rocksdb
rocksdb copied to clipboard
ArenaTest.UnmappedAllocation fails when "transparent hugepage" is enabled.
Note: Please use Issues only for bug reports. For questions, discussions, feature requests, etc. post to dev group: https://groups.google.com/forum/#!forum/rocksdb or https://www.facebook.com/groups/rocksdb.dev
Expected behavior
All tests pass on my desktop.
-` v01dstar@arch
.o+` -------------
`ooo/ OS: Arch Linux x86_64
`+oooo: Kernel: 6.7.2-arch1-2
`+oooooo: Uptime: 3 days, 57 mins
-+oooooo+: Packages: 729 (pacman)
`/:-:++oooo+: Shell: fish 3.7.0
`/++++/+++++++: Resolution: 1440x2560, 2560x1440
`/++++++++++++++: WM: i3
`/+++ooooooooooooo/` Theme: Adwaita [GTK3]
./ooosssso++osssssso+` Icons: Adwaita [GTK3]
.oossssso-````/ossssss+` Terminal: kitty
-osssssso. :ssssssso. CPU: AMD Ryzen 9 5900X (24) @ 3.700GHz
:osssssss/ osssso+++. GPU: AMD ATI Radeon RX 6400/6500 XT/6500M
/ossssssss/ +ssssooo/- GPU: NVIDIA GeForce RTX 3080 Lite Hash Rate
`/ossssso+/:- -:/+osssso+- Memory: 9632MiB / 64210MiB
`+sso+:-` `.-/+oso:
`++:. `-/+/
.` `/
v01dstar@arch ~> cat /sys/kernel/mm/transparent_hugepage/enabled
[always] madvise never
Actual behavior
ArenaTest.UnmappedAllocation fails. (It passes if I change THP to madvise or never).
Steps to reproduce the behavior
Run the test case with THP enabled (always). On my friend's Ubuntu, he also need to set sysctl -w vm.nr_hugepages=256 to reproduce the failure.
This test case expects on-demand Arena paging allocation (IIUC from the PR that introduced it), but seems, it needs to be composed conditionally based on OS, hardware's settings.