Install script does not exit with a non-zero exit code on a broken pipe

[klemmchr]

Our nightly build of our Docker images succeeded, however the pipeline shows the following error:

02:06:38 [DBG] dotnet-install: Attempting to download using aka.ms link https://builds.dotnet.microsoft.com/dotnet/Sdk/8.0.407/dotnet-sdk-8.0.407-linux-x64.tar.gz
02:06:43 [DBG] dotnet-install: Remote file https://builds.dotnet.microsoft.com/dotnet/Sdk/8.0.407/dotnet-sdk-8.0.407-linux-x64.tar.gz size is 212950827 bytes.
02:06:43 [DBG] dotnet-install: Extracting archive from https://builds.dotnet.microsoft.com/dotnet/Sdk/8.0.407/dotnet-sdk-8.0.407-linux-x64.tar.gz
02:06:43 [DBG] dotnet-install: Downloaded file size is 212950827 bytes.
02:06:43 [DBG] dotnet-install: The remote and local file sizes are equal.
02:06:55 [DBG] dotnet-install: Installed version is 8.0.407
02:06:55 [DBG] dotnet-install: Adding to current process PATH: `/opt/.dotnet`. Note: This change will be visible only when sourcing script.
02:06:55 [DBG] dotnet-install: Note that the script does not resolve dependencies during installation.
02:06:55 [DBG] dotnet-install: To check the list of dependencies, go to https://learn.microsoft.com/dotnet/core/install, select your operating system and check the "Dependencies" section.
02:06:55 [DBG] dotnet-install: Installation finished successfully.
02:06:55 [DBG] curl: (56) Send failure: Broken pipe

ENV DOTNET_ROOT /opt/.dotnet
ENV PATH $PATH:$DOTNET_ROOT:$DOTNET_ROOT/tools
RUN apt-get update && \
    DEBIAN_FRONTEND=noninteractive \
    apt-get install -y --no-install-recommends \
        ca-certificates \
        libc6 \
        libgcc-s1 \
        libgssapi-krb5-2 \
        libicu74 \
        liblttng-ust1 \
        libssl3 \
        libstdc++6 \
        libunwind8 \
        libatomic1 \
        tzdata \
        zlib1g && \
    curl -sSL https://dot.net/v1/dotnet-install.sh | bash /dev/stdin --channel 8.0 --install-dir $DOTNET_ROOT && \
    curl -sSL https://dot.net/v1/dotnet-install.sh | bash /dev/stdin --channel 9.0 --install-dir $DOTNET_ROOT && \
    chmod -R 555 $DOTNET_ROOT && \
    rm -rf /var/lib/apt/lists/*

.NET 8 was installed properly while .NET 9 was not installed.

02:12:45 [DBG] Step 32/43 : RUN dotnet --info
02:12:48 [DBG]  ---> Running in cf27c4516910
02:12:48 [DBG] .NET SDK:
02:12:48 [DBG]  Version:           8.0.407
02:12:48 [DBG]  Commit:            cac65f27eb
02:12:48 [DBG]  Workload version:  8.0.400-manifests.20e9918b
02:12:48 [DBG]  MSBuild version:   17.11.26+2b19be476
02:12:48 [DBG] Runtime Environment:
02:12:48 [DBG]  OS Name:     ubuntu
02:12:48 [DBG]  OS Version:  24.04
02:12:48 [DBG]  OS Platform: Linux
02:12:48 [DBG]  RID:         linux-x64
02:12:48 [DBG]  Base Path:   /opt/.dotnet/sdk/8.0.407/
02:12:48 [DBG] .NET workloads installed:
02:12:48 [DBG] Configured to use loose manifests when installing new manifests.
02:12:48 [DBG] There are no installed workloads to display.
02:12:48 [DBG] Host:
02:12:48 [DBG]   Version:      8.0.14
02:12:48 [DBG]   Architecture: x64
02:12:48 [DBG]   Commit:       1584e49360
02:12:48 [DBG] .NET SDKs installed:
02:12:48 [DBG]   8.0.407 [/opt/.dotnet/sdk]
02:12:48 [DBG] .NET runtimes installed:
02:12:48 [DBG]   Microsoft.AspNetCore.App 8.0.14 [/opt/.dotnet/shared/Microsoft.AspNetCore.App]
02:12:48 [DBG]   Microsoft.NETCore.App 8.0.14 [/opt/.dotnet/shared/Microsoft.NETCore.App]
02:12:48 [DBG] Other architectures found:
02:12:48 [DBG]   None
02:12:48 [DBG] Environment variables:
02:12:48 [DBG]   DOTNET_ROOT       [/opt/.dotnet]
02:12:48 [DBG] global.json file:
02:12:48 [DBG]   Not found
02:12:48 [DBG] Learn more:
02:12:48 [DBG]   https://aka.ms/dotnet/info
02:12:48 [DBG] Download .NET:
02:12:48 [DBG]   https://aka.ms/dotnet/download

[JanProvaznik]

thanks for the report, seems like a bad behavior in an important use case, looking into a fix

[JanProvaznik]

Hold on, now I doubt this happens in the install script. We're having intermittent issues with the https://dot.net/v1/dotnet-install.sh link the past 2 days #582 . This might be a symptom of the curl sending an empty message.
e.g. this succeeds and what you're experiencing might be the same case echo "" | bash /dev/stdin --channel 9.0

We're sorry this affects you, you can consider downloading the script to a file and validating that it's present before running it. Or as the aforementioned issue suggests; currently the https://builds.dotnet.microsoft.com/dotnet/scripts/v1/dotnet-install.sh seems more stable.

[baronfel]

Going to close this as a duplicate of #582 - we're working with the internal service to determine the cause of the disruption, and longer term we need to migrate off of this to a more globally-accessible service, which we're tracking in https://github.com/dotnet/release/issues/1176.

[klemmchr]

Hold on, now I doubt this happens in the install script. We're having intermittent issues with the https://dot.net/v1/dotnet-install.sh link the past 2 days #582 . This might be a symptom of the curl sending an empty message. e.g. this succeeds and what you're experiencing might be the same case echo "" | bash /dev/stdin --channel 9.0

We're sorry this affects you, you can consider downloading the script to a file and validating that it's present before running it. Or as the aforementioned issue suggests; currently the https://builds.dotnet.microsoft.com/dotnet/scripts/v1/dotnet-install.sh seems more stable.

Good catch, I was expecting curl to exit with a non-zero code if such an error happens. I'm thankful that you're investigating this issue and I highly apprechiate your work towards making the install script server more resilient.

However, there can be numerous ways why network fails in such instances, also on the client side. Since I've taken the install code from the docs, how do I need to modify my command so that it fails with a non-zero exit code in such instances? I would rather have a failed Docker build than an invalid Docker image.

[JanProvaznik]

@klemmchr https://stackoverflow.com/questions/69684133/dl4006-warning-set-the-shell-option-o-pipefail-before-run-with-a-pipe-in-it this might help you