core: implement resourceUsage()

[r-52]

libuv already provides the uv_getrusage function to get the
resource mesaures for the current process.
This changes implements the new process.resourceUsage() function
to make use of it.

An example output looks like:

{ userCpuTimeUsedSec: 0,
  userCpuTimeUsedMs: 159729,
  systemCpuTimeUsedSec: 0,
  systemCpuTimeUsedMs: 30597,
  maxResidentSetSize: 20426752,
  sharedMemSize: 0,
  integralUnsharedDataSize: 0,
  integralUnsharedStackSize: 0,
  pageReclaims: 3002,
  pageFaults: 2250,
  swaps: 0,
  blockInputOperations: 0,
  blockOutputOperations: 0,
  ipcMessagesSent: 55,
  ipcMessagesReceived: 54,
  signalsReceived: 0,
  voluntaryContextSwitches: 92,
  involuntaryContextSwitches: 374 }

Not supported fields on windows are zero.

[bnoordhuis]

The indentation of the second argument seems off by one space.

[bnoordhuis]

LGTM with comments. Maybe this can get one more pair of eyeballs, @iojs/collaborators?

[cjihrig]

Do you know exactly which fields aren't available on Windows? If so, you could call them out.

[r-52]

@bnoordhuis @cjihrig thanks for your response! I removed the whitespace from the second argument.
The doc mentions now the fields, that are filled on windows.

[bnoordhuis]

LGTM

A question for everyone: do we want to deviate from the UNIX rusage field names? I'd say 'yes' because the UNIX names are pretty cryptic and don't look like idiomatic JS identifiers.

[trevnorris]

@bnoordhuis Using different names sounds appropriate.

[cjihrig]

+1 to more JS-like names.

[domenic]

Can we make the fields not-present on Windows instead? Would be a much better experience.

[trevnorris]

Easy enough. We just need to identify the fields that don't exist on Windows, ifdef them out and update the docs. Anyone know which fields don't work in Windows?

[r-52]

If the fields aren't present on Windows, would that make the behavior different to e.g os.loadavg? This function will return also zero filled fields on Windows.

[domenic]

Hmm, that is true. It would be more consistent with process.(get|set)e?(g|u)id, though. Maybe we should fix os.loadavg to not exist on Windows, like those functions do?

[rlidwka]

I'd prefer unix field names with ru_ prefix stripped. For example, isrss instead of integralUnsharedStackSize. Reasons are:

1. They are shorter.
2. They are more easily recognizable for people who worked with C.
3. They are more likely to be the same for different programming languages (I don't know what e.g. perl and python do, but I'd bet they do not use JS identifiers).
4. Documentation is readily available for them (manual pages + those cryptic nivcsw are google-able)

becauseLongCamelCasedIdentifiersAreUnreadable

If you still want to use those, maybe borrow some identifiers from here (just something I found searching for involuntaryContextSwitches):
http://www.thunderstone.com/site/vortexman/sysinfo.html

At least they are shorter.

[sam-github]

If it uses js-style identifiers, please document the mapping to the underlying getrusage fields, so that its possible to find reference documentation. Should do the same either way for Windows, so its possible to look up more information for Windows, as well.

[r-52]

@sam-github you mean something like a comment behind the example output:

....
maxResidentSetSize: 20426752, // maxrss
sharedMemSize: 0, // ixrss
integralUnsharedDataSize: 0, // idrss
...

@rlidwka thanks! I thought that the unix names are somehow cryptic. (In case of JS names: take messagesReceived instead of `ipcMessagesReceived``?)

@domenic it's not only ``os.loadavg()-os.cpus()` contains a `nice` field that's always filled with zero on Windows. The question is, should io.js as the runtime "hide" fields that aren't available on the current OS?

[domenic]

I guess I don't have strong enough feelings to override the precedent set for other os.* things. It seems nicer if they're absent but it's not like I have some program in mind that would benefit from the change.

[monsanto]

I would prefer if functions that cannot be meaningfully used do not exist, as opposed to returning garbage.

[mscdex]

+1 to keeping the properties on Windows and just setting them to 0. If nothing else it prevents the need to have to branch on process.platform or similar if you're doing running calculations on these fields.

[bnoordhuis]

I agree with @mscdex.

[jbergstroem]

how about returning -1 instead of 0? I agree that keeping properties is the way to go.

[r-52]

@jbergstroem -1 could look like an error code or an error that happened. 0 has the advantage that you could e.g. perform calculations with them and you don't get a negative result for the values.

[ghost]

Copy libuv/libuv#342

[ronkorving]

Agree with @jbergstroem that 0 is a mediocre "not supported" indicator. Why not undefined?
@romankl what good are the calculations on 0 possibly going to do?

[trevnorris]

@bnoordhuis think this is cool to land? seems it was on track before some last minute bikeshedding occurred. It LGTM.

[bnoordhuis]

LGTM. CI: https://jenkins-iojs.nodesource.com/job/node-test-pull-request/98/

[bnoordhuis]

The linter is probably going to complain that this should have 'use strict' at the top and that the imports should be const.

[rvagg]

What are the fields that are not supported on Windows? If it's more than just a few of them then I'm -0 on this I think.

[ghost]

@rvagg, over 50% fields are missing for windows in libuv, even though there is a Windows98-days helpers library (PDH) provided by win CRT till date for measuring low-level CPU related performance stats etc., which apparently emits all the information required to match Unix implementation (else how would game devs pull off so many projects on windows in the past and current century?). I have no dev experience with it therefore this request is blocking: libuv/libuv#342. Can anyone take a stab? Maybe guys from https://github.com/Microsoft/node get excited to chip in on this one; @jianchun.

[rvagg]

over 50% fields are missing for windows

I'm not a fan then, it's going to look second-class, the os.loadavg() precedent is an unfortunate one here that I'm also not a fan of for similar reasons.

@nodejs/platform-windows opinions?

[seishun]

I agree, we shouldn't encourage people to write non-portable code.

[orangemocha]

I have added libuv/libuv#342 to our queue of Windows fixes. Preferably, this PR would wait for that feature to be fully implemented in libuv on all platforms.

If this cannot wait, I would recommend specifying what is supported on which platform in the docs.

Since some of these fields might be platform-specific anyway, I think leaving them undefined where they are not implemented makes more sense than setting them to 0. It makes writing portable code easier.

[jasnell]

@orangemocha @rvagg @bnoordhuis ... what do we want to do with this one? Still pending?

[bnoordhuis]

@jasnell Still pending, I think. libuv/libuv#342 is still open. I don't expect anyone from the libuv team to work on that anytime soon though, it would have to come from outside.

[orangemocha]

@jasnell still planning to work on libuv/libuv#342

[eljefedelrodeodeljefe]

I could work on the windows part. I guess there won't be windows equivalents for the whole struct. So if we want to go further with it, we should be planning on windows specific docs.

[Trott]

libuv/libuv#342 is closed. Should we remove the blocked label from this?

[bnoordhuis]

Removed the label. The PR needs a rebase, though.

[Trott]

Since the repo that this PR is against is now listed as unknown repository and the original author last commented here nearly 2 years ago, I took the liberty of rebasing, resolving conflicts, updating a few things to reflect the current code base, and opening a PR at #11490. I'm going to close this. Hopefully we can get the new PR landed or closed quickly.