Strange issue running image fedora 36 vs 37 - rocker/rstudio::devel works, as does home build

[r-cheologist]

I have just switched my base OS from Fedora 36 to 37 and encountered the following:

1. When executing

     podman run \
       -d \
       -p 127.0.0.1:8787:8787 \
       -v /tmp:/tmp \
       -e ROOT=TRUE   \
       -e DISABLE_AUTH=TRUE \
       docker.io/bioconductor/bioconductor_docker
   

   the container comes up fine, but using a browser, the RStudio server instance can't be reached. The spinner shows up, but then:

   Could not connect to the R session on RStudio server.
   Unable to connect to service (1)

2. When doing the same thing using docker.io/rocker/rstudio:devel, the base image to the bioconductor one, all is fine.

3. Additionally all is fine when I build the bioconductor image locally...

I have been unable to debug this. Does anyone have any insight into why Fedora 37 might bulk at the BC image from dockerhub (and all containers that are based on it and still around from before the 36/37 switch).

[r-cheologist]

So from Fedora 35 to 36 there apparently was a major podman version switch with implied networking backend changes that were not backward compatible... but I can't find anything for the 36/37 switch...

[almahmoud]

Could you give ghcr.io/bioconductor/bioconductor:RELEASE_3_16 a try? The dockerhub images might be the problem especially where you say rebuilding locally works. I've been using GHCR images since taking over, and Dockerhub autobuilder might be out of sync at this point. I am looking at merging the two now

[r-cheologist]

That image gives me exactly the fame trouble as does the docker. Io one...

…

On March 8, 2023 6:36:38 PM UTC, Alexandru Mahmoud ***@***.***> wrote: Could you give ghcr.io/bioconductor/bioconductor:RELEASE_3_16 a try? The dockerhub images might be the problem especially where you say rebuilding locally works. I've been using GHCR images since taking over, and they might be out of sync. I am looking at resyncing them now -- Reply to this email directly or view it on GitHub: #70 (comment) You are receiving this because you authored the thread. Message ID: ***@***.***>

[almahmoud]

I am not entirely sure what the problem could be in that case... Did the podman version change between the two VMs?
I found this issue which might be related containers/podman#11528 ?
If you haven't already, perhaps try with -p 8787:8787 instead of -p 127.0.0.1:8787:8787

[r-cheologist]

As I wrote above: using the docker.io/rocker/rstudio image (which the bioconductor one uses as a base) works fine ... the port remapping issue should hit both ...

[r-cheologist]

The issue persists as of 2023/03/31, BUT docker.io/bioconductor/bioconductor_docker:devel actually works!?

[almahmoud]

I am still not sure what this could be, as it works okay whenever I try to run it, but I am not using podman on Fedora...
One thing to note, but again this is largely a guess, this issue might be related? containers/podman#8132
I can confirm the new image does have empty layers, as we're using that to support various bases for arm64 vs amd64 architectures. The only change in the current devel vs 3.16 published images is the lack of support for arm64, so my best guess is it's either related to the fact that images exist for both archs, or the new build file with empty layers...

[almahmoud]

@r-cheologist Is this still persisting?

[r-cheologist]

Oh boy. I have been neglecting this, but Now I switched from fedora 37 to 38 and end up having exactly the same issues AGAIN!? For both the latestand develimages from docker.io.
The starting call above does not work in my unpriviledged podman setup, neither does setting the password and forgoing the DISABLE_AUTH - as I am unpriviledged using user root and password as given by the environmental variable should work, but it doesn't.