RFR: 8287107: CgroupSubsystemFactory.setCgroupV2Path asserts with freezer controller [v3]

Severin Gehwolf sgehwolf at openjdk.java.net
Wed May 25 16:14:50 UTC 2022


On Wed, 25 May 2022 15:51:04 GMT, Ioi Lam <iklam at openjdk.org> wrote:

>> This PR fixes a bug found on an Ubuntu host that's mostly running with cgroupv2, but there's a controller (freezer) that is mounted in cgroupv1 mode.
>> 
>> The container support code in the VM and JDK checks if we have simultaneously mounted v1 and v2 containers. If so, we revert to "hybrid" mode (which is essentially v1).
>> 
>> However, the "hybrid checks" only considers the following controllers that Java cares about:
>> 
>> - cpu
>> - cpuacct
>> - cpuset
>> - blkio
>> - memory
>> - pids
>> 
>> If only "freezer" is mounted in v1 mode, Java will start in V2 mode. Later, when `setCgroupV2Path()` sees the first line in /proc/self/cgroup, it runs into the assert.
>> 
>> 
>> $ cat /proc/self/cgroup
>> 1:freezer:/
>> 0::/user.slice/user-1001.slice/session-85.scope
>> 
>> 
>> The fix is simple -- when we get to `setCgroupV2Path()`, we have already decided that the system has not mounted any v1 controllers that we care about, so we should just ignore all the v1 controllers (which has a non-empty name).
>
> Ioi Lam has updated the pull request incrementally with one additional commit since the last revision:
> 
>   fixed comments

src/java.base/linux/classes/jdk/internal/platform/CgroupSubsystemFactory.java line 149:

> 147:             // other controllers (such as freezer) are ignored and
> 148:             // are not considered in the checks below for
> 149:             // anyCgroupsV1Controller/anyCgroupsV1Controller.

It still has the `anyCgroupsV1Controller/anyCgroupsV1Controller` typo. Not **V1** twice?

-------------

PR: https://git.openjdk.java.net/jdk/pull/8858


More information about the core-libs-dev mailing list