HeadlessChromium.Puppeteer.Dotnet.OS.Fork 1.2.1.3

HeadlessChromium.Puppeteer.Lambda.Dotnet

Packages everything you need to run PuppeteerSharp in AWS Lambda or a Docker container on Chromium into a Nuget Package

Description

The chromium binary for this project has been extracted from the NPM project Sparticuz/chromium. It is automatically extracted to /tmp/chromium at runtime. For now, the chromium binary only supports x86_64

Supported platforms, auto-detected at runtime:

• AWS Lambda dotnet8/dotnet10 (Amazon Linux 2023 based).
• Ubuntu 22.04 LTS Docker containers (e.g. mcr.microsoft.com/dotnet/runtime:10.0-based images before they moved to 24.04, or any ubuntu:22.04-based image). Ubuntu 22.04 reuses the AL2023 dependency bundle - it was validated to run unmodified in a vanilla ubuntu:22.04 container, since AL2023's bundled shared libraries (NSS, NSPR, expat) satisfy what Ubuntu 22.04 needs and there is no separate Sparticuz-published Ubuntu build.
• Chainguard / Wolfi containers (e.g. cgr.dev/chainguard/aspnet-runtime, or any Wolfi-based image reporting ID=wolfi or ID=chainguard). Wolfi reuses the AL2023 dependency bundle and adds wolfi.tar.br on top of it, which carries libsqlite3.so.0 - see below. End-to-end renders of both a local document and a live HTTPS page on the real image confirm the combination. See the assumptions below.

Why Wolfi needs an extra archive

A dependency walk of the Sparticuz binary resolves every library it links against from either the image or al2023.tar.br, which is why Wolfi first looked like a pure reuse of the AL2023 bundle. It is not. The first time a page needs TLS, NSS loads its PKCS#11 module libsoftokn3.so with dlopen(), and that module needs libsqlite3.so.0 for the SQLite-backed certificate database. Amazon Linux and Ubuntu ship libsqlite3.so.0 in the base image; Chainguard's distroless images do not, and failing NSS initialisation is fatal:

ERROR:crypto/nss_util.cc:256 Error initializing NSS with a persistent database
    (sql:/tmp/.local/share/pki/nssdb): libsqlite3.so.0: cannot open shared object file
FATAL:crypto/nss_util.cc:146  nss_error=-5925

Chromium aborts there, so the symptom reaching the caller is a TargetClosedException or NavigationException ("Target closed", "the remote party closed the WebSocket connection") in the middle of GoToAsync - and only for pages fetched over the network. Rendering a local document via SetContentAsync never touches NSS and keeps working, which is what makes this failure easy to miss.

payloads/wolfi.tar.br supplies the library, is extracted only on Wolfi, and is regenerated by tools/build-wolfi-payload.sh.

Chainguard / Wolfi assumptions

Chainguard support relies on three properties of the shipping images, all verified rather than assumed:

• Non-root, uid 65532. The images run as nonroot with APP_UID=65532. Chromium and its libraries are extracted to /tmp, which is mode 1777 and therefore the one location this uid can write to. A read-only root filesystem is not supported - /tmp is hardcoded.
• Distroless. There is no shell, so nothing can be diagnosed from inside a running container. Every failure path is therefore reported through log output and exception messages; enable DEBUG logging to see them.
• x86_64 only. The Sparticuz Chromium payload is x86_64. Chainguard also publishes arm64 images, which this library cannot serve.

No change to your Dockerfile is required - no apk packages, no hand-installed Chromium. See sample/SampleAspNet-chainguard/ for a working ASP.NET example.

Platform detection

Platform detection runs automatically and needs no configuration in the common case:

1. If the CHROMIUM_PLATFORM_OVERRIDE environment variable is set to a supported value (al2023, ubuntu-22.04, or wolfi - with chainguard accepted as an alias for wolfi), it's used immediately.
2. Otherwise, /etc/system-release-cpe is checked for Amazon Linux 2023.
3. Otherwise, /etc/os-release is checked. Matching is on the ID field (falling back to NAME when ID is absent), compared against known identifiers: ubuntu with VERSION_ID="22.04", or wolfi/chainguard. Wolfi's VERSION_ID is a rolling build date and is deliberately ignored.

If none of the above match - an invalid override value, or an unrecognized OS - ExtractChromium() throws a ChromiumExtractionException. The message names the OS that was detected, lists the supported platform identifiers, and points at CHROMIUM_PLATFORM_OVERRIDE. A caller that knows better can still set ChromiumExtractor.OperatingSystem before calling ExtractChromium(), which bypasses detection entirely.

Troubleshooting

Enable DEBUG-level logging on ChromiumExtractor to see which detection method was used and what platform was resolved. A ChromiumExtractionException mentioning the supported platform list means detection failed (invalid CHROMIUM_PLATFORM_OVERRIDE value, or no matching OS file) - set CHROMIUM_PLATFORM_OVERRIDE explicitly to unblock it, e.g.:

docker run -e CHROMIUM_PLATFORM_OVERRIDE=ubuntu-22.04 myimage
docker run -e CHROMIUM_PLATFORM_OVERRIDE=wolfi myimage

Upgrading

Unrecognized platforms now fail fast. Previously, when platform detection could not resolve a platform, a WARNING was logged and the OS-specific dependency archive was simply skipped. Chromium then failed later with a dynamic-linker error, which on a distroless image is close to undiagnosable. ExtractChromium() now throws a ChromiumExtractionException at that point instead.

• If you already set ChromiumExtractor.OperatingSystem before calling ExtractChromium(), nothing changes - that still bypasses detection.
• If you relied on warn-and-continue and set the platform afterwards, move the assignment before the call, or set CHROMIUM_PLATFORM_OVERRIDE.
• On a supported platform, nothing changes.

The unreleased chainguard-20230214 platform identifier has been removed in favour of wolfi. It never shipped in a release and its detection could never fire, so no consumer can depend on it; chainguard is accepted as an alias for wolfi.

Usage

Screenshot a URL as a byte[].

var browserLauncher = new HeadlessChromiumPuppeteerLauncher(logger);

using(var browser = await browserLauncher.LaunchAsync())
using(var page = await browser.NewPageAsync())
{
    await page.GoToAsync(url);
    return await page.ScreenshotDataAsync();
}

For more use cases see the PuppeteerSharp documentation

Projects using this library

These projects are using this library and are good examples of how you might consume this nuget package

• HtmlToPDF

Building

To build locally:

.\build.ps1 -Target Build