[ML] Disable machine learning on macOS x86_64 #104125
Conversation
As previously advised in elastic#104087, machine learning functionality will no longer be available on macOS x86_64. Machine learning functionality _is_ still available on macOS by using an arm64 machine (Apple silicon). It is also possible to run Elasticsearch with machine learning functionality within a Docker container on macOS x86_64.
| Hi @droberts195, I've created a changelog YAML for you. Note that since this PR is labelled |
| DO NOT MERGE THIS PR IMMEDIATELY! This PR should be merged to main after the branch is split for the last minor release scheduled for before December 2024. For example, suppose 8.17.0 is scheduled for release in November 2024 and 8.18.0 is scheduled for release in January 2025. Then this PR should be merged to main after the 8.17 branch is split. One this PR is merged a followup PR should be opened against the ml-cpp repo to remove the build system for darwin-x86_64. I have confirmed that with this change in place the Elasticsearch build system works with an ml-cpp bundle that does not contain a After the build system for darwin-x86_64 has been removed from the ml-cpp repo, we will be able to do another PyTorch upgrade without having to worry about tweaking the build system to work on Intel macOS. |
| Pinging @elastic/ml-core (Team:ML) |
…ml_on_macos_x86_64
…ml_on_macos_x86_64
…lasticsearch into disable_ml_on_macos_x86_64
As previously advised in elastic#104087, machine learning functionality will no longer be available on macOS x86_64. Machine learning functionality is still available on macOS by using an arm64 machine (Apple silicon). It is also possible to run Elasticsearch with machine learning functionality within a Docker container on macOS x86_64. This PR should be merged to main after the branch is split for the last minor release scheduled for before December 2024. For example, suppose 8.17.0 is scheduled for release in November 2024 and 8.18.0 is scheduled for release in January 2025. Then this PR should be merged to main after the 8.17 branch is split. One this PR is merged a followup PR should be opened against the ml-cpp repo to remove the build system for darwin-x86_64. It has been confirmed that with this change in place the Elasticsearch build system works with an ml-cpp bundle that does not contain a platform/darwin-x86_64 directory. It still produces an Elasticsearch build that will run providing xpack.ml.enabled is not explicitly set to true. After the build system for darwin-x86_64 has been removed from the ml-cpp repo, we will be able to do another PyTorch upgrade without having to worry about tweaking the build system to work on Intel macOS. --------- Co-authored-by: Ed Savage <ed.savage@elastic.co> Co-authored-by: Valeriy Khakhutskyy <1292899+valeriy42@users.noreply.github.com>
From version 9.0.0 we drop support for macos x86_64 builds. Therefore all support for cross compiling for macos x86_64 has also been dropped. Macos builds have been migrated to be executed solely on macos 13 (Ventura) aarch64. In CI, these builds are performed on updated Orka VMs. In addition, 3rd party dependencies (boost, pytorch) are now baked into the Orka VM, so downloading these dependencies at run time is no longer required. Depends on elastic/elasticsearch#104125
As previously advised in #104087, machine learning functionality will no longer be available on macOS x86_64. Machine learning functionality is still available on macOS by using an arm64 machine (Apple silicon). It is also possible to run Elasticsearch with machine learning functionality within a Docker container on macOS x86_64. This PR should be merged to main after the branch is split for the last minor release scheduled for before December 2024. For example, suppose 8.17.0 is scheduled for release in November 2024 and 8.18.0 is scheduled for release in January 2025. Then this PR should be merged to main after the 8.17 branch is split. One this PR is merged a followup PR should be opened against the ml-cpp repo to remove the build system for darwin-x86_64. It has been confirmed that with this change in place the Elasticsearch build system works with an ml-cpp bundle that does not contain a platform/darwin-x86_64 directory. It still produces an Elasticsearch build that will run providing xpack.ml.enabled is not explicitly set to true. After the build system for darwin-x86_64 has been removed from the ml-cpp repo, we will be able to do another PyTorch upgrade without having to worry about tweaking the build system to work on Intel macOS. Backports #104125 --------- Co-authored-by: David Roberts <dave.roberts@elastic.co> Co-authored-by: Valeriy Khakhutskyy <1292899+valeriy42@users.noreply.github.com>
From version 9.0.0 we drop support for macos x86_64 builds. Therefore all support for cross compiling for macos x86_64 has also been dropped. Macos builds have been migrated to be executed solely on macos 13 (Ventura) aarch64. In CI, these builds are performed on updated Orka VMs. In addition, 3rd party dependencies (boost, pytorch) are now baked into the Orka VM, so downloading these dependencies at run time is no longer required. Depends on elastic/elasticsearch#104125
From version 8.18.0 we drop support for macos x86_64 builds. Therefore all support for cross compiling for macos x86_64 has also been dropped. Relates elastic/elasticsearch#104125
| Friendly reminder that this PR seems to be a breaking change for 9.0, but is missing from the 9.0 release note. We may want to add an entry to the breaking change section of 9.0 release note. |
Thanks @samxbr ... We don't offer official support for any macOS platform, see https://www.elastic.co/support/matrix. So I'm not sure if this counts as a breaking change. I will raise it with the wider team to confirm. |
| @droberts195 is this PR relevant to the serverless changelog? [FYI this question is based on 9.0 breaking changes] |
| @leemthompo no, this PR is not relevant for serverless |
8 participants
As previously advised in #104087, machine learning functionality will no longer be available on macOS x86_64.
Machine learning functionality is still available on macOS by using an arm64 machine (Apple silicon). It is also possible to run Elasticsearch with machine learning functionality within a Docker container on macOS x86_64.