep/internal/dagger"
)
type Dep struct {
Dir *dagger.Directory
Ref *dagger.GitRef
}
func New(
dir *dagger.Directory,
) *Dep {
return &Dep{
Dir: dir,
}
}
func (m *Dep) WithRef(
ctx context.Context,
ref *dagger.GitRef,
) (*Dep, error) {
m.Ref = ref
return m, nil
}
func (m *Dep) Fn(
ctx context.Context,
) (*dagger.Directory, error) {
return m.Ref.Tree(), nil
}
main module code with dependency on the above through a git dep:
package main
import (
"context"
"dagger/test/internal/dagger"
)
type Test struct {
//+private
Ref *dagger.GitRef
//+private
Dep *dagger.Dep
}
func New(
// +defaultPath="."
ref *dagger.GitRef,
//+defaultPath="crap"
source *dagger.Directory,
) *Test {
return &Test{
Ref: ref,
Dep: dag.Dep(source),
}
}
func (m *Test) Fn(
ctx context.Context,
//+defaultPath="config/config.local.js"
configFile *dagger.File,
) (*dagger.Directory, error) {
return m.Dep.WithRef(m.Ref).Fn().WithFile("config.js", configFile).Sync(ctx)
}
Then run dagger call fn, modify config/config.local.js, run dagger call fn again and it fails
Fairly sure this is the exact same problem as was fixed here: https://github.com/dagger/dagger/pull/11350, except that PR missed the specific case of "contextual GitRef that's actually sourced from a local dir". So hopefully I can just follow the same pattern as the previous fix but applied to this case
fix https://github.com/dagger/dagger/pull/11688, can't repro anymore with that, unless something unexpected pops up should make it in release tomorrow!
@sturdy ravine @trim prairie Julien is interested in a demo of native CI
let's do it! @eternal gyro do you want to use this scheduling link? Let us know if you can't find a good time. Talk soon!
are you and @sturdy ravine on the PT timezone ? Latest is 8am my time (Sydney time) on the calendar could you guys do 1hr later ? (kid drop off and all)
Hey @trim prairie @sturdy ravine ! Sorry to ping you directly. Would 9am Sydney time (2pm SF time ) work for you ? i can do any day of the work week !
yes I can do that time wednesday or Thursday. Might be too late for kyle otherwise it would be an option on the booking link (he's on east coast). No worries you'll just get a slightly less good demo π
demo gods are usually nicer with him !