Merge pull request #55188 from mindprince/accelerator-monitoring

Automatic merge from submit-queue (batch tested with PRs 55798, 49579, 54862, 55188, 51990). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.

Add monitoring support for hardware accelerators

Currently only NVIDIA GPU monitoring is implemented.

Feature repo issue: https://github.com/kubernetes/features/issues/369
cAdvisor PR: https://github.com/google/cadvisor/pull/1762

/kind feature
/sig node
/sig instrumentation
/area hw-accelerators

**Release note**:
```release-note
Kubelet now exposes metrics for NVIDIA GPUs attached to the containers.
```

Kubernetes-commit: 779105673ab2f00af9e02dc9ac3c6413588aeae4
This commit is contained in:
Kubernetes Publisher
2017-11-16 03:09:21 -08:00
173 changed files with 3350 additions and 1180 deletions
+30 -18
View File
@@ -13,7 +13,6 @@
package unix
import (
"sync/atomic"
"syscall"
"unsafe"
)
@@ -167,7 +166,7 @@ func Getwd() (wd string, err error) {
func Getgroups() (gids []int, err error) {
n, err := getgroups(0, nil)
// Check for error and sanity check group count. Newer versions of
// Check for error and sanity check group count. Newer versions of
// Solaris allow up to 1024 (NGROUPS_MAX).
if n < 0 || n > 1024 {
if err != nil {
@@ -351,7 +350,7 @@ func Futimesat(dirfd int, path string, tv []Timeval) error {
}
// Solaris doesn't have an futimes function because it allows NULL to be
// specified as the path for futimesat. However, Go doesn't like
// specified as the path for futimesat. However, Go doesn't like
// NULL-style string interfaces, so this simple wrapper is provided.
func Futimes(fd int, tv []Timeval) error {
if tv == nil {
@@ -515,6 +514,24 @@ func Acct(path string) (err error) {
return acct(pathp)
}
//sys __makedev(version int, major uint, minor uint) (val uint64)
func Mkdev(major, minor uint32) uint64 {
return __makedev(NEWDEV, uint(major), uint(minor))
}
//sys __major(version int, dev uint64) (val uint)
func Major(dev uint64) uint32 {
return uint32(__major(NEWDEV, dev))
}
//sys __minor(version int, dev uint64) (val uint)
func Minor(dev uint64) uint32 {
return uint32(__minor(NEWDEV, dev))
}
/*
* Expose the ioctl function
*/
@@ -561,6 +578,15 @@ func IoctlGetTermio(fd int, req uint) (*Termio, error) {
return &value, err
}
//sys poll(fds *PollFd, nfds int, timeout int) (n int, err error)
func Poll(fds []PollFd, timeout int) (n int, err error) {
if len(fds) == 0 {
return poll(nil, 0, timeout)
}
return poll(&fds[0], len(fds), timeout)
}
/*
* Exposed directly
*/
@@ -613,6 +639,7 @@ func IoctlGetTermio(fd int, req uint) (*Termio, error) {
//sys Mlock(b []byte) (err error)
//sys Mlockall(flags int) (err error)
//sys Mprotect(b []byte, prot int) (err error)
//sys Msync(b []byte, flags int) (err error)
//sys Munlock(b []byte) (err error)
//sys Munlockall() (err error)
//sys Nanosleep(time *Timespec, leftover *Timespec) (err error)
@@ -699,18 +726,3 @@ func Mmap(fd int, offset int64, length int, prot int, flags int) (data []byte, e
func Munmap(b []byte) (err error) {
return mapper.Munmap(b)
}
//sys sysconf(name int) (n int64, err error)
// pageSize caches the value of Getpagesize, since it can't change
// once the system is booted.
var pageSize int64 // accessed atomically
func Getpagesize() int {
n := atomic.LoadInt64(&pageSize)
if n == 0 {
n, _ = sysconf(_SC_PAGESIZE)
atomic.StoreInt64(&pageSize, n)
}
return int(n)
}