model-writtenrun 4764 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Nazir Bilal Yavuz proposed switching both Windows CI jobs from containers to VMs in #1, reporting much shorter startup times and more than twice the execution speed with the same CPU and memory allocations.
- Justin Pryzby supported the change but questioned its effect on other CI resource consumption in #4; Andres Freund pointed to cfbot's scheduling limits in #5, and Thomas Munro calculated in #6 that non-Windows usage would rise roughly 25% while freeing job slots and reducing delays for new patches.
- Justin Pryzby found missing IPC::Run support in the MinGW image in #7, and Andres Freund explained in #8 that Packer was failing to abort after PowerShell errors, with Nazir working on a workaround.
- Nazir Bilal Yavuz reported the image problem fixed and submitted master and PostgreSQL 15 variants in #9, explaining that the original patch conflicted with the stable branch.
- Andres Freund landed 98811323c8 following #9, with a PostgreSQL 15 backport; the committed configuration changes exactly match the original proposal in #1, although the latest attachments were unavailable for direct comparison.
- In the push announcement #10, Andres Freund reported that MinGW TAP tests still lacked IPC::Run, described this as independent of the VM switch, and requested investigation; the thread contains no subsequent resolution of that issue.
- Thomas Munro observed fewer concurrent jobs and immediate capacity for new patches in #11, and Bharath Rupireddy reported Windows testing falling from about 27 minutes to 13 minutes in #12, closing the thread with the VM switch landed and its performance benefits confirmed.