Summary

• Restructures the handling of cache paths for build caching in static builds. This is an improvement that should have happened long ago, and is being done as part of this work because it makes it easier to ensure that the new builds use a new cache path.
• Adds improved debugging support to the static build process. By settingDEBUG_BUILD_INFRA to a non-empty value when running the build, a failure in the build process will drop to a shell in the container being used for building instead of just dying.
• Switches to the actual Debian based static build images. This also requires changing the Docker platform for our ARMv6 builds, since Debian supports ARMv5 instead of ARMv6 as a base for softfp ARM systems.
• Vendors libmnl and libelf for static builds, as Debian does not currently provide a working static library for either of them.
• Fixes issues with building the static copy of bash we bundle:
  • Forces usage of the copy of readline from the build environment instead of bash’s internal implementation, as the internal one does not build correctly.
  • Disables network port redirection support, as it does not build correctly and we don’t actually use it in any of our scripts (nor are we ever likely to want to).
• Fixes issues with building the static copy of curl we bundle:
  • Disables libpsl support since it would not build correctly. Not expected to cause any issues as we do not appear to use any features of curl that rely on the Public Suffix List.
  • Disables brotli support since it would not build correctly. Minimal impact expected (we do use brotlioptionally in a number of cases, but it’s not mandatory).
  • Disables building documentation, since it should have been disabled already and it happens to not build at all in the new static build environment.
• Consolidates compiler flag handling for Netdata itself by factoring out common flags for all build types (this way it’s easier to keep things properly in-sync).

Test Plan

This requires extensive manual testing far beyond what I’ve already done.

Things that need specific attention andMUST be tested before merging:

• Overall functionality of the ARMv6 builds on actual hardware (the underlying architecture used for this specific build is being changed as part of this PR, and while I don’t expect it to actually break anything, we should at least confirm that it appears to be working).
• Handling of non-ASCII hostnames in locales that use an encoding other than UTF-8, needs tested on both glibcAND musl systems (we rely on iconv for this, which in turn usually relies on being able todlopen() things that may be incompatible between glibc versions and may not exist on musl systems).
• Handling of mapping between user/group names and IDs, needs tested on both glibcAND musl systems (NSS within glibc relies heavily ondlopen()).
• Handling of hostname lookups, needs testing on both glibcAND musl systems (NSS again).
• Confirmation that the systemd-journal plugin works across a wide variety of systemd systems (this also relies ondlopen(), but it does so in a way that appears to work fine across distros).

I have not personally tested the first item since I don’t have hardware for it, and I can’t readily test the second item, but I have confirmed that the last three all appear to work on my glibc+systemd Gentoo development system.

Additionally, regular runtime testing is required for a wide variety of platforms.

Manual builds for testing work exactly the same as they do for existing static builds (ensure you have a working Docker or rootless Podman setup, then runpackaging/makeself/build-static.sh <arch> where<arch> is the CPu architecture to build for).