Use strings.Cut where possible #4470

kolyshkin · 2024-10-23T06:58:54Z

For the most part, this is a switch from strings.Split[N] to strings.Cut.
Using strings.Cut (added in Go 1.18, see 1) results in faster and
cleaner code with less allocations (as we're not using a slice).

There are a few other cleanups and nits here and there.
Please see individual commits for details.

rata · 2024-11-05T13:47:16Z

@kolyshkin I guess this is due to performance? Did you run some numbers or we just tested before this is faster and we are just using strings.Cut in more places? IMHO, it would be nice to say the reason on the commits.

If something breaks due to this, it would be nice to know that the commit just changed to this due to performance, so a revert is probably safe until a new fixed version is cooked. But that is not that easy to know if we don't explain why we change it in the commit.

kolyshkin · 2024-11-06T20:53:52Z

@kolyshkin I guess this is due to performance? Did you run some numbers or we just tested before this is faster and we are just using strings.Cut in more places? IMHO, it would be nice to say the reason on the commits.

If something breaks due to this, it would be nice to know that the commit just changed to this due to performance, so a revert is probably safe until a new fixed version is cooked. But that is not that easy to know if we don't explain why we change it in the commit.

Thanks, this makes sense. Individual commit messages as well as this PR description updated; PTAL.

kolyshkin · 2024-12-05T00:05:43Z

@opencontainers/runc-maintainers PTAL. This is a minor cleanup, aiming for more readable and faster code with less allocations.

lifubang · 2024-12-20T04:31:03Z

exec.go

-			return nil, fmt.Errorf("invalid --cgroup argument: %s", c)
-		}
-		if len(cs) == 1 { // no controller: prefix
+		if ctr, path, ok := strings.Cut(c, ":"); ok {


I havn't checked that whether this will lead to some regressions or not, for example, if c == "a:b:c:d:e":
When using SplitN, cs[1] will be "b";
When using Cut, path will be "b:c:d:e".

So, the old code would error out when c == "a:b:c:d:e".

The new code would not, assigning b:c:d:e to path.

To me, the new code is (ever so slightly) more correct, since : is a valid symbol which should be allowed in path.

Yup, was looking at that as well; agreed, looks like new code is more correct here.

I still need to have a quick look at the splitting on commas, and the if len(args) != 1 { check below. Probably will check out this branch locally

Using strings.Cut (added in Go 1.18, see [1]) results in faster and cleaner code with less allocations (as we're not using a slice). This part of code is covered by tests in tests/integration/exec.bats. [1]: golang/go#46336 Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>

Nowadays strings.Fields are as fast as strings.SplitN so remove TODO. Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>

Using strings.Cut (added in Go 1.18, see [1]) results in faster and cleaner code with less allocations (as we're not using a slice). This also drops the check for extra dash (we're unlikely to get it from the kernel anyway). While at it, rename min/max -> from/to to avoid collision with Go min/max builtins. This code is tested by TestCPUSetStats* tests. [1]: golang/go#46336 Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>

Remove extra global constants that are only used in a single place and make it harder to read the code. Rename nanosecondsInSecond -> nsInSec. This code is tested by unit tests. Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>

For cgroup v2, we always expect /proc/$PID/cgroup contents like this: > 0::/user.slice/user-1000.slice/user@1000.service/app.slice/vte-spawn-f71c3fb8-519d-4e2d-b13e-9252594b1e05.scope So, it does not make sense to parse it using strings.Split, we can just cut the prefix and return the rest. Code tested by TestParseCgroupFromReader. Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>

Using strings.Cut (added in Go 1.18, see [1]) results in faster and cleaner code with less allocations (as we're not using a slice). The code is tested by testCgroupResourcesUnified. [1]: golang/go#46336 Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>

Using strings.Cut (added in Go 1.18, see [1]) results in faster and cleaner code with less allocations (as we're not using a slice). This code is tested by TestStatCPUPSI. [1]: golang/go#46336 Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>

Using strings.Cut (added in Go 1.18, see [1]) results in faster and cleaner code with less allocations (as we're not using a slice). Also, use switch in parseRdmaKV. [1]: golang/go#46336 Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>

Use strings.Cut in ParseKeyValue and GetValueByKey. Using strings.Cut (added in Go 1.18, see [1]) results in faster and cleaner code with less allocations (as we're not using a slice). [1]: golang/go#46336 Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>

thaJeztah · 2024-12-23T13:36:07Z

libcontainer/cgroups/fs/cpuacct.go

+		usageKernelMode = append(usageKernelMode, kernel)
+


Accidental empty line added here

thaJeztah · 2024-12-23T13:40:02Z

libcontainer/cgroups/fs/cpuacct.go

+		fields := strings.SplitN(scanner.Text(), " ", 3)
+		if len(fields) != 3 {
 			continue


I was considering if we had to check for more than 3 fields to happen here, but I guess that would error already once we try to strconv.ParseUint() below, so probably not adding anything.

OTOTH, I wonder if there could ever be more columns added here? Old code would do a SplitN(.., 4), and would skip if there would be less than 3 or more than 3 columns. The new code would silently skip less than 3 columns, but now errors on more than 3 columns.

Not sure if that behavior was correct, or if it should've processed the first 3 columns (and only ignored the 4th+ columns)? Is there any chance there would ever be more columns added?

thaJeztah · 2024-12-23T13:46:21Z

libcontainer/cgroups/fs2/defaultpath.go

-		if len(parts) < 3 {
-			return "", fmt.Errorf("invalid cgroup entry: %q", text)
-		}
-		// text is like "0::/user.slice/user-1001.slice/session-1.scope"
-		if parts[0] == "0" && parts[1] == "" {
-			return parts[2], nil
+		// "0::/user.slice/user-1001.slice/session-1.scope"
+		if path, ok := strings.CutPrefix(s.Text(), "0::"); ok {


Are we missing an error condition for lines that are invalid? Do we need to handle, say 0:::::::/hello.slice/world.scope, or is it fine to let those go through?

thaJeztah · 2024-12-23T14:01:25Z

libcontainer/cgroups/fs2/psi.go

 		case "avg10":
 			pv = &data.Avg10
 		case "avg60":
 			pv = &data.Avg60
 		case "avg300":
 			pv = &data.Avg300
 		case "total":
-			v, err := strconv.ParseUint(kv[1], 10, 64)
+			v, err := strconv.ParseUint(val, 10, 64)


Not for this PR, but I started to look in some other codebases to dismantle these errors. The default error returned is something like;

invalid total PSI value: strconv.ParseUint: parsing "some invalid value": invalid syntax

Which tends to be a bit "implementation detail", and strconv.ParseUint is not very relevant to the user. So in some cases, dismantling the error can be useful; https://cs.opensource.google/go/go/+/refs/tags/go1.23.4:src/strconv/atoi.go;l=20-49

https://go.dev/play/p/E9Dx1YBzPiN

var numErr *strconv.NumError if errors.As(err, &numErr) { fmt.Printf("invalid %s PS value (%s): %w", key, numErr.Num, numErr.Err) }

Which would be something like;

invalid total PSI value (some invalid value): invalid syntax

thaJeztah · 2024-12-23T14:05:31Z

libcontainer/cgroups/fscommon/rdma.go


-	if len(parts) != 2 {
+	if !ok {
 		return errors.New("Unable to parse RDMA entry")


Some of these could be improved as well; in this case we know it's missing a =. Wondering if we must also check for empty values after; looks like the if/else below would fall through to trying to parse empty values as integer, which may produce an obscure error (strconv.ParseUint: parsing "": invalid syntax).

We could check here if it's empty and return that no value is specified;

if !ok || v == "" { return errors.New("Unable to parse RDMA entry: no value specified") }

thaJeztah · 2024-12-23T14:08:54Z

libcontainer/cgroups/fscommon/utils.go

-	parts := strings.SplitN(t, " ", 3)
-	if len(parts) != 2 {
+	key, val, ok := strings.Cut(t, " ")
+	if !ok {


Not necessarily for this PR, but we could check for empty values here as well to return a nicer error than "strconv.ParseUint failed to parse ...";

if !ok || key == "" || val == "" {

thaJeztah · 2024-12-23T14:12:55Z

libcontainer/cgroups/fscommon/utils.go

-		arr := strings.Split(line, " ")
-		if len(arr) == 2 && arr[0] == key {
-			val, err := ParseUint(arr[1], 10, 64)
+		k, v, ok := strings.Cut(line, " ")
+		if ok && k == key {


Not sure if relevant (and probably new code is more correct?); previously, we would ignore cases with more than 2 fields. I.e., key foo bar would be ignored; in the new code, we'd try to parse foo bar as value.

That said; see my other comment; this would potentially be an issue if we try to parse something that may currently have 2 cols, and in future 3 cols; not sure if that should error or silently ignore

thaJeztah · 2024-12-23T14:26:15Z

exec.go

+			// No controller: prefix.
 			if len(args) != 1 {
 				return nil, fmt.Errorf("invalid --cgroup argument: %s (missing <controller>: prefix)", c)
 			}
 			paths[""] = c


Was trying to somewhat grasp what this case is for;

if zero args are passed, we return early (start of function)

if 1 arg is passed, it MUST have a controller prefix

But if 2 (or more) args are passed, then we don't have to?

So;

# valid: single arg --cgroup a,b:/hello-world # valid: single arg without controller --cgroup /hello-world # invalid: multiple args, one without controller --cgroup a,b:/hello-world --cgroup /hello-world --cgroup /hello-world --cgroup a,b:/hello-world

So, if no controller is passed, it's for everything which is only allowed if no other (per-controller) paths are specified, correct? (so not allowed to be combined).

☝️ Perhaps that's something we should capture in a comment on the function

Also;

perhaps we need to change the ok check to also check if controller is not empty, or is ,,,,,:/hello-world something that should be considered valid?

are duplicate controllers valid? They currently overwrite previous values (foo,foo,foo:/bar, foo:/baz)

kolyshkin force-pushed the strings-cut branch from 92d3c96 to bf53e34 Compare October 23, 2024 06:59

kolyshkin added the kind/refactor refactoring label Oct 23, 2024

kolyshkin force-pushed the strings-cut branch from bf53e34 to aeb74e8 Compare October 23, 2024 19:01

kolyshkin requested review from dqminh and rata and removed request for dqminh November 2, 2024 00:37

kolyshkin force-pushed the strings-cut branch from aeb74e8 to 732da9a Compare November 6, 2024 20:53

kolyshkin force-pushed the strings-cut branch from 732da9a to ae23108 Compare November 7, 2024 20:59

kolyshkin force-pushed the strings-cut branch 2 times, most recently from 819df22 to 62a4183 Compare December 5, 2024 00:04

kolyshkin added this to the 1.3.0 milestone Dec 5, 2024

kolyshkin force-pushed the strings-cut branch from 62a4183 to 37fdfdd Compare December 19, 2024 23:32

lifubang reviewed Dec 20, 2024

View reviewed changes

kolyshkin force-pushed the strings-cut branch from 37fdfdd to 247d92a Compare December 21, 2024 01:01

kolyshkin requested a review from lifubang December 23, 2024 01:46

kolyshkin added 9 commits December 22, 2024 17:46

libct/cg/fs: getPercpuUsage: rm TODO

d9e1005

Nowadays strings.Fields are as fast as strings.SplitN so remove TODO. Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>

libcontainer/cgroups/fs: some refactoring

a4ed2d8

Remove extra global constants that are only used in a single place and make it harder to read the code. Rename nanosecondsInSecond -> nsInSec. This code is tested by unit tests. Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>

kolyshkin force-pushed the strings-cut branch from 247d92a to cddca40 Compare December 23, 2024 01:46

thaJeztah reviewed Dec 23, 2024

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Use strings.Cut where possible #4470

Use strings.Cut where possible #4470

kolyshkin commented Oct 23, 2024 •

edited

Loading

rata commented Nov 5, 2024 •

edited

Loading

kolyshkin commented Nov 6, 2024 •

edited

Loading

kolyshkin commented Dec 5, 2024

lifubang Dec 20, 2024 •

edited

Loading

kolyshkin Dec 21, 2024

thaJeztah Dec 23, 2024

thaJeztah Dec 23, 2024

thaJeztah Dec 23, 2024

thaJeztah Dec 23, 2024

thaJeztah Dec 23, 2024

thaJeztah Dec 23, 2024

thaJeztah Dec 23, 2024

thaJeztah Dec 23, 2024

thaJeztah Dec 23, 2024

Use strings.Cut where possible #4470

Are you sure you want to change the base?

Use strings.Cut where possible #4470

Conversation

kolyshkin commented Oct 23, 2024 • edited Loading

rata commented Nov 5, 2024 • edited Loading

kolyshkin commented Nov 6, 2024 • edited Loading

kolyshkin commented Dec 5, 2024

lifubang Dec 20, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kolyshkin commented Oct 23, 2024 •

edited

Loading

rata commented Nov 5, 2024 •

edited

Loading

kolyshkin commented Nov 6, 2024 •

edited

Loading

lifubang Dec 20, 2024 •

edited

Loading