Hi, Some additional information and errata here.
I have split the tests into four stanzas as upstream does [1]. The resources of one GitHub Action runner is actually 4 cores + 16GB memory, not 2 cores + 8GB as I mentioned before. The test could finish within reasonable time (3hrs) with such configuration (although with a few unexpected failures, but I think it could be solved). I am still trying with fewer resources, especially shrinking the memory. [1]: https://salsa.debian.org/zfsonlinux-team/zfs/-/commit/7afeda495fa5b8129dfac45aef6340f46fbaf3a6 -- Thanks, Shengqi Chen.