diff options
| author | fanquake <fanquake@gmail.com> | 2023-08-09 11:46:01 +0200 |
|---|---|---|
| committer | fanquake <fanquake@gmail.com> | 2023-08-09 12:09:00 +0200 |
| commit | 492257019d6d6e0f4051433e510502e8040fd6d2 (patch) | |
| tree | 7ee41238994e0b2bbef088e11ffe5e841c729728 /.cirrus.yml | |
| parent | b565485c24c0feacae559a7f6f7b83d7516ca58d (diff) | |
| parent | fad0b67c212dcb8a16fcbda5a74acc959ed4e284 (diff) | |
| download | bitcoin-492257019d6d6e0f4051433e510502e8040fd6d2.tar.xz | |
Merge bitcoin/bitcoin#28087: ci: Use qemu-user through container engine
fad0b67c212dcb8a16fcbda5a74acc959ed4e284 ci: Use qemu-user through container engine (MarcoFalke)
Pull request description:
Currently the CI containers always run on the host architecture, and only wrap `bitcoind` into `qemu-user` when needed. This has many issues:
* The `i386` tasks can not be run on non-x86 hosts.
* `config.guess` isn't present when building the CI image, which is fine. But it prints a warning, see https://github.com/bitcoin/bitcoin/pull/27739#pullrequestreview-1446580353
* The python tests are run on the host architecture, making it harder to find architecture specific bugs. See for example https://github.com/bitcoin/bitcoin/pull/27529#issuecomment-1521387810
* All modern container engines support automatic dispatch to qemu-user, so it seems redundant to re-invent the wheel.
Fix all issues by:
* removing `HOST` from `ci/test/00_setup_env.sh`.
* removing `QEMU_USER_CMD` and `ci/test/wrap-qemu.sh`.
* removing `DPKG_ADD_ARCH` where possible, and pruning `PACKAGES` where possible.
* specifying the architecture in `CI_IMAGE_NAME_TAG` to be used by the container engine.
ACKs for top commit:
fanquake:
ACK fad0b67c212dcb8a16fcbda5a74acc959ed4e284 - this seems ok to me, and removes complexity from our CI system.
Tree-SHA512: 85e79f9f570e292d70a629d112fd4a6e6217d96226a1b665ed13485f616d84720ad2126b7d4b22fc603049f72fa7f2163b56a6bc276319fcd8b0496304ea4157
Diffstat (limited to '.cirrus.yml')
| -rw-r--r-- | .cirrus.yml | 15 |
1 files changed, 7 insertions, 8 deletions
diff --git a/.cirrus.yml b/.cirrus.yml index 4a6e73ac85..eaaa2c471f 100644 --- a/.cirrus.yml +++ b/.cirrus.yml @@ -30,6 +30,7 @@ persistent_worker_template_env: &PERSISTENT_WORKER_TEMPLATE_ENV # # The following specific types should exist, with the following requirements: # - lunar: For a machine running the Linux kernel shipped with Ubuntu Lunar 23.04. The machine is recommended to have 4 CPUs and 16 GB of memory. +# - arm64: For an aarch64 machine, recommended to have 2 CPUs and 8 GB of memory. persistent_worker_template: &PERSISTENT_WORKER_TEMPLATE persistent_worker: {} # Only use this if the task does not care about the type at all @@ -206,13 +207,11 @@ task: task: name: 'ARM [unit tests, no functional tests] [bullseye]' << : *GLOBAL_TASK_TEMPLATE - container: - docker_arguments: - CI_IMAGE_NAME_TAG: debian:bullseye - FILE_ENV: "./ci/test/00_setup_env_arm.sh" - << : *CREDITS_TEMPLATE + persistent_worker: + labels: + type: arm64 # Use arm64 worker to sidestep qemu and avoid a slow CI: https://github.com/bitcoin/bitcoin/pull/28087#issuecomment-1649399453 env: - << : *CIRRUS_EPHEMERAL_WORKER_TEMPLATE_ENV + FILE_ENV: "./ci/test/00_setup_env_arm.sh" task: name: 'Win64 [unit tests, no gui tests, no boost::process, no functional tests] [jammy]' @@ -230,7 +229,7 @@ task: << : *GLOBAL_TASK_TEMPLATE container: docker_arguments: - CI_IMAGE_NAME_TAG: quay.io/centos/centos:stream9 + CI_IMAGE_NAME_TAG: "quay.io/centos/amd64:stream9" FILE_ENV: "./ci/test/00_setup_env_i686_centos.sh" # For faster CI feedback, immediately schedule one task that runs all tests << : *CREDITS_TEMPLATE @@ -304,7 +303,7 @@ task: cpu: 4 memory: 16G # The default memory is too small, so double everything docker_arguments: - CI_IMAGE_NAME_TAG: ubuntu:focal + CI_IMAGE_NAME_TAG: "docker.io/amd64/ubuntu:focal" FILE_ENV: "./ci/test/00_setup_env_i686_multiprocess.sh" env: << : *CIRRUS_EPHEMERAL_WORKER_TEMPLATE_ENV |