container/common: Replace readUInt64 with optimized version #3406
Conversation
This patch replaces the `readUint64` function with one that optimizes for reading uint64s from files only. It prevents unnecessary calls to `.Stat` because a valid uint64 can be 20 characters at most anyway. It also avoids a few allocations. Overall, benchmarking has shown to reduce the total CPU time by ~15%, which given a total usage of roughly 3% in the entire kubelet, this means we're reducing the kubelet baseline CPU by about ~0.45%.
|
Hi @brancz. Thanks for your PR. I'm waiting for a google member to verify that this patch is reasonable to test. If it is, they should reply with Once the patch is verified, the new status will be reflected by the I understand the commands that are listed here. Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository. |
brancz
changed the title
| func readNBytesFromFile(file string, i int) []byte { | ||
| f, err := os.Open(file) | ||
| if err != nil { | ||
| // Ignore non-existent files | ||
| if !os.IsNotExist(err) { | ||
| klog.Warningf("readNBytesFromFile: Failed to open %q: %s", file, err) | ||
| } | ||
| return nil | ||
| } | ||
| defer f.Close() | ||
|
|
||
| buf := make([]byte, i) | ||
| n, err := io.ReadFull(f, buf) | ||
| if err != nil { | ||
| if err != io.EOF && err != io.ErrUnexpectedEOF { | ||
| klog.Warningf("readNBytesFromFile: Failed to read %q: %s", file, err) | ||
| } | ||
| } | ||
|
|
||
| return bytes.TrimSpace(buf[:n]) | ||
| } |
There was a problem hiding this comment.
Can I have uni tests for this function, please?
|
/ok-to-test |
This patch replaces the
readUint64function with one that optimizesfor reading uint64s from files only. It prevents unnecessary calls to
.Statbecause a valid uint64 can be 20 characters at most anyway. Italso avoids a few allocations.
Overall, benchmarking has shown to reduce the total CPU time by ~15%,
which given a total usage of roughly 3% in the entire kubelet, this
means we're reducing the kubelet baseline CPU by about ~0.45%.
Original profiling data from the Kubelet filtered down to only show
readUInt64: https://pprof.me/5dfadaf/?filter_by_function=readuint64&search_string=readuint64